{"id":5287,"date":"2026-02-24T04:14:15","date_gmt":"2026-02-24T04:14:15","guid":{"rendered":"https:\/\/research.undetectable.ai\/?p=5287"},"modified":"2026-02-24T04:14:50","modified_gmt":"2026-02-24T04:14:50","slug":"i-tested-5-ai-image-detectors-these-are-the-best-in-2026","status":"publish","type":"post","link":"https:\/\/research.undetectable.ai\/fr\/jai-teste-5-detecteurs-dimages-aeriennes-et-voici-les-meilleurs-en-2026\/","title":{"rendered":"J'ai test\u00e9 5 d\u00e9tecteurs d'images IA. Ce sont les meilleurs en 2026 :"},"content":{"rendered":"<p style=\"font-size:17px\"><strong>Vous avez probablement vu plus de photos g\u00e9n\u00e9r\u00e9es par l'IA en ligne que vous ne le pensez. <\/strong><\/p>\n\n\n\n<p style=\"font-size:17px\">Il est parfois \u00e9vident qu'une image a \u00e9t\u00e9 g\u00e9n\u00e9r\u00e9e par l'IA, mais il est de plus en plus difficile de s'en rendre compte \u00e0 mesure que les outils de g\u00e9n\u00e9ration d'images et de vid\u00e9os s'am\u00e9liorent. De nouveaux outils tels que Google <a href=\"https:\/\/blog.google\/innovation-and-ai\/products\/nano-banana-pro\/\" target=\"_blank\" rel=\"noopener\">Nano Banana Pro<\/a>et des mises \u00e0 jour de la base de donn\u00e9es de l'OpenAI <a href=\"https:\/\/openai.com\/index\/new-chatgpt-images-is-here\/\" target=\"_blank\" rel=\"noopener\">Mod\u00e8le d'image ChatGPT<\/a> permettent aux utilisateurs de g\u00e9n\u00e9rer rapidement des images synth\u00e9tiques qui refl\u00e8tent des images r\u00e9elles. Des recherches ant\u00e9rieures ont montr\u00e9 que 85% des Am\u00e9ricains d\u00e9clarent <a href=\"https:\/\/research.undetectable.ai\/fr\/85-des-americains-disent-que-les-deepfakes-ont-erode-leur-confiance-dans-les-informations-en-ligne\/\" data-type=\"link\" data-id=\"https:\/\/research.undetectable.ai\/85-of-americans-say-deepfakes-have-eroded-their-trust-in-online-information\/\">Les \"deepfakes\" minent la confiance en ligne<\/a>.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" style=\"font-size:19px\"><strong>De nombreux outils pr\u00e9tendent d\u00e9tecter les images g\u00e9n\u00e9r\u00e9es par l'IA, mais fonctionnent-ils ? <\/strong><\/h2>\n\n\n\n<p style=\"font-size:17px\">J'ai effectu\u00e9 50 contr\u00f4les de d\u00e9tection sur cinq des d\u00e9tecteurs d'images les plus populaires et j'ai document\u00e9 les r\u00e9sultats. Non seulement je pr\u00e9senterai toutes les donn\u00e9es dans cet article et les expliquerai, mais je mettrai \u00e9galement un lien vers la documentation \u00e0 la fin de l'article.<\/p>\n\n\n\n<p style=\"font-size:17px\">Les cinq d\u00e9tecteurs que j'ai utilis\u00e9s pour ces tests sont les suivants : <a href=\"http:\/\/Truthscan.com\" data-type=\"link\" data-id=\"Truthscan.com\" target=\"_blank\" rel=\"noreferrer noopener\">TruthScan<\/a>, <a href=\"https:\/\/www.aiornot.com\/\" data-type=\"link\" data-id=\"https:\/\/www.aiornot.com\/\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">IA ou pas<\/a>, <a href=\"https:\/\/sightengine.com\/\" data-type=\"link\" data-id=\"https:\/\/sightengine.com\/\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">Moteur de vision<\/a>, <a href=\"https:\/\/wasitai.com\/\" data-type=\"link\" data-id=\"https:\/\/wasitai.com\/\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">WasItAI<\/a>et <a href=\"https:\/\/gowinston.ai\/\" data-type=\"link\" data-id=\"https:\/\/gowinston.ai\/\" target=\"_blank\" rel=\"noreferrer noopener\">Winston AI<\/a>. J'ai test\u00e9 ces outils en leur faisant analyser deux images g\u00e9n\u00e9r\u00e9es par ChatGPT, six images g\u00e9n\u00e9r\u00e9es par Nano Banana et deux images g\u00e9n\u00e9r\u00e9es par Midjourney. J'ai utilis\u00e9 plusieurs styles et techniques d'incitation (que je d\u00e9taillerai plus tard lorsque j'analyserai chaque image analys\u00e9e).<\/p>\n\n\n\n<p style=\"font-size:17px\">Les cinq d\u00e9tecteurs ont \u00e9t\u00e9 charg\u00e9s de rep\u00e9rer des images cr\u00e9\u00e9es par l'IA dans des cat\u00e9gories telles que la fraude, la d\u00e9sinformation, la photographie g\u00e9n\u00e9rale et les deepfakes. Comme on pouvait s'y attendre, tous les d\u00e9tecteurs n'ont pas obtenu de bons r\u00e9sultats. <strong>TruthScan a \u00e9t\u00e9 le seul d\u00e9tecteur \u00e0 classer syst\u00e9matiquement tous les contenus que j'ai soumis \u00e0 97% ou plus.<\/strong><\/p>\n\n\n\n<!DOCTYPE html>\n<html lang=\"en\">\n<head>\n<meta charset=\"UTF-8\">\n<meta name=\"viewport\" content=\"width=device-width, initial-scale=1.0\">\n<title>R\u00e9sultats du test du d\u00e9tecteur d'images AI<\/title>\n<style>\n  @import url('https:\/\/fonts.googleapis.com\/css2?family=DM+Sans:wght@400;500;600;700&family=Outfit:wght@600;700;800&display=swap');\n\n  .ai-detector-wrap {\n    max-width: 780px;\n    margin: 2rem auto;\n    font-family: 'DM Sans', sans-serif;\n    -webkit-font-smoothing: antialiased;\n  }\n\n  .ai-detector-wrap h2 {\n    font-family: 'Outfit', sans-serif;\n    font-size: 1.75rem;\n    font-weight: 800;\n    color: #1a1a2e;\n    margin-bottom: 1.25rem;\n    letter-spacing: -0.02em;\n  }\n\n  .ai-detector-table {\n    width: 100%;\n    border-collapse: separate;\n    border-spacing: 0;\n    border-radius: 14px;\n    overflow: hidden;\n    box-shadow:\n      0 1px 3px rgba(0,0,0,0.06),\n      0 8px 24px rgba(0,0,0,0.08),\n      0 0 0 1px rgba(0,0,0,0.04);\n  }\n\n  .ai-detector-table thead th {\n    background: #1a1a2e;\n    color: #ffffff;\n    font-family: 'Outfit', sans-serif;\n    font-weight: 700;\n    font-size: 0.85rem;\n    text-transform: uppercase;\n    letter-spacing: 0.06em;\n    padding: 16px 20px;\n    text-align: center;\n    border: none;\n  }\n\n  .ai-detector-table thead th:first-child {\n    text-align: left;\n    padding-left: 24px;\n  }\n\n  .ai-detector-table tbody tr {\n    transition: background 0.2s ease;\n  }\n\n  .ai-detector-table tbody tr:hover {\n    background: rgba(0,0,0,0.02);\n  }\n\n  .ai-detector-table tbody td {\n    padding: 14px 20px;\n    text-align: center;\n    font-size: 0.95rem;\n    font-weight: 500;\n    color: #2d2d3a;\n    border-bottom: 1px solid #eee;\n  }\n\n  .ai-detector-table tbody tr:last-child td {\n    border-bottom: none;\n  }\n\n  .ai-detector-table tbody td:first-child {\n    text-align: left;\n    padding-left: 24px;\n    font-weight: 600;\n    color: #1a1a2e;\n  }\n\n  \/* Colored cells *\/\n  .cell-green {\n    background: #22c55e;\n    color: #fff !important;\n    font-weight: 700 !important;\n    border-radius: 6px;\n    padding: 6px 12px;\n    display: inline-block;\n    min-width: 32px;\n    text-align: center;\n    font-size: 0.9rem;\n  }\n\n  .cell-red {\n    background: #ef4444;\n    color: #fff !important;\n    font-weight: 700 !important;\n    border-radius: 6px;\n    padding: 6px 12px;\n    display: inline-block;\n    min-width: 32px;\n    text-align: center;\n    font-size: 0.9rem;\n  }\n\n  \/* Accuracy badges *\/\n  .badge-100  { background: #16a34a; color: #fff; }\n  .badge-80   { background: #a3e635; color: #1a1a2e; }\n  .badge-70   { background: #facc15; color: #1a1a2e; }\n  .badge-60   { background: #fb923c; color: #fff; }\n  .badge-30   { background: #ef4444; color: #fff; }\n\n  .accuracy-badge {\n    font-weight: 800;\n    font-family: 'Outfit', sans-serif;\n    font-size: 0.95rem;\n    border-radius: 8px;\n    padding: 6px 16px;\n    display: inline-block;\n    min-width: 60px;\n    text-align: center;\n    letter-spacing: 0.02em;\n  }\n\n  \/* Row rank indicator *\/\n  .rank-icon {\n    margin-right: 6px;\n    font-size: 1rem;\n  }\n\n  \/* Asterisk note *\/\n  .table-note {\n    margin-top: 0.75rem;\n    font-size: 0.8rem;\n    color: #888;\n    padding-left: 4px;\n  }\n\n  \/* Responsive *\/\n  @media (max-width: 600px) {\n    .ai-detector-table thead th,\n    .ai-detector-table tbody td {\n      padding: 10px 10px;\n      font-size: 0.8rem;\n    }\n    .ai-detector-wrap h2 {\n      font-size: 1.3rem;\n    }\n  }\n<\/style>\n<\/head>\n<body style=\"background:#f8f8fa; padding: 2rem;\">\n\n<div class=\"ai-detector-wrap\">\n  <h2>R\u00e9sultats du test du d\u00e9tecteur d'images AI<\/h2>\n\n  <table class=\"ai-detector-table\">\n    <thead>\n      <tr>\n        <th>Outil de d\u00e9tection<\/th>\n        <th>Total des tests<\/th>\n        <th>Correctement d\u00e9tect\u00e9<\/th>\n        <th>Rat\u00e9 (\u00c9chec)<\/th>\n        <th>Pr\u00e9cision<\/th>\n      <\/tr>\n    <\/thead>\n    <tbody>\n      <tr>\n        <td><span class=\"rank-icon\">\ud83e\udd47<\/span>TruthScan<\/td>\n        <td>10<\/td>\n        <td><span class=\"cell-green\">10<\/span><\/td>\n        <td><span class=\"cell-green\">0<\/span><\/td>\n        <td><span class=\"accuracy-badge badge-100\">100%<\/span><\/td>\n      <\/tr>\n      <tr>\n        <td><span class=\"rank-icon\">\ud83e\udd48<\/span>IA ou pas<\/td>\n        <td>10<\/td>\n        <td><span class=\"cell-green\">8<\/span><\/td>\n        <td><span class=\"cell-red\">2*<\/span><\/td>\n        <td><span class=\"accuracy-badge badge-80\">80%<\/span><\/td>\n      <\/tr>\n      <tr>\n        <td><span class=\"rank-icon\">\ud83e\udd49<\/span>Moteur de vision<\/td>\n        <td>10<\/td>\n        <td><span class=\"cell-green\">7<\/span><\/td>\n        <td><span class=\"cell-red\">3<\/span><\/td>\n        <td><span class=\"accuracy-badge badge-70\">70%<\/span><\/td>\n      <\/tr>\n      <tr>\n        <td><span class=\"rank-icon\">&nbsp;&nbsp;&nbsp;<\/span>WasItAI<\/td>\n        <td>10<\/td>\n        <td><span class=\"cell-green\">6<\/span><\/td>\n        <td><span class=\"cell-red\">4<\/span><\/td>\n        <td><span class=\"accuracy-badge badge-60\">60%<\/span><\/td>\n      <\/tr>\n      <tr>\n        <td><span class=\"rank-icon\">&nbsp;&nbsp;&nbsp;<\/span>Winston AI<\/td>\n        <td>10<\/td>\n        <td><span class=\"cell-green\">3<\/span><\/td>\n        <td><span class=\"cell-red\">7<\/span><\/td>\n        <td><span class=\"accuracy-badge badge-30\">30%<\/span><\/td>\n      <\/tr>\n    <\/tbody>\n  <\/table>\n\n  <p class=\"table-note\">* AI or Not a \u00e9tiquet\u00e9 une g\u00e9n\u00e9ration AI comme 78%, une comme 85% et une comme 89%. Comme le test exige au moins 90%, nous comptons les deux premiers comme des \u00e9checs, mais nous consid\u00e9rons le 89% comme un quasi \u00e9chec, d'o\u00f9 l'ast\u00e9risque.<\/p>\n<\/div>\n\n<\/body>\n<\/html>\n\n\n\n<p style=\"font-size:17px\">AI or Not a \u00e9galement obtenu d'excellents r\u00e9sultats. En fait, m\u00eame si AI or Not a class\u00e9 plusieurs \u00e9l\u00e9ments g\u00e9n\u00e9r\u00e9s par l'IA en dessous du niveau de certitude 90%, il a \u00e9t\u00e9 le deuxi\u00e8me d\u00e9tecteur le plus pr\u00e9cis et le plus coh\u00e9rent lors de mes tests, derri\u00e8re TruthScan. M\u00eame lorsque AI or Not a \u00e9chou\u00e9, ces \u00e9checs n'ont pas \u00e9t\u00e9 catastrophiques, mais des am\u00e9liorations sont encore possibles. <\/p>\n\n\n\n<p style=\"font-size:17px\">Les autres d\u00e9tecteurs d'images d'IA que j'ai essay\u00e9s ont obtenu des r\u00e9sultats bien pires. Par exemple, Sight Engine a mal class\u00e9 3 images d'IA li\u00e9es \u00e0 la fraude en les qualifiant d'authentiques.<\/p>\n\n\n\n<p style=\"font-size:17px\">Je vais maintenant vous montrer chaque image (sur les 10 que j'ai g\u00e9n\u00e9r\u00e9es), expliquer comment je l'ai cr\u00e9\u00e9e et montrer comment chaque mod\u00e8le l'a not\u00e9e.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">#1. \"Man On a Ledge\" (g\u00e9n\u00e9r\u00e9 par ChatGPT)<\/h2>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"683\" height=\"1024\" src=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/CGPT-Generation-1-Man-on-ledge-683x1024.jpg\" alt=\"\" class=\"wp-image-5289\" title=\"\" srcset=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/CGPT-Generation-1-Man-on-ledge-683x1024.jpg 683w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/CGPT-Generation-1-Man-on-ledge-200x300.jpg 200w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/CGPT-Generation-1-Man-on-ledge-768x1152.jpg 768w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/CGPT-Generation-1-Man-on-ledge-8x12.jpg 8w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/CGPT-Generation-1-Man-on-ledge.jpg 1024w\" sizes=\"auto, (max-width: 683px) 100vw, 683px\" \/><\/figure>\n\n\n\n<style>\n  @import url('https:\/\/fonts.googleapis.com\/css2?family=DM+Sans:wght@400;500;600;700&family=Outfit:wght@600;700;800&display=swap');\n\n  .detector-scores-wrap {\n    max-width: 780px;\n    margin: 2rem auto;\n    font-family: 'DM Sans', sans-serif;\n    -webkit-font-smoothing: antialiased;\n  }\n\n  .detector-scores-wrap h3 {\n    font-family: 'Outfit', sans-serif;\n    font-size: 1.4rem;\n    font-weight: 800;\n    color: #1a1a2e;\n    margin-bottom: 1.25rem;\n    letter-spacing: -0.02em;\n  }\n\n  .detector-scores-table {\n    width: 100%;\n    border-collapse: separate;\n    border-spacing: 0;\n    border-radius: 14px;\n    overflow: hidden;\n    table-layout: fixed;\n    box-shadow:\n      0 1px 3px rgba(0,0,0,0.06),\n      0 8px 24px rgba(0,0,0,0.08),\n      0 0 0 1px rgba(0,0,0,0.04);\n  }\n\n  .detector-scores-table thead th {\n    background: #1a1a2e;\n    color: #ffffff;\n    font-family: 'Outfit', sans-serif;\n    font-weight: 700;\n    font-size: 0.7rem;\n    text-transform: uppercase;\n    letter-spacing: 0.04em;\n    padding: 14px 6px;\n    text-align: center;\n    border: none;\n  }\n\n  .detector-scores-table thead th:first-child {\n    text-align: left;\n    padding-left: 14px;\n    width: 18%;\n  }\n\n  .detector-scores-table thead th:nth-child(2) {\n    width: 12%;\n  }\n\n  .detector-scores-table tbody tr {\n    transition: background 0.2s ease;\n  }\n\n  .detector-scores-table tbody tr:hover {\n    background: rgba(0,0,0,0.02);\n  }\n\n  .detector-scores-table tbody td {\n    padding: 14px 6px;\n    text-align: center;\n    font-size: 0.9rem;\n    font-weight: 500;\n    color: #2d2d3a;\n    border-bottom: 1px solid #eee;\n  }\n\n  .detector-scores-table tbody tr:last-child td {\n    border-bottom: none;\n  }\n\n  .detector-scores-table tbody td:first-child {\n    text-align: left;\n    padding-left: 