{"id":4365,"date":"2026-01-11T12:37:29","date_gmt":"2026-01-11T11:37:29","guid":{"rendered":"https:\/\/laurentchani.fr\/?p=4365"},"modified":"2026-01-11T12:44:04","modified_gmt":"2026-01-11T11:44:04","slug":"when-the-details-shimmer","status":"publish","type":"post","link":"https:\/\/laurentchani.fr\/en\/quand-lia-miroite-des-details\/","title":{"rendered":"When AI reflects details"},"content":{"rendered":"<p>When AI \u201cmirrors\u201d a detail\u2026 and breaks a character&#039;s consistency<br>I conducted a simple, but very revealing test on the consistency of characters in image generation.<br>Case: an asymmetrical character (e.g., a bag carried on the character&#039;s left shoulder). We start with a reference image (left profile), then we ask for the opposite profile\u2026 and the bag \u201cmoves\u201d to the other shoulder, as if the model were mirroring rather than truly rotating.<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-medium\"><img loading=\"lazy\" decoding=\"async\" width=\"300\" height=\"196\" src=\"https:\/\/laurentchani.fr\/wp-content\/uploads\/@Big-Daddy-100-P-300x196.jpg\" alt=\"\" class=\"wp-image-4366\" srcset=\"https:\/\/laurentchani.fr\/wp-content\/uploads\/@Big-Daddy-100-P-300x196.jpg 300w, https:\/\/laurentchani.fr\/wp-content\/uploads\/@Big-Daddy-100-P-768x502.jpg 768w, https:\/\/laurentchani.fr\/wp-content\/uploads\/@Big-Daddy-100-P-18x12.jpg 18w, https:\/\/laurentchani.fr\/wp-content\/uploads\/@Big-Daddy-100-P.jpg 784w\" sizes=\"auto, (max-width: 300px) 100vw, 300px\" \/><\/figure>\n\n\n\n<p><br>\u27a1\ufe0f This is not a \u201cprompt wave\u201d problem. I have tried multiple methods (constraints \u201cleft of character, not left of camera\u201d, \u201cno mirror\u201d, \u201cstrictly the same person\u201d, etc.) on several AI image generators\u2026 without obtaining robust consistency in this specific case.<br>Thanks to Baptiste Rognon (The AI guy) <a href=\"https:\/\/www.linkedin.com\/in\/baptisterognon\/\"><strong>https:\/\/www.linkedin.com\/in\/baptisterognon\/<\/strong><\/a> ) who reproduced tests on his side and confirms: we are reaching a current limit of the models (for now).<\/p>\n\n\n\n<p><br>My takeaway from this:<br>1. Some lateralized (left\/right) details remain unstable as soon as the angle is changed.<br>2. The model may \u201cprefer\u201d an implicit symmetry rather than an anatomical continuity.<br>3. The right reflex is not to \u201ccry bug\u201d, but to design a workflow (editing\/inpainting, double referencing, targeted retouching) when consistency is critical.<br>I&#039;ve added an image below that clearly illustrates the mirror effect.<br>If you have found a reliable workaround (locked seed, local editing, multi-reference method\u2026), I&#039;m all ears: I&#039;ll compile the best leads and give feedback.<\/p>\n\n\n\n<p><br><a href=\"https:\/\/www.linkedin.com\/feed\/hashtag\/?keywords=ia\"><strong>##IA<\/strong><\/a> <a href=\"https:\/\/www.linkedin.com\/feed\/hashtag\/?keywords=generativeai\"><strong>##GenerativeAI<\/strong><\/a> <a href=\"https:\/\/www.linkedin.com\/feed\/hashtag\/?keywords=promptengineering\"><strong>##PromptEngineering<\/strong><\/a> <a href=\"https:\/\/www.linkedin.com\/feed\/hashtag\/?keywords=creationvisuelle\"><strong>##CreationVisuelle<\/strong><\/a> <a href=\"https:\/\/www.linkedin.com\/feed\/hashtag\/?keywords=imagegeneration\"><strong>hashtag#ImageGeneration<\/strong><\/a> <a href=\"https:\/\/www.linkedin.com\/feed\/hashtag\/?keywords=workflow\"><strong>hashtag#Workflow<\/strong><\/a><\/p>","protected":false},"excerpt":{"rendered":"<p>Quand l\u2019IA \u201cmiroir\u201d un d\u00e9tail\u2026 et casse la coh\u00e9rence d\u2019un personnageJ\u2019ai men\u00e9 un test simple, mais tr\u00e8s r\u00e9v\u00e9lateur, sur la consistance des personnages en g\u00e9n\u00e9ration d\u2019images.Cas : un personnage non sym\u00e9trique (ex. sac port\u00e9 \u00e0 l\u2019\u00e9paule gauche du personnage). On part d\u2019une image de r\u00e9f\u00e9rence (profil gauche), puis on demande le profil oppos\u00e9\u2026 et le [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":4366,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_uag_custom_page_level_css":"","footnotes":""},"categories":[43,10],"tags":[],"class_list":["post-4365","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-autoedition