14px;\n    font-weight: 600;\n    color: #1a1a2e;\n  }\n\n  .score-pill {\n    font-weight: 700;\n    font-family: 'Outfit', sans-serif;\n    font-size: 0.78rem;\n    border-radius: 6px;\n    padding: 5px 8px;\n    display: inline-block;\n    min-width: 0;\n    text-align: center;\n    white-space: nowrap;\n  }\n\n  .score-pass {\n    background: #22c55e;\n    color: #fff;\n  }\n\n  .score-warn {\n    background: #fb923c;\n    color: #fff;\n  }\n\n  .score-fail {\n    background: #ef4444;\n    color: #fff;\n  }\n\n  @media (max-width: 600px) {\n    .detector-scores-table thead th,\n    .detector-scores-table tbody td {\n      padding: 10px 4px;\n      font-size: 0.7rem;\n    }\n    .score-pill {\n      padding: 4px 5px;\n      font-size: 0.68rem;\n    }\n    .detector-scores-wrap h3 {\n      font-size: 1.15rem;\n    }\n  }\n<\/style>\n\n<div class=\"detector-scores-wrap\">\n  <h3>Comment les d\u00e9tecteurs d'images l'ont not\u00e9<\/h3>\n\n  <table class=\"detector-scores-table\">\n    <thead>\n      <tr>\n        <th>Mod\u00e8le d'image<\/th>\n        <th>Cat\u00e9gorie<\/th>\n        <th>TruthScan<\/th>\n        <th>Moteur de vision<\/th>\n        <th>IA ou pas<\/th>\n        <th>Winston AI<\/th>\n        <th>WasItAI<\/th>\n      <\/tr>\n    <\/thead>\n    <tbody>\n      <tr>\n        <td>ChatGPT <span style=\"font-size:0.75rem; color:#888; display:block;\">G\u00e9n\u00e9ration compl\u00e8te<\/span><\/td>\n        <td>G\u00e9n\u00e9ral<\/td>\n        <td><span class=\"score-pill score-pass\">99.00% AI<\/span><\/td>\n        <td><span class=\"score-pill score-pass\">99.00% AI<\/span><\/td>\n        <td><span class=\"score-pill score-warn\">77.99% AI<\/span><\/td>\n        <td><span class=\"score-pill score-fail\">0,98% AI<\/span><\/td>\n        <td><span class=\"score-pill score-fail\">1.00% AI<\/span><\/td>\n      <\/tr>\n    <\/tbody>\n  <\/table>\n<\/div>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p style=\"font-size:17px\">Pour moi (et je vois beaucoup de contenu g\u00e9n\u00e9r\u00e9 par l'IA tous les jours), l'image semble assez convaincante \u00e0 premi\u00e8re vue. Si vous cliquiez sur le profil de quelqu'un sur les m\u00e9dias sociaux, que vous parcouriez son fil d'actualit\u00e9 et que vous voyiez cette image, seriez-vous imm\u00e9diatement convaincu qu'il s'agit d'un faux g\u00e9n\u00e9r\u00e9 par l'IA ? Pour \u00eatre honn\u00eate, j'ai essay\u00e9 d'\u00eatre cr\u00e9atif avec le message. Voici celle que j'ai utilis\u00e9e pour g\u00e9n\u00e9rer cette image :<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p class=\"has-text-align-center has-vivid-cyan-blue-color has-text-color has-link-color has-medium-font-size wp-elements-99b76fc78211de13a760aff177dd504c\">&#8220;<em><strong>Cr\u00e9ez-moi une photo de style appareil photo instantan\u00e9 d'un homme debout au sommet d'un toit regardant la rue, la nuit, il est \u00e9clair\u00e9 par le flash de l'appareil photo, il y a des traces de lumi\u00e8re provenant des voitures en contrebas, une femme se tient derri\u00e8re lui, la main sur les l\u00e8vres, souriante, l'esth\u00e9tique de la photo ressemble \u00e0 une photo prise sur un appareil photo instantan\u00e9 en 2009 par des amis d'universit\u00e9 en train de s'amuser.<\/strong><\/em>&#8220;<\/p>\n<\/blockquote>\n<\/blockquote>\n<\/blockquote>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p style=\"font-size:17px\">Quelle que soit ma cr\u00e9ativit\u00e9 ou le nombre de d\u00e9tails que j'ai inclus dans mon message, TruthScan et Sight Engine ont toujours correctement class\u00e9 la sortie comme \u00e9tant g\u00e9n\u00e9r\u00e9e par l'IA. Le d\u00e9tecteur AI or Not s'en est approch\u00e9, mais a tout de m\u00eame manqu\u00e9 la cible. <\/p>\n\n\n\n<p style=\"font-size:17px\">Winston et WasItAI se sont carr\u00e9ment tromp\u00e9s en classant l'image ChatGPT comme r\u00e9elle.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">#2. \"The Fake Receipt\" (g\u00e9n\u00e9r\u00e9 par ChatGPT)<\/h2>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"683\" height=\"1024\" src=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/CGPT2Bestbuyreceipttest-683x1024.jpg\" alt=\"Une image montrant ce qui semble \u00eatre un re\u00e7u tach\u00e9\" class=\"wp-image-5291\" title=\"\" srcset=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/CGPT2Bestbuyreceipttest-683x1024.jpg 683w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/CGPT2Bestbuyreceipttest-200x300.jpg 200w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/CGPT2Bestbuyreceipttest-768x1152.jpg 768w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/CGPT2Bestbuyreceipttest-8x12.jpg 8w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/CGPT2Bestbuyreceipttest.jpg 1024w\" sizes=\"auto, (max-width: 683px) 100vw, 683px\" \/><\/figure>\n\n\n\n<style>\n  @import url('https:\/\/fonts.googleapis.com\/css2?family=DM+Sans:wght@400;500;600;700&family=Outfit:wght@600;700;800&display=swap');\n\n  .detector-scores-wrap {\n    max-width: 780px;\n    margin: 2rem auto;\n    font-family: 'DM Sans', sans-serif;\n    -webkit-font-smoothing: antialiased;\n  }\n\n  .detector-scores-wrap h3 {\n    font-family: 'Outfit', sans-serif;\n    font-size: 1.4rem;\n    font-weight: 800;\n    color: #1a1a2e;\n    margin-bottom: 1.25rem;\n    letter-spacing: -0.02em;\n  }\n\n  .detector-scores-table {\n    width: 100%;\n    border-collapse: separate;\n    border-spacing: 0;\n    border-radius: 14px;\n    overflow: hidden;\n    table-layout: fixed;\n    box-shadow:\n      0 1px 3px rgba(0,0,0,0.06),\n      0 8px 24px rgba(0,0,0,0.08),\n      0 0 0 1px rgba(0,0,0,0.04);\n  }\n\n  .detector-scores-table thead th {\n    background: #1a1a2e;\n    color: #ffffff;\n    font-family: 'Outfit', sans-serif;\n    font-weight: 700;\n    font-size: 0.7rem;\n    text-transform: uppercase;\n    letter-spacing: 0.04em;\n    padding: 14px 6px;\n    text-align: center;\n    border: none;\n  }\n\n  .detector-scores-table thead th:first-child {\n    text-align: left;\n    padding-left: 14px;\n    width: 18%;\n  }\n\n  .detector-scores-table thead th:nth-child(2) {\n    width: 12%;\n  }\n\n  .detector-scores-table tbody tr {\n    transition: background 0.2s ease;\n  }\n\n  .detector-scores-table tbody tr:hover {\n    background: rgba(0,0,0,0.02);\n  }\n\n  .detector-scores-table tbody td {\n    padding: 14px 6px;\n    text-align: center;\n    font-size: 0.9rem;\n    font-weight: 500;\n    color: #2d2d3a;\n    border-bottom: 1px solid #eee;\n  }\n\n  .detector-scores-table tbody tr:last-child td {\n    border-bottom: none;\n  }\n\n  .detector-scores-table tbody td:first-child {\n    text-align: left;\n    padding-left: 14px;\n    font-weight: 600;\n    color: #1a1a2e;\n  }\n\n  .score-pill {\n    font-weight: 700;\n    font-family: 'Outfit', sans-serif;\n    font-size: 0.78rem;\n    border-radius: 6px;\n    padding: 5px 8px;\n    display: inline-block;\n    min-width: 0;\n    text-align: center;\n    white-space: nowrap;\n  }\n\n  .score-pass {\n    background: #22c55e;\n    color: #fff;\n  }\n\n  .score-warn {\n    background: #fb923c;\n    color: #fff;\n  }\n\n  .score-fail {\n    background: #ef4444;\n    color: #fff;\n  }\n\n  @media (max-width: 600px) {\n    .detector-scores-table thead th,\n    .detector-scores-table tbody td {\n      padding: 10px 4px;\n      font-size: 0.7rem;\n    }\n    .score-pill {\n      padding: 4px 5px;\n      font-size: 0.68rem;\n    }\n    .detector-scores-wrap h3 {\n      font-size: 1.15rem;\n    }\n  }\n<\/style>\n\n<div class=\"detector-scores-wrap\">\n  <h3>Comment les d\u00e9tecteurs d'images l'ont not\u00e9<\/h3>\n\n  <table class=\"detector-scores-table\">\n    <thead>\n      <tr>\n        <th>Mod\u00e8le d'image<\/th>\n        <th>Cat\u00e9gorie<\/th>\n        <th>TruthScan<\/th>\n        <th>Moteur de vision<\/th>\n        <th>IA ou pas<\/th>\n        <th>Winston AI<\/th>\n        <th>WasItAI<\/th>\n      <\/tr>\n    <\/thead>\n    <tbody>\n      <tr>\n        <td>ChatGPT <span style=\"font-size:0.75rem; color:#888; display:block;\">G\u00e9n\u00e9ration compl\u00e8te<\/span><\/td>\n        <td>Fraude<\/td>\n        <td><span class=\"score-pill score-pass\">99.00% AI<\/span><\/td>\n        <td><span class=\"score-pill score-fail\">19.00% AI<\/span><\/td>\n        <td><span class=\"score-pill score-pass\">94.48% AI<\/span><\/td>\n        <td><span class=\"score-pill score-fail\">0,04% AI<\/span><\/td>\n        <td><span class=\"score-pill score-fail\">1.00% AI<\/span><\/td>\n      <\/tr>\n    <\/tbody>\n  <\/table>\n<\/div>\n\n\n\n<p style=\"font-size:17px\">J'ai voulu g\u00e9n\u00e9rer un faux re\u00e7u avec ChatGPT pour voir \u00e0 quel point il serait r\u00e9aliste. Vous remarquerez que l'image n'a pas d'adresse l\u00e9gitime, mais la texture semble encore un peu r\u00e9aliste. Je dirais que cette image est moins convaincante que la premi\u00e8re que j'ai fait g\u00e9n\u00e9rer par ChatGPT, mais la plupart des d\u00e9tecteurs ont tout de m\u00eame \u00e9chou\u00e9 en l'analysant.<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li style=\"font-size:17px\">Parmi toutes les v\u00e9rifications que j'ai effectu\u00e9es sur cette image, <strong>TruthScan<\/strong> \u00e9tait le plus pr\u00e9cis. <\/li>\n\n\n\n<li style=\"font-size:17px\"><strong>Moteur de vision<\/strong> a vraiment \u00e9chou\u00e9 ici, faisant preuve d'un mauvais jugement sur une image d'IA li\u00e9e \u00e0 un document.<\/li>\n\n\n\n<li style=\"font-size:17px\"><strong>IA ou pas<\/strong> a obtenu des r\u00e9sultats nettement meilleurs dans cette deuxi\u00e8me analyse.<\/li>\n\n\n\n<li style=\"font-size:17px\"><strong>Winston AI <\/strong>a obtenu les pires r\u00e9sultats, et WasItAI a \u00e9galement compl\u00e8tement \u00e9chou\u00e9.<\/li>\n<\/ul>\n\n\n\n<p style=\"font-size:17px\">\u00c9tant donn\u00e9 que cette image est associ\u00e9e \u00e0 la cat\u00e9gorie fraude, il est inqui\u00e9tant de constater que la plupart des d\u00e9tecteurs l'ont mal class\u00e9e. <\/p>\n\n\n\n<h3 class=\"wp-block-heading\">L'invite que j'ai utilis\u00e9e pour g\u00e9n\u00e9rer l'image<\/h3>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p class=\"has-vivid-cyan-blue-color has-text-color has-link-color wp-elements-b58fc16e4a944c203389dba029f958d9\"><strong>&#8220;<em>G\u00e9n\u00e9rer l'image d'un ticket de caisse qui indique que le montant total d\u00e9pens\u00e9 est d'un milliard de dollars et qui comporte une tache de caf\u00e9.<\/em>&#8220;<\/strong><\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p style=\"font-size:17px\">Il est clair que j'ai \u00e9crit une invite simple et rapide. N\u00e9anmoins, la qualit\u00e9 de la sortie du ChatGPT est visuellement convaincante \u00e0 l'\u0153il nu. Elle ne l'est pas pour deux des cinq d\u00e9tecteurs que j'ai test\u00e9s, semble-t-il.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">#3 : \"Hazard Package\" (g\u00e9n\u00e9r\u00e9 avec Nano Banana)<\/h2>\n\n\n\n<p><strong><em>Au lieu de laisser Nano Banana g\u00e9n\u00e9rer une image originale, je lui en ai donn\u00e9 une \u00e0 modifier. <\/em><\/strong><\/p>\n\n\n\n<p style=\"font-size:17px\">L'id\u00e9e \u00e9tait de montrer comment quelqu'un pouvait facilement utiliser l'IA pour cr\u00e9er de fausses preuves et pr\u00e9tendre avoir re\u00e7u un colis endommag\u00e9\/dangereux. Tout d'abord, j'ai pris une photo avec un iPhone 15 d'un paquet Amazon vide que j'ai trouv\u00e9.<\/p>\n\n\n\n<p><strong>Voici la photo que j'ai prise :<\/strong><\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"759\" height=\"1024\" src=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/Amazon-package-759x1024.jpg\" alt=\"Photo d&#039;un colis Amazon\" class=\"wp-image-5293\" title=\"\" srcset=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/Amazon-package-759x1024.jpg 759w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/Amazon-package-222x300.jpg 222w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/Amazon-package-768x1036.jpg 768w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/Amazon-package.jpg 1024w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/Amazon-package-9x12.jpg 9w\" sizes=\"auto, (max-width: 759px) 100vw, 759px\" \/><\/figure>\n\n\n\n<p style=\"font-size:17px\">Ensuite, j'ai t\u00e9l\u00e9charg\u00e9 la photo que j'avais prise sur Nano Banana et je lui ai donn\u00e9 l'instruction suivante :<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p class=\"has-vivid-cyan-blue-color has-text-color has-link-color wp-elements-41ee9a66efaa63aab6aedf0d2e80cce1\"><strong>&#8220;<em>Je veux que tu modifies cette photo, ne change rien sauf l'endroit o\u00f9 se trouve l'\u00e9tiquette, ajoute des dommages et des taches de boue noire sur la bo\u00eete qui ressemblent vraiment \u00e0 des taches de graisse.<\/em>&#8220;<\/strong><\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h3 class=\"wp-block-heading\">Production de Nano Bana :<\/h3>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"747\" height=\"1024\" src=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Amazon-Package-Damaged-747x1024.jpg\" alt=\"Une photo montrant un paquet\" class=\"wp-image-5292\" title=\"\" srcset=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Amazon-Package-Damaged-747x1024.jpg 747w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Amazon-Package-Damaged-219x300.jpg 219w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Amazon-Package-Damaged-768x1052.jpg 768w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Amazon-Package-Damaged-9x12.jpg 9w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Amazon-Package-Damaged.jpg 864w\" sizes=\"auto, (max-width: 747px) 100vw, 747px\" \/><\/figure>\n\n\n\n<style>\n  @import url('https:\/\/fonts.googleapis.com\/css2?family=DM+Sans:wght@400;500;600;700&family=Outfit:wght@600;700;800&display=swap');\n\n  .detector-scores-wrap {\n    max-width: 780px;\n    margin: 2rem auto;\n    font-family: 'DM Sans', sans-serif;\n    -webkit-font-smoothing: antialiased;\n  }\n\n  .detector-scores-wrap h3 {\n    font-family: 'Outfit', sans-serif;\n    font-size: 1.4rem;\n    font-weight: 800;\n    color: #1a1a2e;\n    margin-bottom: 1.25rem;\n    letter-spacing: -0.02em;\n  }\n\n  .detector-scores-table {\n    width: 100%;\n    border-collapse: separate;\n    border-spacing: 0;\n    border-radius: 14px;\n    overflow: hidden;\n    table-layout: fixed;\n    box-shadow:\n      0 1px 3px rgba(0,0,0,0.06),\n      0 8px 24px rgba(0,0,0,0.08),\n      0 0 0 1px rgba(0,0,0,0.04);\n  }\n\n  .detector-scores-table thead th {\n    background: #1a1a2e;\n    color: #ffffff;\n    font-family: 'Outfit', sans-serif;\n    font-weight: 700;\n    font-size: 0.7rem;\n    text-transform: uppercase;\n    letter-spacing: 0.04em;\n    padding: 14px 6px;\n    text-align: center;\n    border: none;\n  }\n\n  .detector-scores-table thead th:first-child {\n    text-align: left;\n    padding-left: 14px;\n    width: 18%;\n  }\n\n  .detector-scores-table thead th:nth-child(2) {\n    width: 12%;\n  }\n\n  .detector-scores-table tbody tr {\n    transition: background 0.2s ease;\n  }\n\n  .detector-scores-table tbody tr:hover {\n    background: rgba(0,0,0,0.02);\n  }\n\n  .detector-scores-table tbody td {\n    padding: 14px 6px;\n    text-align: center;\n    font-size: 0.9rem;\n    font-weight: 500;\n    color: #2d2d3a;\n    border-bottom: 1px solid #eee;\n  }\n\n  .detector-scores-table tbody tr:last-child td {\n    border-bottom: none;\n  }\n\n  .detector-scores-table tbody td:first-child {\n    text-align: left;\n    padding-left: 14px;\n    font-weight: 600;\n    color: #1a1a2e;\n  }\n\n  .score-pill {\n    font-weight: 700;\n    font-family: 'Outfit', sans-serif;\n    font-size: 0.78rem;\n    border-radius: 6px;\n    padding: 5px 8px;\n    display: inline-block;\n    min-width: 0;\n    text-align: center;\n    white-space: nowrap;\n  }\n\n  .score-pass {\n    background: #22c55e;\n    color: #fff;\n  }\n\n  .score-warn {\n    background: #fb923c;\n    color: #fff;\n  }\n\n  .score-fail {\n    background: #ef4444;\n    color: #fff;\n  }\n\n  @media (max-width: 600px) {\n    .detector-scores-table thead th,\n    .detector-scores-table tbody td {\n      padding: 10px 4px;\n      font-size: 0.7rem;\n    }\n    .score-pill {\n      padding: 4px 5px;\n      font-size: 0.68rem;\n    }\n    .detector-scores-wrap h3 {\n      font-size: 1.15rem;\n    }\n  }\n<\/style>\n\n<div class=\"detector-scores-wrap\">\n  <h3>Comment les d\u00e9tecteurs d'images l'ont not\u00e9<\/h3>\n\n  <table class=\"detector-scores-table\">\n    <thead>\n      <tr>\n        <th>Mod\u00e8le d'image<\/th>\n        <th>Cat\u00e9gorie<\/th>\n        <th>TruthScan<\/th>\n        <th>Moteur de vision<\/th>\n        <th>IA ou pas<\/th>\n        <th>Winston AI<\/th>\n        <th>WasItAI<\/th>\n      <\/tr>\n    <\/thead>\n    <tbody>\n      <tr>\n        <td>Nano Banane <span style=\"font-size:0.75rem; color:#888; display:block;\">Deepfake Edit<\/span><\/td>\n        <td>Fraude<\/td>\n        <td><span class=\"score-pill score-pass\">99.00% AI<\/span><\/td>\n        <td><span class=\"score-pill score-pass\">98.00% AI<\/span><\/td>\n        <td><span class=\"score-pill score-pass\">99.10% AI<\/span><\/td>\n        <td><span class=\"score-pill score-fail\">31.98% AI<\/span><\/td>\n        <td><span class=\"score-pill score-pass\">99.00% AI<\/span><\/td>\n      <\/tr>\n    <\/tbody>\n  <\/table>\n<\/div>\n\n\n\n<p style=\"font-size:17px\">La boue semblable \u00e0 de l'huile que Nano Banana a g\u00e9n\u00e9r\u00e9e en sortie avait un reflet crasseux convaincant, avec un effet de d\u00e9trempe - pas du tout cartoonesque. Heureusement, la plupart des d\u00e9tecteurs que j'ai utilis\u00e9s ont correctement identifi\u00e9 l'image comme \u00e9tant un faux. Le seul cas aberrant est celui de Winston, dont les performances sont encore une fois m\u00e9diocres, puisqu'il a mal class\u00e9 l'image en la qualifiant d'authentique. AI or Not, TruthScan et WasItAI ont \u00e9t\u00e9 les d\u00e9tecteurs les plus pr\u00e9cis pour rep\u00e9rer cette image.<\/p>\n\n\n\n<p style=\"font-size:17px\">Il est inqui\u00e9tant de voir \u00e0 quel point il semble simple de t\u00e9l\u00e9charger une image vers un chatbot et de la voir compl\u00e8tement modifi\u00e9e. Je peux imaginer que des fraudeurs et des escrocs utilisent des outils de g\u00e9n\u00e9ration d'images pour tenter de faire des demandes de remboursement frauduleuses ou de se livrer \u00e0 des fraudes sur le march\u00e9.<\/p>\n\n\n\n<p style=\"font-size:17px\">Si les plateformes de commerce \u00e9lectronique n'exigent qu'une image comme preuve pour initier une r\u00e9clamation de colis endommag\u00e9, alors, sans aucun moyen fiable de d\u00e9tecter l'alt\u00e9ration de l'IA, elles s'exposent \u00e0 un vecteur d'attaque majeur. Par exemple, le site officiel d'Amazon <a href=\"https:\/\/sellercentral.amazon.com\/help\/hub\/reference\/external\/GYW2EV9FEMJ3JJGR?locale=en-US\" target=\"_blank\" rel=\"noopener\">politique de retour des articles<\/a> litt\u00e9ralement<strong>, &#8220;<em>Les mati\u00e8res dangereuses, y compris les liquides ou les gaz inflammables, ne sont pas consign\u00e9es.<\/em>.&#8221;<\/strong><\/p>\n\n\n\n<p style=\"font-size:17px\">Ainsi, si vous recevez un colis couvert d'huile ou de boue noire graisseuse, vous n'\u00eates pas oblig\u00e9 de le renvoyer, mais vous pouvez obtenir un remboursement. Si, pour une raison quelconque, cela vous arrivait, Amazon vous demanderait probablement une preuve photographique. Voyez-vous le probl\u00e8me que peuvent poser les images g\u00e9n\u00e9r\u00e9es par l'IA lorsqu'elles sont impossibles \u00e0 distinguer des images r\u00e9elles ?<\/p>\n\n\n\n<h2 class=\"wp-block-heading has-text-align-left\">#4. \"Repas de cafard\" (g\u00e9n\u00e9r\u00e9 avec Nano Banana)<\/h2>\n\n\n\n<p style=\"font-size:17px\">La prochaine image d'IA pourrait vous faire perdre l'app\u00e9tit. Il s'av\u00e8re qu'il existe des cas document\u00e9s de personnes utilisant l'intelligence artificielle pour recevoir des informations sur la sant\u00e9. <a href=\"https:\/\/dailydot.com\/ai-generated-undercooked-food-doordash\" data-type=\"link\" data-id=\"https:\/\/dailydot.com\/ai-generated-undercooked-food-doordash\" target=\"_blank\" rel=\"noopener\">remboursements frauduleux<\/a> sur les applications de livraison de nourriture. Pour ce quatri\u00e8me test, j'ai suivi les m\u00eames \u00e9tapes que pour le troisi\u00e8me. Cette fois, j'ai pris une photo de ma bo\u00eete \u00e0 emporter vide apr\u00e8s le d\u00e9jeuner.<\/p>\n\n\n\n<p><strong>Voici l'image r\u00e9elle :<\/strong><\/p>\n\n\n\n<figure class=\"wp-block-image size-large is-resized\"><img loading=\"lazy\" decoding=\"async\" width=\"768\" height=\"1024\" src=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/IMG_6145-768x1024.jpg\" alt=\"Une photo d&#039;une bo\u00eete de nourriture \u00e0 emporter vide.\" class=\"wp-image-5296\" style=\"width:462px;height:auto\" title=\"\" srcset=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/IMG_6145-768x1024.jpg 768w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/IMG_6145-225x300.jpg 225w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/IMG_6145-scaled.jpg 1024w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/IMG_6145-9x12.jpg 9w\" sizes=\"auto, (max-width: 768px) 100vw, 768px\" \/><\/figure>\n\n\n\n<p style=\"font-size:17px\">J'ai ensuite plac\u00e9 la photo r\u00e9elle que j'ai prise dans Nano Banana et je lui ai donn\u00e9 l'instruction suivante :<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p class=\"has-vivid-cyan-blue-color has-text-color has-link-color wp-elements-e9f00ca4e893e7742bce463148a639f5\"><strong><em>\"Modifiez cette image. Ne changez rien, sauf ajoutez des nouilles \u00e0 l'endroit o\u00f9 se trouve le papier d'aluminium et ajoutez un tas de petits cafards dans la nourriture. Faites en sorte que tout cela ait l'air r\u00e9aliste et non caricatural.\" <\/em><\/strong><\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h3 class=\"wp-block-heading\">La sortie de la nano-banane :<\/h3>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"747\" height=\"1024\" src=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Cockroach-meal.jpg\" alt=\"Une photo modifi\u00e9e par l&#039;IA de la bo\u00eete de nourriture \u00e0 emporter avec des nouilles et des insectes \u00e0 l&#039;int\u00e9rieur de la bo\u00eete.\" class=\"wp-image-5295\" title=\"\" srcset=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Cockroach-meal.jpg 747w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Cockroach-meal-219x300.jpg 219w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Cockroach-meal-9x12.jpg 9w\" sizes=\"auto, (max-width: 747px) 100vw, 747px\" \/><\/figure>\n\n\n\n<style>\n  @import url('https:\/\/fonts.googleapis.com\/css2?family=DM+Sans:wght@400;500;600;700&family=Outfit:wght@600;700;800&display=swap');\n\n  .detector-scores-wrap {\n    max-width: 780px;\n    margin: 2rem auto;\n    font-family: 'DM Sans', sans-serif;\n    -webkit-font-smoothing: antialiased;\n  }\n\n  .detector-scores-wrap h3 {\n    font-family: 'Outfit', sans-serif;\n    font-size: 1.4rem;\n    font-weight: 800;\n    color: #1a1a2e;\n    margin-bottom: 1.25rem;\n    letter-spacing: -0.02em;\n  }\n\n  .detector-scores-table {\n    width: 100%;\n    border-collapse: separate;\n    border-spacing: 0;\n    border-radius: 14px;\n    overflow: hidden;\n    table-layout: fixed;\n    box-shadow:\n      0 1px 3px rgba(0,0,0,0.06),\n      0 8px 24px rgba(0,0,0,0.08),\n      0 0 0 1px rgba(0,0,0,0.04);\n  }\n\n  .detector-scores-table thead th {\n    background: #1a1a2e;\n    color: #ffffff;\n    font-family: 'Outfit', sans-serif;\n    font-weight: 700;\n    font-size: 0.7rem;\n    text-transform: uppercase;\n    letter-spacing: 0.04em;\n    padding: 14px 6px;\n    text-align: center;\n    border: none;\n  }\n\n  .detector-scores-table thead th:first-child {\n    text-align: left;\n    padding-left: 14px;\n    width: 18%;\n  }\n\n  .detector-scores-table thead th:nth-child(2) {\n    width: 12%;\n  }\n\n  .detector-scores-table tbody tr {\n    transition: background 0.2s ease;\n  }\n\n  .detector-scores-table tbody tr:hover {\n    background: rgba(0,0,0,0.02);\n  }\n\n  .detector-scores-table tbody td {\n    padding: 14px 6px;\n    text-align: center;\n    font-size: 0.9rem;\n    font-weight: 500;\n    color: #2d2d3a;\n    border-bottom: 1px solid #eee;\n  }\n\n  .detector-scores-table tbody tr:last-child td {\n    border-bottom: none;\n  }\n\n  .detector-scores-table tbody td:first-child {\n    text-align: left;\n    padding-left: 14px;\n    font-weight: 600;\n    color: #1a1a2e;\n  }\n\n  .score-pill {\n    font-weight: 700;\n    font-family: 'Outfit', sans-serif;\n    font-size: 0.78rem;\n    border-radius: 6px;\n    padding: 5px 8px;\n    display: inline-block;\n    min-width: 0;\n    text-align: center;\n    white-space: nowrap;\n  }\n\n  .score-pass {\n    background: #22c55e;\n    color: #fff;\n  }\n\n  .score-warn {\n    background: #fb923c;\n    color: #fff;\n  }\n\n  .score-fail {\n    background: #ef4444;\n    color: #fff;\n  }\n\n  @media (max-width: 600px) {\n    .detector-scores-table thead th,\n    .detector-scores-table tbody td {\n      padding: 10px 4px;\n      font-size: 0.7rem;\n    }\n    .score-pill {\n      padding: 4px 5px;\n      font-size: 0.68rem;\n    }\n    .detector-scores-wrap h3 {\n      font-size: 1.15rem;\n    }\n  }\n<\/style>\n\n<div class=\"detector-scores-wrap\">\n  <h3>Comment les d\u00e9tecteurs d'images l'ont not\u00e9<\/h3>\n\n  <table class=\"detector-scores-table\">\n    <thead>\n      <tr>\n        <th>Mod\u00e8le d'image<\/th>\n        <th>Cat\u00e9gorie<\/th>\n        <th>TruthScan<\/th>\n        <th>Moteur de vision<\/th>\n        <th>IA ou pas<\/th>\n        <th>Winston AI<\/th>\n        <th>WasItAI<\/th>\n      <\/tr>\n    <\/thead>\n    <tbody>\n      <tr>\n        <td>Nano Banane <span style=\"font-size:0.75rem; color:#888; display:block;\">Deepfake Edit<\/span><\/td>\n        <td>Fraude<\/td>\n        <td><span class=\"score-pill score-pass\">97.15% AI<\/span><\/td>\n        <td><span class=\"score-pill score-fail\">18.00% AI<\/span><\/td>\n        <td><span class=\"score-pill score-warn\">85.87% AI<\/span><\/td>\n        <td><span class=\"score-pill score-warn\">72.29% AI<\/span><\/td>\n        <td><span class=\"score-pill score-fail\">1.00% AI<\/span><\/td>\n      <\/tr>\n    <\/tbody>\n  <\/table>\n<\/div>\n\n\n\n<p style=\"font-size:17px\">Une fois de plus, TruthScan a d\u00e9tect\u00e9 en toute confiance les m\u00e9dias g\u00e9n\u00e9r\u00e9s par l'IA. Sight Engine et WasitAI ont compl\u00e8tement \u00e9chou\u00e9. AI or Not a \u00e9t\u00e9 le deuxi\u00e8me d\u00e9tecteur le plus pr\u00e9cis (avec 85% de certitude d'IA) lors de l'analyse de cette image d'IA, et Winston a obtenu de meilleurs r\u00e9sultats que lors du troisi\u00e8me test (avec 72% de certitude d'IA).<\/p>\n\n\n\n<p style=\"font-size:17px\">Dans cette image, les nouilles et les cafards ont tous \u00e9t\u00e9 enti\u00e8rement g\u00e9n\u00e9r\u00e9s par l'IA ; pourtant, deux des d\u00e9tecteurs ne semblaient pas le penser, et un seul (TruthScan) avait une certitude sup\u00e9rieure \u00e0 90% quant \u00e0 l'alt\u00e9ration par l'IA.<\/p>\n\n\n\n<p style=\"font-size:17px\">Ce qui m'a interpell\u00e9, c'est la simplicit\u00e9 avec laquelle j'ai cr\u00e9\u00e9 cette fausse image. Le t\u00e9l\u00e9chargement de la photo que j'ai prise, le lancement de Nano Banana et la r\u00e9ception du r\u00e9sultat n'ont pris en tout et pour tout que deux minutes. Un cauchemar pour les applications de livraison de nourriture, un r\u00eave pour les menteurs affam\u00e9s. Un \"facteur de fraude\".<\/p>\n\n\n\n<p class=\"has-vivid-red-color has-text-color has-link-color wp-elements-4c89b6ee3afcbdd6b6617c150e5b2e03\"><strong>DIVULGATION : Certaines des images g\u00e9n\u00e9r\u00e9es par l'IA incluses dans les tests suivants avaient pour but de d\u00e9montrer comment les images g\u00e9n\u00e9r\u00e9es par l'IA peuvent \u00eatre utilis\u00e9es \u00e0 des fins de d\u00e9sinformation. Elles contiennent des images controvers\u00e9es et des sujets sensibles<\/strong>, <strong>et n'ont PAS pour but de repr\u00e9senter un parti politique particulier.<\/strong> <strong>des points de vue ou des id\u00e9ologies.<\/strong><\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>#5<\/strong>. \"Autovandalisme\" (g\u00e9n\u00e9r\u00e9 avec Nano Banana)<\/h2>\n\n\n\n<p style=\"font-size:17px\">J'ai trouv\u00e9 une image r\u00e9elle d'une voiture \u00e0 vendre en ligne. Plut\u00f4t banal. Mais que se passerait-il si quelqu'un voulait utiliser cette image pour diffuser de la d\u00e9sinformation ou d\u00e9poser une fausse d\u00e9claration d'assurance automobile ? Malheureusement, l'IA g\u00e9n\u00e9rative rend cela facile et rapide.<\/p>\n\n\n\n<p style=\"font-size:17px\"><strong>Pour ce cinqui\u00e8me test, j'ai pris une image r\u00e9elle d'une voiture et j'ai donn\u00e9 \u00e0 Nano Banana l'instruction suivante :<\/strong><\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p class=\"has-vivid-cyan-blue-color has-text-color has-link-color wp-elements-d06074fe9533ce61c08bcf9750cd398d\">&#8220;<strong><em>Ne changez rien \u00e0 la voiture, sauf que je veux que vous y mettiez de la peinture en a\u00e9rosol avec l'inscription \"Palestine libre\" et que le pare-brise soit fissur\u00e9. Les phares sont cass\u00e9s, le r\u00e9troviseur est bris\u00e9 et pendouille, et les deux fen\u00eatres sur le c\u00f4t\u00e9 de la voiture sont<\/em><\/strong> <strong>\u00e9cras\u00e9<\/strong>.&#8221;<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Voici l'image r\u00e9elle :<\/strong><\/h3>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"940\" height=\"627\" src=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/2000_honda_civic-si_2000_honda_civic-si_bbeed9f1-7b50-4314-a504-40be9411b8ab-mXhf1y-2.webp\" alt=\"Une photo d&#039;une voiture bleue\" class=\"wp-image-5298\" title=\"\" srcset=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/2000_honda_civic-si_2000_honda_civic-si_bbeed9f1-7b50-4314-a504-40be9411b8ab-mXhf1y-2.webp 940w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/2000_honda_civic-si_2000_honda_civic-si_bbeed9f1-7b50-4314-a504-40be9411b8ab-mXhf1y-2-300x200.webp 300w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/2000_honda_civic-si_2000_honda_civic-si_bbeed9f1-7b50-4314-a504-40be9411b8ab-mXhf1y-2-768x512.webp 768w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/2000_honda_civic-si_2000_honda_civic-si_bbeed9f1-7b50-4314-a504-40be9411b8ab-mXhf1y-2-18x12.webp 18w\" sizes=\"auto, (max-width: 940px) 100vw, 940px\" \/><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\">Image de Nano Banana AI :<\/h3>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"683\" src=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Protest-Car-Damage.jpg\" alt=\"Photo de la voiture bleue modifi\u00e9e par l&#039;IA, montrant un pare-brise fissur\u00e9 et des d\u00e9g\u00e2ts de peinture sur le c\u00f4t\u00e9 de la voiture.\" class=\"wp-image-5299\" title=\"\" srcset=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Protest-Car-Damage.jpg 1024w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Protest-Car-Damage-300x200.jpg 300w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Protest-Car-Damage-768x512.jpg 768w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Protest-Car-Damage-18x12.jpg 18w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<style>\n  @import url('https:\/\/fonts.googleapis.com\/css2?family=DM+Sans:wght@400;500;600;700&family=Outfit:wght@600;700;800&display=swap');\n\n  .detector-scores-wrap {\n    max-width: 780px;\n    margin: 2rem auto;\n    font-family: 'DM Sans', sans-serif;\n    -webkit-font-smoothing: antialiased;\n  }\n\n  .detector-scores-wrap h3 {\n    font-family: 'Outfit', sans-serif;\n    font-size: 1.4rem;\n    font-weight: 800;\n    color: #1a1a2e;\n    margin-bottom: 1.25rem;\n    letter-spacing: -0.02em;\n  }\n\n  .detector-scores-table {\n    width: 100%;\n    border-collapse: separate;\n    border-spacing: 0;\n    border-radius: 14px;\n    overflow: hidden;\n    table-layout: fixed;\n    box-shadow:\n      0 1px 3px rgba(0,0,0,0.06),\n      0 8px 24px rgba(0,0,0,0.08),\n      0 0 0 1px rgba(0,0,0,0.04);\n  }\n\n  .detector-scores-table thead th {\n    background: #1a1a2e;\n    color: #ffffff;\n    font-family: 'Outfit', sans-serif;\n    font-weight: 700;\n    font-size: 0.7rem;\n    text-transform: uppercase;\n    letter-spacing: 0.04em;\n    padding: 14px 6px;\n    text-align: center;\n    border: none;\n  }\n\n  .detector-scores-table thead th:first-child {\n    text-align: left;\n    padding-left: 14px;\n    width: 18%;\n  }\n\n  .detector-scores-table thead th:nth-child(2) {\n    width: 14%;\n  }\n\n  .detector-scores-table tbody tr {\n    transition: background 0.2s ease;\n  }\n\n  .detector-scores-table tbody tr:hover {\n    background: rgba(0,0,0,0.02);\n  }\n\n  .detector-scores-table tbody td {\n    padding: 14px 6px;\n    text-align: center;\n    font-size: 0.9rem;\n    font-weight: 500;\n    color: #2d2d3a;\n    border-bottom: 1px solid #eee;\n  }\n\n  .detector-scores-table tbody tr:last-child td {\n    border-bottom: none;\n  }\n\n  .detector-scores-table tbody td:first-child {\n    text-align: left;\n    padding-left: 14px;\n    font-weight: 600;\n    color: #1a1a2e;\n  }\n\n  .score-pill {\n    font-weight: 700;\n    font-family: 'Outfit', sans-serif;\n    font-size: 0.78rem;\n    border-radius: 6px;\n    padding: 5px 8px;\n    display: inline-block;\n    min-width: 0;\n    text-align: center;\n    white-space: nowrap;\n  }\n\n  .score-pass {\n    background: #22c55e;\n    color: #fff;\n  }\n\n  .score-warn {\n    background: #fb923c;\n    color: #fff;\n  }\n\n  .score-fail {\n    background: #ef4444;\n    color: #fff;\n  }\n\n  @media (max-width: 600px) {\n    .detector-scores-table thead th,\n    .detector-scores-table tbody td {\n      padding: 10px 4px;\n      font-size: 0.7rem;\n    }\n    .score-pill {\n      padding: 4px 5px;\n      font-size: 0.68rem;\n    }\n    .detector-scores-wrap h3 {\n      font-size: 1.15rem;\n    }\n  }\n<\/style>\n\n<div class=\"detector-scores-wrap\">\n  <h3>Comment les d\u00e9tecteurs d'images l'ont not\u00e9<\/h3>\n\n  <table class=\"detector-scores-table\">\n    <thead>\n      <tr>\n        <th>Mod\u00e8le d'image<\/th>\n        <th>Cat\u00e9gorie<\/th>\n        <th>TruthScan<\/th>\n        <th>Moteur de vision<\/th>\n        <th>IA ou pas<\/th>\n        <th>Winston AI<\/th>\n        <th>WasItAI<\/th>\n      <\/tr>\n    <\/thead>\n    <tbody>\n      <tr>\n        <td>Nano Banane <span style=\"font-size:0.75rem; color:#888; display:block;\">Deepfake Edit<\/span><\/td>\n        <td style=\"font-size:0.82rem;\">Fraude \/ D\u00e9sinformation<\/td>\n        <td><span class=\"score-pill score-pass\">97.48% AI<\/span><\/td>\n        <td><span class=\"score-pill score-fail\">18.00% AI<\/span><\/td>\n        <td><span class=\"score-pill score-warn\">89.52% AI<\/span><\/td>\n        <td><span class=\"score-pill score-fail\">0,02% AI<\/span><\/td>\n        <td><span class=\"score-pill score-pass\">99.00% AI<\/span><\/td>\n      <\/tr>\n    <\/tbody>\n  <\/table>\n<\/div>\n\n\n\n<p style=\"font-size:17px\">Les r\u00e9sultats des d\u00e9tecteurs pour cette sortie sont mitig\u00e9s. Pour ce test, WasItAI a (\u00e9tonnamment) obtenu le meilleur score de d\u00e9tection de l'IA (99%). TruthScan a obtenu le deuxi\u00e8me meilleur score (97%). Sight Engine n'a pas r\u00e9ussi \u00e0 d\u00e9tecter d'\u00e9l\u00e9ments synth\u00e9tiques dans l'image, et Winston a obtenu le pire score de tous les d\u00e9tecteurs (moins de 1% de chances d'implication). AI or Not s'est bien comport\u00e9, mais la pr\u00e9cision est rest\u00e9e inf\u00e9rieure \u00e0 90%.<br><br>Parlons de l'image. Les deux aspects les plus pr\u00e9occupants de ce type de contenu cr\u00e9\u00e9 ou modifi\u00e9 par l'IA sont qu'ils impliquent tous deux une tromperie. Tout d'abord, l'utilisation de l'IA pour modifier des images banales afin de cr\u00e9er un contenu \u00e0 connotation politique est rapide, facile et semble r\u00e9elle. Que les auteurs de ce type de contenu cherchent simplement \u00e0 susciter l'engagement ou qu'ils soient des provocateurs politiques, le probl\u00e8me est qu'il s'agit de fausses informations pr\u00e9sent\u00e9es comme authentiques.<\/p>\n\n\n\n<p style=\"font-size:17px\">La deuxi\u00e8me fa\u00e7on d'abuser des outils de g\u00e9n\u00e9ration d'images est la fraude. Si un escroc veut d\u00e9poser une fausse demande d'assurance automobile, il devra produire de fausses preuves. Au lieu de passer des heures sur Photoshop \u00e0 retoucher des images et \u00e0 falsifier des preuves, il peut utiliser des g\u00e9n\u00e9rateurs d'images pour produire rapidement de fausses preuves.<\/p>\n\n\n\n<p style=\"font-size:17px\">Dans l'exemple de l'image d'IA ci-dessus, remarquez qu'un escroc pourrait utiliser un processus similaire pour d\u00e9poser une fausse demande d'indemnisation, conform\u00e9ment \u00e0 l'avis de Progressive. <a href=\"https:\/\/www.progressive.com\/answers\/does-car-insurance-cover-vandalism\/\" data-type=\"link\" data-id=\"https:\/\/www.progressive.com\/answers\/does-car-insurance-cover-vandalism\/\" target=\"_blank\" rel=\"noopener\">Politique de couverture globale<\/a>qui couvre sp\u00e9cifiquement<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>R\u00e9duit ou&nbsp;<a href=\"https:\/\/www.progressive.com\/answers\/does-car-insurance-cover-tire-damage\/\" target=\"_blank\" rel=\"noopener\">pneus endommag\u00e9s<\/a><\/strong><\/li>\n\n\n\n<li><strong>Vitres, phares ou feux arri\u00e8re cass\u00e9s<\/strong><\/li>\n\n\n\n<li><strong>Dommages caus\u00e9s par la peinture en a\u00e9rosol<\/strong><\/li>\n\n\n\n<li><strong><a href=\"https:\/\/www.progressive.com\/answers\/does-car-insurance-cover-scratches\/\" target=\"_blank\" rel=\"noopener\">Bosses ou rayures<\/a>&nbsp;de quelqu'un qui met la cl\u00e9 dans votre voiture<\/strong><\/li>\n\n\n\n<li><strong>Mettre du sucre ou d'autres substances dans le r\u00e9servoir d'essence<\/strong><\/li>\n<\/ul>\n\n\n\n<p>Bien entendu, pr\u00e9senter de fausses preuves n'est pas seulement trompeur, c'est aussi tout simplement ill\u00e9gal. Dans un monde id\u00e9al, personne n'enfreindrait la loi ; en r\u00e9alit\u00e9, la criminalit\u00e9 est quotidienne. Et sans d\u00e9tection fiable, les compagnies d'assurance automobile risquent de se faire escroquer de plusieurs millions d'euros \u00e0 mesure que les demandes d'indemnisation frauduleuses fond\u00e9es sur l'IA se multiplient.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">#6. \"Soda Musk and Don\" (g\u00e9n\u00e9r\u00e9 avec Nano Banana)<\/h2>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"559\" src=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Elon-Spilling-Soda-1024x559.jpeg\" alt=\"Une image d&#039;Elon Musk et de Donald Trump g\u00e9n\u00e9r\u00e9e par l&#039;IA\" class=\"wp-image-5300\" title=\"\" srcset=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Elon-Spilling-Soda-1024x559.jpeg 1024w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Elon-Spilling-Soda-300x164.jpeg 300w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Elon-Spilling-Soda-768x419.jpeg 768w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Elon-Spilling-Soda-1536x838.jpeg 1536w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Elon-Spilling-Soda-2048x1117.jpeg 1877w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Elon-Spilling-Soda-18x10.jpeg 18w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Elon-Spilling-Soda-scaled.jpeg 1878w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<style>\n  @import url('https:\/\/fonts.googleapis.com\/css2?family=DM+Sans:wght@400;500;600;700&family=Outfit:wght@600;700;800&display=swap');\n\n  .detector-scores-wrap {\n    max-width: 780px;\n    margin: 2rem auto;\n    font-family: 'DM Sans', sans-serif;\n    -webkit-font-smoothing: antialiased;\n  }\n\n  .detector-scores-wrap h3 {\n    font-family: 'Outfit', sans-serif;\n    font-size: 1.4rem;\n    font-weight: 800;\n    color: #1a1a2e;\n    margin-bottom: 1.25rem;\n    letter-spacing: -0.02em;\n  }\n\n  .detector-scores-table {\n    width: 100%;\n    border-collapse: separate;\n    border-spacing: 0;\n    border-radius: 14px;\n    overflow: hidden;\n    table-layout: fixed;\n    box-shadow:\n      0 1px 3px rgba(0,0,0,0.06),\n      0 8px 24px rgba(0,0,0,0.08),\n      0 0 0 1px rgba(0,0,0,0.04);\n  }\n\n  .detector-scores-table thead th {\n    background: #1a1a2e;\n    color: #ffffff;\n    font-family: 'Outfit', sans-serif;\n    font-weight: 700;\n    font-size: 0.7rem;\n    text-transform: uppercase;\n    letter-spacing: 0.04em;\n    padding: 14px 6px;\n    text-align: center;\n    border: none;\n  }\n\n  .detector-scores-table thead th:first-child {\n    text-align: left;\n    padding-left: 14px;\n    width: 18%;\n  }\n\n  .detector-scores-table thead th:nth-child(2) {\n    width: 14%;\n  }\n\n  .detector-scores-table tbody tr {\n    transition: background 0.2s ease;\n  }\n\n  .detector-scores-table tbody tr:hover {\n    background: rgba(0,0,0,0.02);\n  }\n\n  .detector-scores-table tbody td {\n    padding: 14px 6px;\n    text-align: center;\n    font-size: 0.9rem;\n    font-weight: 500;\n    color: #2d2d3a;\n    border-bottom: 1px solid #eee;\n  }\n\n  .detector-scores-table tbody tr:last-child td {\n    border-bottom: none;\n  }\n\n  .detector-scores-table tbody td:first-child {\n    text-align: left;\n    padding-left: 14px;\n    font-weight: 600;\n    color: #1a1a2e;\n  }\n\n  .score-pill {\n    font-weight: 700;\n    font-family: 'Outfit', sans-serif;\n    font-size: 0.78rem;\n    border-radius: 6px;\n    padding: 5px 8px;\n    display: inline-block;\n    min-width: 0;\n    text-align: center;\n    white-space: nowrap;\n  }\n\n  .score-pass {\n    background: #22c55e;\n    color: #fff;\n  }\n\n  .score-warn {\n    background: #fb923c;\n    color: #fff;\n  }\n\n  .score-fail {\n    background: #ef4444;\n    color: #fff;\n  }\n\n  @media (max-width: 600px) {\n    .detector-scores-table thead th,\n    .detector-scores-table tbody td {\n      padding: 10px 4px;\n      font-size: 0.7rem;\n    }\n    .score-pill {\n      padding: 4px 5px;\n      font-size: 0.68rem;\n    }\n    .detector-scores-wrap h3 {\n      font-size: 1.15rem;\n    }\n  }\n<\/style>\n\n<div class=\"detector-scores-wrap\">\n  <h3>Comment les d\u00e9tecteurs d'images l'ont not\u00e9<\/h3>\n\n  <table class=\"detector-scores-table\">\n    <thead>\n      <tr>\n        <th>Mod\u00e8le d'image<\/th>\n        <th>Cat\u00e9gorie<\/th>\n        <th>TruthScan<\/th>\n        <th>Moteur de vision<\/th>\n        <th>IA ou pas<\/th>\n        <th>Winston AI<\/th>\n        <th>WasItAI<\/th>\n      <\/tr>\n    <\/thead>\n    <tbody>\n      <tr>\n        <td>Nano Banane <span style=\"font-size:0.75rem; color:#888; display:block;\">G\u00e9n\u00e9ration compl\u00e8te<\/span><\/td>\n        <td style=\"font-size:0.82rem;\">D\u00e9sinformation<\/td>\n        <td><span class=\"score-pill score-pass\">99.00% AI<\/span><\/td>\n        <td><span class=\"score-pill score-pass\">99.00% AI<\/span><\/td>\n        <td><span class=\"score-pill score-pass\">99.00% AI<\/span><\/td>\n        <td><span class=\"score-pill score-fail\">0,24% AI<\/span><\/td>\n        <td><span class=\"score-pill score-warn\">86.00% AI<\/span><\/td>\n      <\/tr>\n    <\/tbody>\n  <\/table>\n<\/div>\n\n\n\n<p style=\"font-size:17px\">Une fois de plus, j'ai utilis\u00e9 Nano Banana pour ce test, et l'objectif \u00e9tait de montrer de fausses images politiquement charg\u00e9es avec un sous-entendu plus stupide. TruthScan, Sight Engine et AIorNot ont tous signal\u00e9 l'image g\u00e9n\u00e9r\u00e9e avec une note de d\u00e9tection d'IA de 99%. WasitAI n'\u00e9tait pas aussi confiant - il n'a d\u00e9tect\u00e9 l'image g\u00e9n\u00e9r\u00e9e par l'IA qu'avec une certitude de 86%. Enfin, Winston a compl\u00e8tement \u00e9chou\u00e9, attribuant \u00e0 l'IA un score de 0,2%. Pour g\u00e9n\u00e9rer l'image, j'ai donn\u00e9 \u00e0 Nano Banana l'instruction suivante :<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p class=\"has-vivid-cyan-blue-color has-text-color has-link-color wp-elements-4f7e002ecaefec24bbebdca3101fff42\"><strong>&#8220;<em>Cr\u00e9ez une photo d'Elon Musk renversant un grand gobelet de soda comme un Big Gulp sur sa chemise et paniquant tandis que Donald Trump est assis \u00e0 c\u00f4t\u00e9 de lui en train de rire. Ils sont dans un avion \u00e0 r\u00e9action. Voil\u00e0 la sc\u00e8ne. La photo doit avoir l'air r\u00e9aliste, comme si elle avait \u00e9t\u00e9 prise avec un appareil photo instantan\u00e9, et inclure des artefacts tels que le reflet du flash de l'appareil sur le hublot de l'avion et \u00e0 l'ext\u00e9rieur de l'avion. Il fait nuit. La photo doit \u00eatre hyperr\u00e9aliste.<\/em>&#8220;<\/strong><\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p style=\"font-size:17px\">Chaque fois qu'un de ces logiciels ne parvient pas \u00e0 d\u00e9tecter l'IA, il le qualifie d'authentique. Aussi stupide que soit cet exemple d'image tir\u00e9e du test, il n'y a rien d'amusant \u00e0 penser qu'un deepfake est r\u00e9el.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">#7. \"Still Alive\" (g\u00e9n\u00e9r\u00e9 par Nano Banana)<\/h2>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"765\" height=\"1024\" src=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3Christian_Ransom_Photo.jpg\" alt=\"Une image g\u00e9n\u00e9r\u00e9e par l&#039;IA d&#039;un homme tenant une pancarte sur laquelle est \u00e9crit &quot;help&quot; (aide).\" class=\"wp-image-5301\" title=\"\" srcset=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3Christian_Ransom_Photo.jpg 765w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3Christian_Ransom_Photo-224x300.jpg 224w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3Christian_Ransom_Photo-9x12.jpg 9w\" sizes=\"auto, (max-width: 765px) 100vw, 765px\" \/><\/figure>\n\n\n\n<style>\n  @import url('https:\/\/fonts.googleapis.com\/css2?family=DM+Sans:wght@400;500;600;700&family=Outfit:wght@600;700;800&display=swap');\n\n  .detector-scores-wrap {\n    max-width: 780px;\n    margin: 2rem auto;\n    font-family: 'DM Sans', sans-serif;\n    -webkit-font-smoothing: antialiased;\n  }\n\n  .detector-scores-wrap h3 {\n    font-family: 'Outfit', sans-serif;\n    font-size: 1.4rem;\n    font-weight: 800;\n    color: #1a1a2e;\n    margin-bottom: 1.25rem;\n    letter-spacing: -0.02em;\n  }\n\n  .detector-scores-table {\n    width: 100%;\n    border-collapse: separate;\n    border-spacing: 0;\n    border-radius: 14px;\n    overflow: hidden;\n    table-layout: fixed;\n    box-shadow:\n      0 1px 3px rgba(0,0,0,0.06),\n      0 8px 24px rgba(0,0,0,0.08),\n      0 0 0 1px rgba(0,0,0,0.04);\n  }\n\n  .detector-scores-table thead th {\n    background: #1a1a2e;\n    color: #ffffff;\n    font-family: 'Outfit', sans-serif;\n    font-weight: 700;\n    font-size: 0.7rem;\n    text-transform: uppercase;\n    letter-spacing: 0.04em;\n    padding: 14px 6px;\n    text-align: center;\n    border: none;\n  }\n\n  .detector-scores-table thead th:first-child {\n    text-align: left;\n    padding-left: 14px;\n    width: 18%;\n  }\n\n  .detector-scores-table thead th:nth-child(2) {\n    width: 14%;\n  }\n\n  .detector-scores-table tbody tr {\n    transition: background 0.2s ease;\n  }\n\n  .detector-scores-table tbody tr:hover {\n    background: rgba(0,0,0,0.02);\n  }\n\n  .detector-scores-table tbody td {\n    padding: 14px 6px;\n    text-align: center;\n    font-size: 0.9rem;\n    font-weight: 500;\n    color: #2d2d3a;\n    border-bottom: 1px solid #eee;\n  }\n\n  .detector-scores-table tbody tr:last-child td {\n    border-bottom: none;\n  }\n\n  .detector-scores-table tbody td:first-child {\n    text-align: left;\n    padding-left: 14px;\n    font-weight: 600;\n    color: #1a1a2e;\n  }\n\n  .score-pill {\n    font-weight: 700;\n    font-family: 'Outfit', sans-serif;\n    font-size: 0.78rem;\n    border-radius: 6px;\n    padding: 5px 8px;\n    display: inline-block;\n    min-width: 0;\n    text-align: center;\n    white-space: nowrap;\n  }\n\n  .score-pass {\n    background: #22c55e;\n    color: #fff;\n  }\n\n  .score-warn {\n    background: #fb923c;\n    color: #fff;\n  }\n\n  .score-fail {\n    background: #ef4444;\n    color: #fff;\n  }\n\n  @media (max-width: 600px) {\n    .detector-scores-table thead th,\n    .detector-scores-table tbody td {\n      padding: 10px 4px;\n      font-size: 0.7rem;\n    }\n    .score-pill {\n      padding: 4px 5px;\n      font-size: 0.68rem;\n    }\n    .detector-scores-wrap h3 {\n      font-size: 1.15rem;\n    }\n  }\n<\/style>\n\n<div class=\"detector-scores-wrap\">\n  <h3>Comment les d\u00e9tecteurs d'images l'ont not\u00e9<\/h3>\n\n  <table class=\"detector-scores-table\">\n    <thead>\n      <tr>\n        <th>Mod\u00e8le d'image<\/th>\n        <th>Cat\u00e9gorie<\/th>\n        <th>TruthScan<\/th>\n        <th>Moteur de vision<\/th>\n        <th>IA ou pas<\/th>\n        <th>Winston AI<\/th>\n        <th>WasItAI<\/th>\n      <\/tr>\n    <\/thead>\n    <tbody>\n      <tr>\n        <td>Nano Banane <span style=\"font-size:0.75rem; color:#888; display:block;\">G\u00e9n\u00e9ration compl\u00e8te<\/span><\/td>\n        <td style=\"font-size:0.82rem;\">Faux-semblant<\/td>\n        <td><span class=\"score-pill score-pass\">97.49% AI<\/span><\/td>\n        <td><span class=\"score-pill score-pass\">95.00% AI<\/span><\/td>\n        <td><span class=\"score-pill score-pass\">99.27% AI<\/span><\/td>\n        <td><span class=\"score-pill score-fail\">0,15% AI<\/span><\/td>\n        <td><span class=\"score-pill score-fail\">1.00% AI<\/span><\/td>\n      <\/tr>\n    <\/tbody>\n  <\/table>\n<\/div>\n\n\n\n<p style=\"font-size:17px\">Trouvez-vous cette image d\u00e9rangeante ? Pour g\u00e9n\u00e9rer ce deepfake, j'ai utilis\u00e9 Nano Banana et je lui ai donn\u00e9 une photo de moi. Ensuite, je lui ai donn\u00e9 une longue invite destin\u00e9e \u00e0 me mettre dans une situation qui impliquait que j'avais \u00e9t\u00e9 enlev\u00e9. C'est peut-\u00eatre parce que cette photo montre <em>moi<\/em> dans une position compromettante. Et si quelqu'un avait fait quelque chose comme \u00e7a, l'avait envoy\u00e9 \u00e0 ma m\u00e8re et l'avait menac\u00e9e de me faire du mal si elle ne lui envoyait pas d'argent ? C'est effrayant. <\/p>\n\n\n\n<p class=\"has-small-font-size\"><em>(L'invite compl\u00e8te que j'ai utilis\u00e9e \u00e0 cet effet se trouve dans l'ensemble des donn\u00e9es de test \u00e0 la fin de cet article).<\/em><\/p>\n\n\n\n<p style=\"font-size:17px\"><br>Heureusement pour cette image, bien qu'elle ait un aspect tr\u00e8s r\u00e9aliste, <strong>trois d\u00e9tecteurs sur cinq l'ont identifi\u00e9 comme \u00e9tant g\u00e9n\u00e9r\u00e9 par l'IA.<\/strong> <\/p>\n\n\n\n<p style=\"font-size:17px\">AI or Not a obtenu un excellent score de 99%, TruthScan 97% et Sight Engine 95%. Malheureusement, Winston et WasItAI ont class\u00e9 l'image comme r\u00e9elle.<\/p>\n\n\n\n<p style=\"font-size:17px\">Je tiens \u00e0 pr\u00e9ciser que ces tests n'ont pas \u00e9t\u00e9 effectu\u00e9s sur mon compte Gemini personnel. Nano Banana ne m'a jamais demand\u00e9 de prouver que j'\u00e9tais la personne sur l'image. N'importe qui aurait pu t\u00e9l\u00e9charger une photo de moi sur Internet, la mettre dans Nano Banana et lui faire cr\u00e9er ce type d'image. <\/p>\n\n\n\n<h2 class=\"wp-block-heading\">#8. \"Abuse of a Ben\" (g\u00e9n\u00e9r\u00e9 avec Nano Banana)<\/h2>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"200\" height=\"200\" src=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/c945d528-730f-48c1-bdca-30e5799fc1aa_thumb-1.jpg\" alt=\"\" class=\"wp-image-5307\" title=\"\" srcset=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/c945d528-730f-48c1-bdca-30e5799fc1aa_thumb-1.jpg 200w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/c945d528-730f-48c1-bdca-30e5799fc1aa_thumb-1-150x150.jpg 150w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/c945d528-730f-48c1-bdca-30e5799fc1aa_thumb-1-12x12.jpg 12w\" sizes=\"auto, (max-width: 200px) 100vw, 200px\" \/><\/figure>\n<\/div>\n\n\n<p style=\"font-size:17px\">Voici Ben. Il passe le plus clair de son temps \u00e0 g\u00e9rer les diff\u00e9rents services des entreprises dans lesquelles il travaille. L'image ci-dessus est la (vraie) photo d'identit\u00e9 de Ben. Si vous ne le voyez pas, Ben est un professionnel. Quelles que soient ses opinions politiques, il les garde pour lui, et ses conversations professionnelles ne portent que sur le travail. Mais \u00e0 tout moment, l'image de Ben pourrait \u00eatre utilis\u00e9e pour diffuser de la propagande ou des messages politiques sans son consentement.<\/p>\n\n\n\n<p style=\"font-size:17px\">On a beaucoup parl\u00e9 r\u00e9cemment de la diffusion en ligne de vid\u00e9os g\u00e9n\u00e9r\u00e9es par l'IA impliquant des agents de l'ICE et des manifestations. Sur la base de tous les tests effectu\u00e9s jusqu'\u00e0 pr\u00e9sent, j'ai conclu qu'il \u00e9tait tr\u00e8s facile de cr\u00e9er de la fausse propagande politique \u00e0 l'aide de l'IA. J'ai donc d\u00e9cid\u00e9 de refaire un test, cette fois-ci avec Ben. <\/p>\n\n\n\n<p style=\"font-size:17px\">Avec la permission de Ben, j'ai utilis\u00e9 sa photo pour cr\u00e9er un deepfake. <strong>Voici l'image g\u00e9n\u00e9r\u00e9e par l'IA que Nano Banana a cr\u00e9\u00e9e :<\/strong><\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full is-resized\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"1024\" src=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/Gemini_Generated_Image_qd4x5sqd4x5sqd4x.jpg\" alt=\"Une photo modifi\u00e9e par l&#039;IA d&#039;un homme nomm\u00e9 Ben, tenant une pancarte sur laquelle est \u00e9crit &quot;Abolir l&#039;ICE&quot;.\" class=\"wp-image-5308\" style=\"width:537px;height:auto\" title=\"\" srcset=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/Gemini_Generated_Image_qd4x5sqd4x5sqd4x.jpg 1024w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/Gemini_Generated_Image_qd4x5sqd4x5sqd4x-300x300.jpg 300w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/Gemini_Generated_Image_qd4x5sqd4x5sqd4x-150x150.jpg 150w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/Gemini_Generated_Image_qd4x5sqd4x5sqd4x-768x768.jpg 768w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/Gemini_Generated_Image_qd4x5sqd4x5sqd4x-12x12.jpg 12w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n<\/div>\n\n\n<style>\n  @import url('https:\/\/fonts.googleapis.com\/css2?family=DM+Sans:wght@400;500;600;700&family=Outfit:wght@600;700;800&display=swap');\n\n  .detector-scores-wrap {\n    max-width: 780px;\n    margin: 2rem auto;\n    font-family: 'DM Sans', sans-serif;\n    -webkit-font-smoothing: antialiased;\n  }\n\n  .detector-scores-wrap h3 {\n    font-family: 'Outfit', sans-serif;\n    font-size: 1.4rem;\n    font-weight: 800;\n    color: #1a1a2e;\n    margin-bottom: 1.25rem;\n    letter-spacing: -0.02em;\n  }\n\n  .detector-scores-table {\n    width: 100%;\n    border-collapse: separate;\n    border-spacing: 0;\n    border-radius: 14px;\n    overflow: hidden;\n    table-layout: fixed;\n    box-shadow:\n      0 1px 3px rgba(0,0,0,0.06),\n      0 8px 24px rgba(0,0,0,0.08),\n      0 0 0 1px rgba(0,0,0,0.04);\n  }\n\n  .detector-scores-table thead th {\n    background: #1a1a2e;\n    color: #ffffff;\n    font-family: 'Outfit', sans-serif;\n    font-weight: 700;\n    font-size: 0.7rem;\n    text-transform: uppercase;\n    letter-spacing: 0.04em;\n    padding: 14px 6px;\n    text-align: center;\n    border: none;\n  }\n\n  .detector-scores-table thead th:first-child {\n    text-align: left;\n    padding-left: 14px;\n    width: 18%;\n  }\n\n  .detector-scores-table thead th:nth-child(2) {\n    width: 14%;\n  }\n\n  .detector-scores-table tbody tr {\n    transition: background 0.2s ease;\n  }\n\n  .detector-scores-table tbody tr:hover {\n    background: rgba(0,0,0,0.02);\n  }\n\n  .detector-scores-table tbody td {\n    padding: 14px 6px;\n    text-align: center;\n    font-size: 0.9rem;\n    font-weight: 500;\n    color: #2d2d3a;\n    border-bottom: 1px solid #eee;\n  }\n\n  .detector-scores-table tbody tr:last-child td {\n    border-bottom: none;\n  }\n\n  .detector-scores-table tbody td:first-child {\n    text-align: left;\n    padding-left: 14px;\n    font-weight: 600;\n    color: #1a1a2e;\n  }\n\n  .score-pill {\n    font-weight: 700;\n    font-family: 'Outfit', sans-serif;\n    font-size: 0.78rem;\n    border-radius: 6px;\n    padding: 5px 8px;\n    display: inline-block;\n    min-width: 0;\n    text-align: center;\n    white-space: nowrap;\n  }\n\n  .score-pass {\n    background: #22c55e;\n    color: #fff;\n  }\n\n  .score-warn {\n    background: #fb923c;\n    color: #fff;\n  }\n\n  .score-fail {\n    background: #ef4444;\n    color: #fff;\n  }\n\n  @media (max-width: 600px) {\n    .detector-scores-table thead th,\n    .detector-scores-table tbody td {\n      padding: 10px 4px;\n      font-size: 0.7rem;\n    }\n    .score-pill {\n      padding: 4px 5px;\n      font-size: 0.68rem;\n    }\n    .detector-scores-wrap h3 {\n      font-size: 1.15rem;\n    }\n  }\n<\/style>\n\n<div class=\"detector-scores-wrap\">\n  <h3>Comment les d\u00e9tecteurs d'images l'ont not\u00e9<\/h3>\n\n  <table class=\"detector-scores-table\">\n    <thead>\n      <tr>\n        <th>Mod\u00e8le d'image<\/th>\n        <th>Cat\u00e9gorie<\/th>\n        <th>TruthScan<\/th>\n        <th>Moteur de vision<\/th>\n        <th>IA ou pas<\/th>\n        <th>Winston AI<\/th>\n        <th>WasItAI<\/th>\n      <\/tr>\n    <\/thead>\n    <tbody>\n      <tr>\n        <td>Nano Banane<\/td>\n        <td style=\"font-size:0.78rem;\">Deepfake \/ Disinfo<\/td>\n        <td><span class=\"score-pill score-pass\">99.00% AI<\/span><\/td>\n        <td><span class=\"score-pill score-pass\">97.00% AI<\/span><\/td>\n        <td><span class=\"score-pill score-pass\">98.88% AI<\/span><\/td>\n        <td><span class=\"score-pill score-fail\">5.74% AI<\/span><\/td>\n        <td><span class=\"score-pill score-pass\">99.00% AI<\/span><\/td>\n      <\/tr>\n    <\/tbody>\n  <\/table>\n<\/div>\n\n\n\n<p style=\"font-size:17px\">Cela a pris trente secondes. Tout ce que j'ai eu \u00e0 faire, c'est de glisser et de d\u00e9poser sa photo dans Gemini et de lui demander de \"brandir une pancarte indiquant l'abolition de l'ICE\". Le r\u00e9sultat de Nano Banana n'est pas aussi photor\u00e9aliste que les autres, mais il est suffisamment authentique pour \u00eatre convaincant.<\/p>\n\n\n\n<p style=\"font-size:17px\"><span style=\"box-sizing: border-box; margin: 0px; padding: 0px;\">Heureusement, quatre des d\u00e9tecteurs ont identifi\u00e9 l'image avec pr\u00e9cision :\u00a0<strong>TruthScan<\/strong> (9<\/span>9% AI), Was It AI (99% AI), Sight Engine (97% AI), et AI or Not (97% AI). Winston AI a \u00e9chou\u00e9, n'obtenant qu'un score de 5% AI.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">#9. \"Club Void\" (g\u00e9n\u00e9r\u00e9 par Midjourney)<\/h2>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"771\" src=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/MJ-Girl-in-club-1024x771.jpg\" alt=\"Une femme souriante v\u00eatue d&#039;une robe blanche \u00e0 manches courtes se tient dans une pi\u00e8ce peu \u00e9clair\u00e9e et bond\u00e9e. La photo semble avoir \u00e9t\u00e9 prise avec un flash, cr\u00e9ant une atmosph\u00e8re brumeuse et bleut\u00e9e avec de faibles particules visibles dans l&#039;air. D&#039;autres personnes en tenue d\u00e9contract\u00e9e se tiennent \u00e0 l&#039;arri\u00e8re-plan, sous un plafond endommag\u00e9 ou d&#039;aspect industriel.\" class=\"wp-image-5309\" title=\"\" srcset=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/MJ-Girl-in-club-1024x771.jpg 1024w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/MJ-Girl-in-club-300x226.jpg 300w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/MJ-Girl-in-club-768x578.jpg 768w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/MJ-Girl-in-club-16x12.jpg 16w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/MJ-Girl-in-club.jpg 1232w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<style>\n  @import url('https:\/\/fonts.googleapis.com\/css2?family=DM+Sans:wght@400;500;600;700&family=Outfit:wght@600;700;800&display=swap');\n\n  .detector-scores-wrap {\n    max-width: 780px;\n    margin: 2rem auto;\n    font-family: 'DM Sans', sans-serif;\n    -webkit-font-smoothing: antialiased;\n  }\n\n  .detector-scores-wrap h3 {\n    font-family: 'Outfit', sans-serif;\n    font-size: 1.4rem;\n    font-weight: 800;\n    color: #1a1a2e;\n    margin-bottom: 1.25rem;\n    letter-spacing: -0.02em;\n  }\n\n  .detector-scores-table {\n    width: 100%;\n    border-collapse: separate;\n    border-spacing: 0;\n    border-radius: 14px;\n    overflow: hidden;\n    table-layout: fixed;\n    box-shadow:\n      0 1px 3px rgba(0,0,0,0.06),\n      0 8px 24px rgba(0,0,0,0.08),\n      0 0 0 1px rgba(0,0,0,0.04);\n  }\n\n  .detector-scores-table thead th {\n    background: #1a1a2e;\n    color: #ffffff;\n    font-family: 'Outfit', sans-serif;\n    font-weight: 700;\n    font-size: 0.7rem;\n    text-transform: uppercase;\n    letter-spacing: 0.04em;\n    padding: 14px 6px;\n    text-align: center;\n    border: none;\n  }\n\n  .detector-scores-table thead th:first-child {\n    text-align: left;\n    padding-left: 14px;\n    width: 18%;\n  }\n\n  .detector-scores-table thead th:nth-child(2) {\n    width: 14%;\n  }\n\n  .detector-scores-table tbody tr {\n    transition: background 0.2s ease;\n  }\n\n  .detector-scores-table tbody tr:hover {\n    background: rgba(0,0,0,0.02);\n  }\n\n  .detector-scores-table tbody td {\n    padding: 14px 6px;\n    text-align: center;\n    font-size: 0.9rem;\n    font-weight: 500;\n    color: #2d2d3a;\n    border-bottom: 1px solid #eee;\n  }\n\n  .detector-scores-table tbody tr:last-child td {\n    border-bottom: none;\n  }\n\n  .detector-scores-table tbody td:first-child {\n    text-align: left;\n    padding-left: 14px;\n    font-weight: 600;\n    color: #1a1a2e;\n  }\n\n  .score-pill {\n    font-weight: 700;\n    font-family: 'Outfit', sans-serif;\n    font-size: 0.78rem;\n    border-radius: 6px;\n    padding: 5px 8px;\n    display: inline-block;\n    min-width: 0;\n    text-align: center;\n    white-space: nowrap;\n  }\n\n  .score-pass {\n    background: #22c55e;\n    color: #fff;\n  }\n\n  .score-warn {\n    background: #fb923c;\n    color: #fff;\n  }\n\n  .score-fail {\n    background: #ef4444;\n    color: #fff;\n  }\n\n  @media (max-width: 600px) {\n    .detector-scores-table thead th,\n    .detector-scores-table tbody td {\n      padding: 10px 4px;\n      font-size: 0.7rem;\n    }\n    .score-pill {\n      padding: 4px 5px;\n      font-size: 0.68rem;\n    }\n    .detector-scores-wrap h3 {\n      font-size: 1.15rem;\n    }\n  }\n<\/style>\n\n<div class=\"detector-scores-wrap\">\n  <h3>Comment les d\u00e9tecteurs d'images l'ont not\u00e9<\/h3>\n\n  <table class=\"detector-scores-table\">\n    <thead>\n      <tr>\n        <th>Mod\u00e8le d'image<\/th>\n        <th>Cat\u00e9gorie<\/th>\n        <th>TruthScan<\/th>\n        <th>Moteur de vision<\/th>\n        <th>IA ou pas<\/th>\n        <th>Winston AI<\/th>\n        <th>WasItAI<\/th>\n      <\/tr>\n    <\/thead>\n    <tbody>\n      <tr>\n        <td>Voyage \u00e0 mi-parcours <span style=\"font-size:0.75rem; color:#888; display:block;\">G\u00e9n\u00e9ration compl\u00e8te<\/span><\/td>\n        <td>G\u00e9n\u00e9ral<\/td>\n        <td><span class=\"score-pill score-pass\">99.00% AI<\/span><\/td>\n        <td><span class=\"score-pill score-pass\">99.00% AI<\/span><\/td>\n        <td><span class=\"score-pill score-warn\">89.21% AI<\/span><\/td>\n        <td><span class=\"score-pill score-fail\">1.56% AI<\/span><\/td>\n        <td><span class=\"score-pill score-warn\">70.00% AI<\/span><\/td>\n      <\/tr>\n    <\/tbody>\n  <\/table>\n<\/div>\n\n\n\n<p style=\"font-size:17px\">J'ai g\u00e9n\u00e9r\u00e9 l'image ci-dessus en utilisant Midjourney. Plus je la regardais, plus elle me paraissait troublante. L'esth\u00e9tique est sombre et inqui\u00e9tante. Une pr\u00e9sence obscure semble \u00e9merger de derri\u00e8re le sujet. Pour cr\u00e9er cette image, j'ai utilis\u00e9 la fonction de r\u00e9f\u00e9rence esth\u00e9tique de Midjourney et j'ai simplement demand\u00e9 qu'elle soit une \"fille debout dans un club en train de sourire\". J'ai montr\u00e9 l'image \u00e0 quelques personnes, qui l'ont toutes trouv\u00e9e r\u00e9elle. <br><br>La plupart des d\u00e9tecteurs ont signal\u00e9 l'image comme \u00e9tant g\u00e9n\u00e9r\u00e9e par l'IA. TruthScan et Sight Engine l'ont marqu\u00e9e avec un score AI de 99%, AI or Not a indiqu\u00e9 89%. <\/p>\n\n\n\n<p style=\"font-size:17px\">WasItAI n'\u00e9tait pas aussi s\u00fbr que les d\u00e9tecteurs, classant l'image avec seulement un score AI de 70%. <\/p>\n\n\n\n<p style=\"font-size:17px\">Le d\u00e9tecteur de Winston AI a totalement \u00e9chou\u00e9, donnant \u00e0 l'image un score AI de 1,56%. <\/p>\n\n\n\n<h2 class=\"wp-block-heading\">#10 \"Ice Box\" (bo\u00eete \u00e0 glace)<strong> <\/strong>(g\u00e9n\u00e9r\u00e9 par Midjourney)<\/h2>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"771\" src=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/MJ-ICE-1024x771.jpg\" alt=\"Une photo floue montrant un homme avec une chemise tach\u00e9e de rouge, debout devant quelqu&#039;un.\" class=\"wp-image-5311\" title=\"\" srcset=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/MJ-ICE-1024x771.jpg 1024w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/MJ-ICE-300x226.jpg 300w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/MJ-ICE-768x578.jpg 768w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/MJ-ICE-16x12.jpg 16w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/MJ-ICE.jpg 1232w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<style>\n  @import url('https:\/\/fonts.googleapis.com\/css2?family=DM+Sans:wght@400;500;600;700&family=Outfit:wght@600;700;800&display=swap');\n\n  .detector-scores-wrap {\n    max-width: 780px;\n    margin: 2rem auto;\n    font-family: 'DM Sans', sans-serif;\n    -webkit-font-smoothing: antialiased;\n  }\n\n  .detector-scores-wrap h3 {\n    font-family: 'Outfit', sans-serif;\n    font-size: 1.4rem;\n    font-weight: 800;\n    color: #1a1a2e;\n    margin-bottom: 1.25rem;\n    letter-spacing: -0.02em;\n  }\n\n  .detector-scores-table {\n    width: 100%;\n    border-collapse: separate;\n    border-spacing: 0;\n    border-radius: 14px;\n    overflow: hidden;\n    table-layout: fixed;\n    box-shadow:\n      0 1px 3px rgba(0,0,0,0.06),\n      0 8px 24px rgba(0,0,0,0.08),\n      0 0 0 1px rgba(0,0,0,0.04);\n  }\n\n  .detector-scores-table thead th {\n    background: #1a1a2e;\n    color: #ffffff;\n    font-family: 'Outfit', sans-serif;\n    font-weight: 700;\n    font-size: 0.7rem;\n    text-transform: uppercase;\n    letter-spacing: 0.04em;\n    padding: 14px 6px;\n    text-align: center;\n    border: none;\n  }\n\n  .detector-scores-table thead th:first-child {\n    text-align: left;\n    padding-left: 14px;\n    width: 18%;\n  }\n\n  .detector-scores-table thead th:nth-child(2) {\n    width: 14%;\n  }\n\n  .detector-scores-table tbody tr {\n    transition: background 0.2s ease;\n  }\n\n  .detector-scores-table tbody tr:hover {\n    background: rgba(0,0,0,0.02);\n  }\n\n  .detector-scores-table tbody td {\n    padding: 14px 6px;\n    text-align: center;\n    font-size: 0.9rem;\n    font-weight: 500;\n    color: #2d2d3a;\n    border-bottom: 1px solid #eee;\n  }\n\n  .detector-scores-table tbody tr:last-child td {\n    border-bottom: none;\n  }\n\n  .detector-scores-table tbody td:first-child {\n    text-align: left;\n    padding-left: 14px;\n    font-weight: 600;\n    color: #1a1a2e;\n  }\n\n  .score-pill {\n    font-weight: 700;\n    font-family: 'Outfit', sans-serif;\n    font-size: 0.78rem;\n    border-radius: 6px;\n    padding: 5px 8px;\n    display: inline-block;\n    min-width: 0;\n    text-align: center;\n    white-space: nowrap;\n  }\n\n  .score-pass {\n    background: #22c55e;\n    color: #fff;\n  }\n\n  .score-warn {\n    background: #fb923c;\n    color: #fff;\n  }\n\n  .score-fail {\n    background: #ef4444;\n    color: #fff;\n  }\n\n  @media (max-width: 600px) {\n    .detector-scores-table thead th,\n    .detector-scores-table tbody td {\n      padding: 10px 4px;\n      font-size: 0.7rem;\n    }\n    .score-pill {\n      padding: 4px 5px;\n      font-size: 0.68rem;\n    }\n    .detector-scores-wrap h3 {\n      font-size: 1.15rem;\n    }\n  }\n<\/style>\n\n<div class=\"detector-scores-wrap\">\n  <h3>Comment les d\u00e9tecteurs d'images l'ont not\u00e9<\/h3>\n\n  <table class=\"detector-scores-table\">\n    <thead>\n      <tr>\n        <th>Mod\u00e8le d'image<\/th>\n        <th>Cat\u00e9gorie<\/th>\n        <th>TruthScan<\/th>\n        <th>Moteur de vision<\/th>\n        <th>IA ou pas<\/th>\n        <th>Winston AI<\/th>\n        <th>WasItAI<\/th>\n      <\/tr>\n    <\/thead>\n    <tbody>\n      <tr>\n        <td>Voyage \u00e0 mi-parcours <span style=\"font-size:0.75rem; color:#888; display:block;\">G\u00e9n\u00e9ration compl\u00e8te<\/span><\/td>\n        <td style=\"font-size:0.82rem;\">D\u00e9sinformation<\/td>\n        <td><span class=\"score-pill score-pass\">99.00% AI<\/span><\/td>\n        <td><span class=\"score-pill score-pass\">92.00% AI<\/span><\/td>\n        <td><span class=\"score-pill score-pass\">97.94% AI<\/span><\/td>\n        <td><span class=\"score-pill score-fail\">41.20% AI<\/span><\/td>\n        <td><span class=\"score-pill score-warn\">72.00% AI<\/span><\/td>\n      <\/tr>\n    <\/tbody>\n  <\/table>\n<\/div>\n\n\n\n<p style=\"font-size:17px\">J'ai cr\u00e9\u00e9 cette image en utilisant la m\u00eame m\u00e9thode \u00e0 mi-parcours que lors du test pr\u00e9c\u00e9dent. Cette fois, l'objectif \u00e9tait de cr\u00e9er une image un peu plus sombre. TruthScan a \u00e9t\u00e9 le d\u00e9tecteur le plus pr\u00e9cis, \u00e9tiquetant l'image comme 99% AI. Sight Engine et AI or Not ont tous deux correctement d\u00e9tect\u00e9 l'image. Winston AI a compl\u00e8tement \u00e9chou\u00e9, et WasItAI a obtenu un score de confiance bien inf\u00e9rieur \u00e0 celui de TruthScan, Sight Engine et AI or Not.<\/p>\n\n\n\n<p style=\"font-size:17px\">Je dois dire que dans ce cas, l'image que j'ai g\u00e9n\u00e9r\u00e9e pourrait \u00eatre utilis\u00e9e ou encadr\u00e9e de deux mani\u00e8res. D'une part, quelqu'un pourrait utiliser ce type de m\u00e9dia (ou tout autre m\u00e9dia ayant pour th\u00e8me la politique ou les troubles civils) dans le cadre d'un projet artistique ou cr\u00e9atif pour faire une d\u00e9claration. Dans ce cas, le risque semble moindre. La principale inqui\u00e9tude que m'inspire cette cat\u00e9gorie de m\u00e9dias g\u00e9n\u00e9r\u00e9s par l'IA est qu'ils soient utilis\u00e9s par des individus malveillants qui pr\u00e9tendent qu'ils sont r\u00e9els.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">R\u00e9capitulation et r\u00e9sultats : Les meilleurs d\u00e9tecteurs d'images par IA<\/h2>\n\n\n\n<p style=\"font-size:17px\">Bon, maintenant que nous avons affich\u00e9 tous les tests, je vais r\u00e9capituler : J'ai g\u00e9n\u00e9r\u00e9 10 images d'IA<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li style=\"font-size:17px\">J'ai utilis\u00e9 ChatGPT, Nano Banana et Midjourney pour g\u00e9n\u00e9rer 10 images d'IA.<\/li>\n\n\n\n<li style=\"font-size:17px\">J'ai test\u00e9 cinq d\u00e9tecteurs d'images d'IA en y pla\u00e7ant toutes les images d'IA que j'ai g\u00e9n\u00e9r\u00e9es.<\/li>\n\n\n\n<li style=\"font-size:17px\">TruthScan a r\u00e9ussi tous les tests et a \u00e9t\u00e9 le d\u00e9tecteur le plus pr\u00e9cis. AI or Not a r\u00e9ussi 8 tests sur 10, faisant preuve d'une certaine fiabilit\u00e9. Sight Engine a \u00e9chou\u00e9 \u00e0 3 tests sur 10 et a fait preuve d'une pr\u00e9cision douteuse. Was It AI a \u00e9chou\u00e9 \u00e0 4 tests sur 10 et a fait preuve d'une pr\u00e9cision m\u00e9diocre dans l'ensemble. Winston AI \u00e9tait le d\u00e9tecteur d'images le moins pr\u00e9cis, ne r\u00e9ussissant que 3 tests sur 10 et classant syst\u00e9matiquement mal les images.<\/li>\n<\/ul>\n\n\n\n<!DOCTYPE html>\n<html lang=\"en\">\n<head>\n<meta charset=\"UTF-8\">\n<meta name=\"viewport\" content=\"width=device-width, initial-scale=1.0\">\n<title>AI Image Detector - R\u00e9sultats complets des tests<\/title>\n<style>\n  @import url('https:\/\/fonts.googleapis.com\/css2?family=DM+Sans:wght@400;500;600;700&family=Outfit:wght@600;700;800&display=swap');\n\n  *,*::before,*::after{box-sizing:border-box;margin:0;padding:0}\n\n  body {\n    background: #f4f3f0;\n    min-height: 100vh;\n    font-family: 'DM Sans', sans-serif;\n    -webkit-font-smoothing: antialiased;\n    color: #2d2d3a;\n  }\n\n  \/* \u2500\u2500 wrapper \u2500\u2500 *\/\n  .results-page {\n    max-width: 860px;\n    margin: 0 auto;\n    padding: 3rem 1.25rem 4rem;\n  }\n\n  \/* \u2500\u2500 hero header \u2500\u2500 *\/\n  .hero {\n    text-align: center;\n    margin-bottom: 2.75rem;\n  }\n  .hero-badge {\n    display: inline-block;\n    font-family: 'Outfit', sans-serif;\n    font-weight: 700;\n    font-size: 0.65rem;\n    text-transform: uppercase;\n    letter-spacing: 0.12em;\n    color: #fff;\n    background: #1a1a2e;\n    padding: 6px 16px;\n    border-radius: 100px;\n    margin-bottom: 1.1rem;\n  }\n  .hero h1 {\n    font-family: 'Outfit', sans-serif;\n    font-weight: 800;\n    font-size: 2.4rem;\n    color: #1a1a2e;\n    letter-spacing: -0.03em;\n    line-height: 1.15;\n    margin-bottom: 0.6rem;\n  }\n  .hero p {\n    font-size: 1.02rem;\n    color: #6b6b7b;\n    max-width: 560px;\n    margin: 0 auto;\n    line-height: 1.55;\n  }\n\n  \/* \u2500\u2500 accuracy cards row \u2500\u2500 *\/\n  .accuracy-cards {\n    display: grid;\n    grid-template-columns: repeat(5, 1fr);\n    gap: 10px;\n    margin-bottom: 2.75rem;\n  }\n  .acc-card {\n    background: #fff;\n    border-radius: 14px;\n    padding: 1.15rem 0.6rem 1rem;\n    text-align: center;\n    box-shadow:\n      0 1px 3px rgba(0,0,0,0.05),\n      0 6px 20px rgba(0,0,0,0.07),\n      0 0 0 1px rgba(0,0,0,0.03);\n    position: relative;\n    overflow: hidden;\n    transition: transform 0.2s ease, box-shadow 0.2s ease;\n  }\n  .acc-card:hover {\n    transform: translateY(-3px);\n    box-shadow:\n      0 2px 6px rgba(0,0,0,0.06),\n      0 12px 32px rgba(0,0,0,0.1),\n      0 0 0 1px rgba(0,0,0,0.04);\n  }\n  .acc-card::before {\n    content: '';\n    position: absolute;\n    top: 0; left: 0; right: 0;\n    height: 4px;\n  }\n  .acc-card.rank-1::before { background: #16a34a; }\n  .acc-card.rank-2::before { background: #a3e635; }\n  .acc-card.rank-3::before { background: #facc15; }\n  .acc-card.rank-4::before { background: #fb923c; }\n  .acc-card.rank-5::before { background: #ef4444; }\n\n  .acc-card .rank-num {\n    font-family: 'Outfit', sans-serif;\n    font-weight: 800;\n    font-size: 0.62rem;\n    text-transform: uppercase;\n    letter-spacing: 0.08em;\n    color: #999;\n    margin-bottom: 0.35rem;\n  }\n  .acc-card .tool-name {\n    font-family: 'Outfit', sans-serif;\n    font-weight: 700;\n    font-size: 0.92rem;\n    color: #1a1a2e;\n    margin-bottom: 0.65rem;\n    line-height: 1.2;\n  }\n  .acc-ring {\n    width: 72px; height: 72px;\n    margin: 0 auto 0.55rem;\n    position: relative;\n  }\n  .acc-ring svg { transform: rotate(-90deg); }\n  .acc-ring .ring-bg { fill: none; stroke: #eee; stroke-width: 5; }\n  .acc-ring .ring-fg { fill: none; stroke-width: 5; stroke-linecap: round; transition: stroke-dashoffset 1s ease; }\n  .acc-ring .ring-label {\n    position: absolute;\n    inset: 0;\n    display: flex;\n    align-items: center;\n    justify-content: center;\n    font-family: 'Outfit', sans-serif;\n    font-weight: 800;\n    font-size: 1.1rem;\n    color: #1a1a2e;\n  }\n  .acc-card .acc-fraction {\n    font-size: 0.78rem;\n    color: #888;\n    font-weight: 500;\n  }\n\n  \/* \u2500\u2500 section titles \u2500\u2500 *\/\n  .section-title {\n    font-family: 'Outfit', sans-serif;\n    font-weight: 800;\n    font-size: 1.45rem;\n    color: #1a1a2e;\n    letter-spacing: -0.02em;\n    margin-bottom: 1rem;\n  }\n  .section-subtitle {\n    font-size: 0.88rem;\n    color: #888;\n    margin-bottom: 1.25rem;\n    line-height: 1.5;\n  }\n\n  \/* \u2500\u2500 master table \u2500\u2500 *\/\n  .master-table-wrap {\n    border-radius: 14px;\n    overflow: hidden;\n    box-shadow:\n      0 1px 3px rgba(0,0,0,0.06),\n      0 8px 24px rgba(0,0,0,0.08),\n      0 0 0 1px rgba(0,0,0,0.04);\n    margin-bottom: 2.75rem;\n    background: #fff;\n  }\n  .master-table {\n    width: 100%;\n    border-collapse: collapse;\n    table-layout: fixed;\n  }\n  .master-table thead th {\n    background: #1a1a2e;\n    color: #fff;\n    font-family: 'Outfit', sans-serif;\n    font-weight: 700;\n    font-size: 0.62rem;\n    text-transform: uppercase;\n    letter-spacing: 0.05em;\n    padding: 13px 5px;\n    text-align: center;\n    border: none;\n    position: sticky;\n    top: 0;\n    z-index: 2;\n  }\n  .master-table thead th:first-child {\n    text-align: left;\n    padding-left: 14px;\n    width: 14%;\n  }\n  .master-table thead th:nth-child(2) { width: 13%; }\n  .master-table thead th:nth-child(3) { width: 9%; }\n\n  .master-table tbody tr {\n    transition: background 0.18s ease;\n  }\n  .master-table tbody tr:hover {\n    background: rgba(26,26,46,0.025);\n  }\n  .master-table tbody td {\n    padding: 11px 5px;\n    text-align: center;\n    font-size: 0.82rem;\n    font-weight: 500;\n    color: #2d2d3a;\n    border-bottom: 1px solid #f0f0ec;\n  }\n  .master-table tbody tr:last-child td { border-bottom: none; }\n  .master-table tbody td:first-child {\n    text-align: left;\n    padding-left: 14px;\n    font-weight: 600;\n    color: #1a1a2e;\n    font-size: 0.78rem;\n    line-height: 1.3;\n  }\n  .test-num {\n    display: inline-block;\n    font-family: 'Outfit', sans-serif;\n    font-weight: 800;\n    font-size: 0.6rem;\n    background: #1a1a2e;\n    color: #fff;\n    width: 18px; height: 18px;\n    line-height: 18px;\n    text-align: center;\n    border-radius: 5px;\n    margin-right: 5px;\n    vertical-align: middle;\n  }\n  .img-model-sub {\n    display: block;\n    font-size: 0.68rem;\n    color: #999;\n    font-weight: 400;\n    margin-top: 1px;\n  }\n  .cat-badge {\n    font-family: 'Outfit', sans-serif;\n    font-size: 0.6rem;\n    font-weight: 700;\n    text-transform: uppercase;\n    letter-spacing: 0.03em;\n    padding: 3px 6px;\n    border-radius: 5px;\n    display: inline-block;\n    white-space: normal;\n    line-height: 1.35;\n    max-width: 100%;\n  }\n  .cat-fraud    { background: #fef2f2; color: #dc2626; }\n  .cat-disinfo  { background: #fefce8; color: #a16207; }\n  .cat-general  { background: #f0fdf4; color: #16a34a; }\n  .cat-deepfake { background: #faf5ff; color: #7c3aed; }\n  .cat-mixed    { background: #fff7ed; color: #c2410c; }\n\n  \/* score pills *\/\n  .pill {\n    font-weight: 700;\n    font-family: 'Outfit', sans-serif;\n    font-size: 0.7rem;\n    border-radius: 6px;\n    padding: 4px 7px;\n    display: inline-block;\n    text-align: center;\n    white-space: nowrap;\n    min-width: 68px;\n  }\n  .pill-pass { background: #22c55e; color: #fff; }\n  .pill-warn { background: #fb923c; color: #fff; }\n  .pill-fail { background: #ef4444; color: #fff; }\n\n  \/* \u2500\u2500 legend \u2500\u2500 *\/\n  .legend {\n    display: flex;\n    gap: 1.25rem;\n    align-items: center;\n    margin-bottom: 1.25rem;\n    flex-wrap: wrap;\n  }\n  .legend-item {\n    display: flex;\n    align-items: center;\n    gap: 6px;\n    font-size: 0.78rem;\n    font-weight: 500;\n    color: #666;\n  }\n  .legend-dot {\n    width: 12px; height: 12px;\n    border-radius: 4px;\n  }\n  .legend-dot.lg-pass { background: #22c55e; }\n  .legend-dot.lg-warn { background: #fb923c; }\n  .legend-dot.lg-fail { background: #ef4444; }\n\n  \/* \u2500\u2500 bottom summary strip \u2500\u2500 *\/\n  .summary-strip {\n    display: grid;\n    grid-template-columns: repeat(3, 1fr);\n    gap: 10px;\n    margin-bottom: 2.5rem;\n  }\n  .strip-card {\n    background: #1a1a2e;\n    border-radius: 12px;\n    padding: 1.1rem 1rem;\n    color: #fff;\n    text-align: center;\n  }\n  .strip-card .strip-val {\n    font-family: 'Outfit', sans-serif;\n    font-weight: 800;\n    font-size: 1.65rem;\n    letter-spacing: -0.02em;\n    margin-bottom: 0.15rem;\n  }\n  .strip-card .strip-label {\n    font-size: 0.72rem;\n    color: rgba(255,255,255,0.55);\n    font-weight: 500;\n    text-transform: uppercase;\n    letter-spacing: 0.05em;\n  }\n  .strip-val .hl-green { color: #4ade80; }\n  .strip-val .hl-red   { color: #f87171; }\n  .strip-val .hl-amber { color: #fbbf24; }\n\n  \/* \u2500\u2500 footer note \u2500\u2500 *\/\n  .foot-note {\n    font-size: 0.78rem;\n    color: #999;\n    line-height: 1.6;\n    text-align: center;\n    max-width: 640px;\n    margin: 0 auto;\n  }\n  .foot-note strong { color: #666; font-weight: 600; }\n\n  \/* \u2500\u2500 responsive \u2500\u2500 *\/\n  @media (max-width: 700px) {\n    .hero h1 { font-size: 1.6rem; }\n    .accuracy-cards { grid-template-columns: repeat(2, 1fr); }\n    .accuracy-cards .acc-card:last-child { grid-column: span 2; justify-self: center; max-width: 200px; }\n    .summary-strip { grid-template-columns: 1fr; }\n    .master-table thead th,\n    .master-table tbody td { padding: 9px 3px; font-size: 0.68rem; }\n    .pill { font-size: 0.6rem; padding: 3px 4px; min-width: 54px; }\n    .test-num { width: 15px; height: 15px; line-height: 15px; font-size: 0.52rem; }\n    .section-title { font-size: 1.15rem; }\n  }\n\n  \/* \u2500\u2500 entrance animations \u2500\u2500 *\/\n  @keyframes fadeUp {\n    from { opacity: 0; transform: translateY(18px); }\n    to   { opacity: 1; transform: translateY(0); }\n  }\n  .anim { animation: fadeUp 0.6s ease both; }\n  .anim-d1 { animation-delay: 0.08s; }\n  .anim-d2 { animation-delay: 0.16s; }\n  .anim-d3 { animation-delay: 0.24s; }\n  .anim-d4 { animation-delay: 0.32s; }\n  .anim-d5 { animation-delay: 0.4s; }\n<\/style>\n<\/head>\n<body>\n\n<div class=\"results-page\">\n\n  <!-- \u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550 HERO \u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550 -->\n  <header class=\"hero anim\">\n    <span class=\"hero-badge\">50 contr\u00f4les de d\u00e9tection - 5 d\u00e9tecteurs - 10 images AI<\/span>\n    <h1>R\u00e9sultats complets des tests<\/h1>\n    <p>Les performances de cinq d\u00e9tecteurs d'images IA populaires dans les domaines de la fraude, de la d\u00e9sinformation, des deepfakes et de la photographie en g\u00e9n\u00e9ral.<\/p>\n  <\/header>\n\n  <!-- \u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550 ACCURACY RING CARDS \u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550 -->\n  <div class=\"accuracy-cards anim anim-d1\">\n\n    <!-- 1 \u2014 TruthScan 100% -->\n    <div class=\"acc-card rank-1\">\n      <div class=\"rank-num\">#1<\/div>\n      <div class=\"tool-name\">TruthScan<\/div>\n      <div class=\"acc-ring\">\n        <svg viewbox=\"0 0 80 80\" width=\"72\" height=\"72\">\n          <circle class=\"ring-bg\" cx=\"40\" cy=\"40\" r=\"34\"\/>\n          <circle class=\"ring-fg\" cx=\"40\" cy=\"40\" r=\"34\"\n                  stroke=\"#16a34a\"\n                  stroke-dasharray=\"213.6\"\n                  stroke-dashoffset=\"0\"\/>\n        <\/svg>\n        <span class=\"ring-label\">100%<\/span>\n      <\/div>\n      <div class=\"acc-fraction\">10 \/ 10 d\u00e9tect\u00e9<\/div>\n    <\/div>\n\n    <!-- 2 \u2014 AI or Not 80% -->\n    <div class=\"acc-card rank-2\">\n      <div class=\"rank-num\">#2<\/div>\n      <div class=\"tool-name\">IA ou pas<\/div>\n      <div class=\"acc-ring\">\n        <svg viewbox=\"0 0 80 80\" width=\"72\" height=\"72\">\n          <circle class=\"ring-bg\" cx=\"40\" cy=\"40\" r=\"34\"\/>\n          <circle class=\"ring-fg\" cx=\"40\" cy=\"40\" r=\"34\"\n                  stroke=\"#a3e635\"\n                  stroke-dasharray=\"213.6\"\n                  stroke-dashoffset=\"42.72\"\/>\n        <\/svg>\n        <span class=\"ring-label\">80%<\/span>\n      <\/div>\n      <div class=\"acc-fraction\">8 \/ 10 d\u00e9tect\u00e9<\/div>\n    <\/div>\n\n    <!-- 3 \u2014 Sight Engine 70% -->\n    <div class=\"acc-card rank-3\">\n      <div class=\"rank-num\">#3<\/div>\n      <div class=\"tool-name\">Moteur de vision<\/div>\n      <div class=\"acc-ring\">\n        <svg viewbox=\"0 0 80 80\" width=\"72\" height=\"72\">\n          <circle class=\"ring-bg\" cx=\"40\" cy=\"40\" r=\"34\"\/>\n          <circle class=\"ring-fg\" cx=\"40\" cy=\"40\" r=\"34\"\n                  stroke=\"#facc15\"\n                  stroke-dasharray=\"213.6\"\n                  stroke-dashoffset=\"64.08\"\/>\n        <\/svg>\n        <span class=\"ring-label\">70%<\/span>\n      <\/div>\n      <div class=\"acc-fraction\">7 \/ 10 d\u00e9tect\u00e9<\/div>\n    <\/div>\n\n    <!-- 4 \u2014 WasItAI 60% -->\n    <div class=\"acc-card rank-4\">\n      <div class=\"rank-num\">#4<\/div>\n      <div class=\"tool-name\">WasItAI<\/div>\n      <div class=\"acc-ring\">\n        <svg viewbox=\"0 0 80 80\" width=\"72\" height=\"72\">\n          <circle class=\"ring-bg\" cx=\"40\" cy=\"40\" r=\"34\"\/>\n          <circle class=\"ring-fg\" cx=\"40\" cy=\"40\" r=\"34\"\n                  stroke=\"#fb923c\"\n                  stroke-dasharray=\"213.6\"\n                  stroke-dashoffset=\"85.44\"\/>\n        <\/svg>\n        <span class=\"ring-label\">60%<\/span>\n      <\/div>\n      <div class=\"acc-fraction\">6 \/ 10 d\u00e9tect\u00e9<\/div>\n    <\/div>\n\n    <!-- 5 \u2014 Winston AI 30% -->\n    <div class=\"acc-card rank-5\">\n      <div class=\"rank-num\">#5<\/div>\n      <div class=\"tool-name\">Winston AI<\/div>\n      <div class=\"acc-ring\">\n        <svg viewbox=\"0 0 80 80\" width=\"72\" height=\"72\">\n          <circle class=\"ring-bg\" cx=\"40\" cy=\"40\" r=\"34\"\/>\n          <circle class=\"ring-fg\" cx=\"40\" cy=\"40\" r=\"34\"\n                  stroke=\"#ef4444\"\n                  stroke-dasharray=\"213.6\"\n                  stroke-dashoffset=\"149.52\"\/>\n        <\/svg>\n        <span class=\"ring-label\">30%<\/span>\n      <\/div>\n      <div class=\"acc-fraction\">3 \/ 10 d\u00e9tect\u00e9<\/div>\n    <\/div>\n\n  <\/div>\n\n  <!-- \u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550 STATS STRIP \u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550 -->\n  <div class=\"summary-strip anim anim-d2\">\n    <div class=\"strip-card\">\n      <div class=\"strip-val\"><span class=\"hl-green\">34<\/span> \/ 50<\/div>\n      <div class=\"strip-label\">Total des d\u00e9tections correctes<\/div>\n    <\/div>\n    <div class=\"strip-card\">\n      <div class=\"strip-val\"><span class=\"hl-red\">16<\/span> \/ 50<\/div>\n      <div class=\"strip-label\">Total des d\u00e9tections manqu\u00e9es<\/div>\n    <\/div>\n    <div class=\"strip-card\">\n      <div class=\"strip-val\"><span class=\"hl-amber\">68%<\/span><\/div>\n      <div class=\"strip-label\">Pr\u00e9cision moyenne sur l'ensemble des<\/div>\n    <\/div>\n  <\/div>\n\n  <!-- \u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550 LEGEND \u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550 -->\n  <h3 class=\"section-title anim anim-d3\">Les 10 tests - D\u00e9composition compl\u00e8te<\/h3>\n  <p class=\"section-subtitle anim anim-d3\">Chaque score repr\u00e9sente la cote de confiance de l'IA du d\u00e9tecteur. Seuil de r\u00e9ussite : \u2265 90% AI.<\/p>\n\n  <div class=\"legend anim anim-d3\">\n    <div class=\"legend-item\"><span class=\"legend-dot lg-pass\"><\/span> Pass (\u2265 90%)<\/div>\n    <div class=\"legend-item\"><span class=\"legend-dot lg-warn\"><\/span> Presque rat\u00e9 (70-89%)<\/div>\n    <div class=\"legend-item\"><span class=\"legend-dot lg-fail\"><\/span> \u00c9chec (&lt; 70%)<\/div>\n  <\/div>\n\n  <!-- \u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550 MASTER TABLE \u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550 -->\n  <div class=\"master-table-wrap anim anim-d4\">\n    <table class=\"master-table\">\n      <thead>\n        <tr>\n          <th>Test<\/th>\n          <th>Cat\u00e9gorie<\/th>\n          <th>Source<\/th>\n          <th>TruthScan<\/th>\n          <th>Moteur de vision<\/th>\n          <th>IA ou pas<\/th>\n          <th>Winston AI<\/th>\n          <th>WasItAI<\/th>\n        <\/tr>\n      <\/thead>\n      <tbody>\n\n        <!-- #1 -->\n        <tr>\n          <td><span class=\"test-num\">1<\/span> L'homme sur la corde raide<\/td>\n          <td><span class=\"cat-badge cat-general\">G\u00e9n\u00e9ral<\/span><\/td>\n          <td><span class=\"img-model-sub\">ChatGPT<\/span><\/td>\n          <td><span class=\"pill pill-pass\">99.00%<\/span><\/td>\n          <td><span class=\"pill pill-pass\">99.00%<\/span><\/td>\n          <td><span class=\"pill pill-warn\">78.00%<\/span><\/td>\n          <td><span class=\"pill pill-fail\">0.98%<\/span><\/td>\n          <td><span class=\"pill pill-fail\">1.00%<\/span><\/td>\n        <\/tr>\n\n        <!-- #2 -->\n        <tr>\n          <td><span class=\"test-num\">2<\/span> Le faux re\u00e7u<\/td>\n          <td><span class=\"cat-badge cat-fraud\">Fraude<\/span><\/td>\n          <td><span class=\"img-model-sub\">ChatGPT<\/span><\/td>\n          <td><span class=\"pill pill-pass\">99.00%<\/span><\/td>\n          <td><span class=\"pill pill-fail\">19.00%<\/span><\/td>\n          <td><span class=\"pill pill-pass\">94.48%<\/span><\/td>\n          <td><span class=\"pill pill-fail\">0.04%<\/span><\/td>\n          <td><span class=\"pill pill-fail\">1.00%<\/span><\/td>\n        <\/tr>\n\n        <!-- #3 -->\n        <tr>\n          <td><span class=\"test-num\">3<\/span> Paquet de mesures de pr\u00e9caution<\/td>\n          <td><span class=\"cat-badge cat-fraud\">Fraude<\/span><\/td>\n          <td><span class=\"img-model-sub\">Nano Banane<\/span><\/td>\n          <td><span class=\"pill pill-pass\">99.00%<\/span><\/td>\n          <td><span class=\"pill pill-pass\">98.00%<\/span><\/td>\n          <td><span class=\"pill pill-pass\">99.10%<\/span><\/td>\n          <td><span class=\"pill pill-fail\">31.98%<\/span><\/td>\n          <td><span class=\"pill pill-pass\">99.00%<\/span><\/td>\n        <\/tr>\n\n        <!-- #4 -->\n        <tr>\n          <td><span class=\"test-num\">4<\/span> Repas pour cafards<\/td>\n          <td><span class=\"cat-badge cat-fraud\">Fraude<\/span><\/td>\n          <td><span class=\"img-model-sub\">Nano Banane<\/span><\/td>\n          <td><span class=\"pill pill-pass\">97.15%<\/span><\/td>\n          <td><span class=\"pill pill-fail\">18.00%<\/span><\/td>\n          <td><span class=\"pill pill-warn\">85.87%<\/span><\/td>\n          <td><span class=\"pill pill-warn\">72.29%<\/span><\/td>\n          <td><span class=\"pill pill-fail\">1.00%<\/span><\/td>\n        <\/tr>\n\n        <!-- #5 -->\n        <tr>\n          <td><span class=\"test-num\">5<\/span> Autovandalisme<\/td>\n          <td><span class=\"cat-badge cat-mixed\">Fraude \/ D\u00e9sinformation<\/span><\/td>\n          <td><span class=\"img-model-sub\">Nano Banane<\/span><\/td>\n          <td><span class=\"pill pill-pass\">97.48%<\/span><\/td>\n          <td><span class=\"pill pill-fail\">18.00%<\/span><\/td>\n          <td><span class=\"pill pill-warn\">89.52%<\/span><\/td>\n          <td><span class=\"pill pill-fail\">0.02%<\/span><\/td>\n          <td><span class=\"pill pill-pass\">99.00%<\/span><\/td>\n        <\/tr>\n\n        <!-- #6 -->\n        <tr>\n          <td><span class=\"test-num\">6<\/span> Soda Musk &amp; Don<\/td>\n          <td><span class=\"cat-badge cat-disinfo\">D\u00e9sinformation<\/span><\/td>\n          <td><span class=\"img-model-sub\">Nano Banane<\/span><\/td>\n          <td><span class=\"pill pill-pass\">99.00%<\/span><\/td>\n          <td><span class=\"pill pill-pass\">99.00%<\/span><\/td>\n          <td><span class=\"pill pill-pass\">99.00%<\/span><\/td>\n          <td><span class=\"pill pill-fail\">0.24%<\/span><\/td>\n          <td><span class=\"pill pill-warn\">86.00%<\/span><\/td>\n        <\/tr>\n\n        <!-- #7 -->\n        <tr>\n          <td><span class=\"test-num\">7<\/span> Toujours en vie<\/td>\n          <td><span class=\"cat-badge cat-deepfake\">Faux-semblant<\/span><\/td>\n          <td><span class=\"img-model-sub\">Nano Banane<\/span><\/td>\n          <td><span class=\"pill pill-pass\">97.49%<\/span><\/td>\n          <td><span class=\"pill pill-pass\">95.00%<\/span><\/td>\n          <td><span class=\"pill pill-pass\">99.27%<\/span><\/td>\n          <td><span class=\"pill pill-fail\">0.15%<\/span><\/td>\n          <td><span class=\"pill pill-fail\">1.00%<\/span><\/td>\n        <\/tr>\n\n        <!-- #8 -->\n        <tr>\n          <td><span class=\"test-num\">8<\/span> Abus d'un Ben<\/td>\n          <td><span class=\"cat-badge cat-mixed\">Deepfake \/ Disinfo<\/span><\/td>\n          <td><span class=\"img-model-sub\">Nano Banane<\/span><\/td>\n          <td><span class=\"pill pill-pass\">99.00%<\/span><\/td>\n          <td><span class=\"pill pill-pass\">97.00%<\/span><\/td>\n          <td><span class=\"pill pill-pass\">98.88%<\/span><\/td>\n          <td><span class=\"pill pill-fail\">5.74%<\/span><\/td>\n          <td><span class=\"pill pill-pass\">99.00%<\/span><\/td>\n        <\/tr>\n\n        <!-- #9 -->\n        <tr>\n          <td><span class=\"test-num\">9<\/span> Club Void<\/td>\n          <td><span class=\"cat-badge cat-general\">G\u00e9n\u00e9ral<\/span><\/td>\n          <td><span class=\"img-model-sub\">Voyage \u00e0 mi-parcours<\/span><\/td>\n          <td><span class=\"pill pill-pass\">99.00%<\/span><\/td>\n          <td><span class=\"pill pill-pass\">99.00%<\/span><\/td>\n          <td><span class=\"pill pill-warn\">89.21%<\/span><\/td>\n          <td><span class=\"pill pill-fail\">1.56%<\/span><\/td>\n          <td><span class=\"pill pill-warn\">70.00%<\/span><\/td>\n        <\/tr>\n\n        <!-- #10 -->\n        <tr>\n          <td><span class=\"test-num\">10<\/span> La glaci\u00e8re<\/td>\n          <td><span class=\"cat-badge cat-disinfo\">D\u00e9sinformation<\/span><\/td>\n          <td><span class=\"img-model-sub\">Voyage \u00e0 mi-parcours<\/span><\/td>\n          <td><span class=\"pill pill-pass\">99.00%<\/span><\/td>\n          <td><span class=\"pill pill-pass\">92.00%<\/span><\/td>\n          <td><span class=\"pill pill-pass\">97.94%<\/span><\/td>\n          <td><span class=\"pill pill-fail\">41.20%<\/span><\/td>\n          <td><span class=\"pill pill-warn\">72.00%<\/span><\/td>\n        <\/tr>\n\n      <\/tbody>\n    <\/table>\n  <\/div>\n\n  <!-- \u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550 FOOTNOTES \u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550 -->\n  <p class=\"foot-note anim anim-d5\">\n    <strong>M\u00e9thodologie :<\/strong> Chaque image a \u00e9t\u00e9 soumise une fois \u00e0 chaque d\u00e9tecteur. Un score de <strong>\u2265 90% AI<\/strong> est consid\u00e9r\u00e9e comme une d\u00e9tection correcte. Les scores compris entre 70 et 89% sont des \"quasi-erreurs\". Tout r\u00e9sultat inf\u00e9rieur \u00e0 70% est un \u00e9chec. AI or Not a \u00e9tiquet\u00e9 une g\u00e9n\u00e9ration \u00e0 78%, une \u00e0 85% et une \u00e0 89% - la 89% est trait\u00e9e comme un quasi-\u00e9chec avec un ast\u00e9risque dans le tableau r\u00e9capitulatif.\n  <\/p>\n\n<\/div>\n\n<\/body>\n<\/html>\n\n\n\n<h2 class=\"wp-block-heading\">Remarques et donn\u00e9es de cl\u00f4ture<\/h2>\n\n\n\n<p style=\"font-size:17px\">Au total, entre la r\u00e9daction de cet article et les tests rigoureux, j'ai consacr\u00e9 25 heures \u00e0 l'\u00e9laboration de ce rapport. Les outils de d\u00e9tection multimodale de l'IA sont encore en cours de d\u00e9veloppement, mais il est clair que certains sont plus pr\u00e9cis que d'autres. Apr\u00e8s avoir test\u00e9 tous les outils, TruthScan est le plus pr\u00e9cis. <a href=\"https:\/\/truthscan.com\/ai-image-detector\" target=\"_blank\" rel=\"noopener\">D\u00e9tecteur d'images AI<\/a>. Les tests parlent d'eux-m\u00eames. <\/p>\n\n\n\n<p style=\"font-size:17px\">Si vous souhaitez acc\u00e9der \u00e0 une copie CSV des donn\u00e9es issues des tests que j'ai effectu\u00e9s dans cet article, vous pouvez le faire. <a href=\"https:\/\/docs.google.com\/spreadsheets\/d\/1iuapCicNcdG6dZCFd-dCmSRgUJ9GVCS2v9kh4PShRew\/edit?usp=sharing\" data-type=\"link\" data-id=\"https:\/\/docs.google.com\/spreadsheets\/d\/1iuapCicNcdG6dZCFd-dCmSRgUJ9GVCS2v9kh4PShRew\/edit?usp=sharing\" target=\"_blank\" rel=\"noopener\">ici.<\/a> La feuille de calcul que j'ai mise en lien contient toutes les invites originales et les r\u00e9sultats de d\u00e9tection des tests pr\u00e9sent\u00e9s dans cet article. <\/p>","protected":false},"excerpt":{"rendered":"<p>Vous avez probablement vu plus de photos g\u00e9n\u00e9r\u00e9es par l'IA en ligne que vous ne le pensez. Parfois, il est \u00e9vident qu'une [...]<\/p>","protected":false},"author":15,"featured_media":5326,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"rank_math_lock_modified_date":false,"_themeisle_gutenberg_block_has_review":false,"footnotes":""},"categories":[34],"tags":[19,96,97],"class_list":["post-5287","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-news","tag-ai-detection","tag-ai-images","tag-detecting-deepfakes"],"_links":{"self":[{"href":"https:\/\/research.undetectable.ai\/fr\/wp-json\/wp\/v2\/posts\/5287","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/research.undetectable.ai\/fr\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/research.undetectable.ai\/fr\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/research.undetectable.ai\/fr\/wp-json\/wp\/v2\/users\/15"}],"replies":[{"embeddable":true,"href":"https:\/\/research.undetectable.ai\/fr\/wp-json\/wp\/v2\/comments?post=5287"}],"version-history":[{"count":27,"href":"https:\/\/research.undetectable.ai\/fr\/wp-json\/wp\/v2\/posts\/5287\/revisions"}],"predecessor-version":[{"id":5334,"href":"https:\/\/research.undetectable.ai\/fr\/wp-json\/wp\/v2\/posts\/5287\/revisions\/5334"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/research.undetectable.ai\/fr\/wp-json\/wp\/v2\/media\/5326"}],"wp:attachment":[{"href":"https:\/\/research.undetectable.ai\/fr\/wp-json\/wp\/v2\/media?parent=5287"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/research.undetectable.ai\/fr\/wp-json\/wp\/v2\/categories?post=5287"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/research.undetectable.ai\/fr\/wp-json\/wp\/v2\/tags?post=5287"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}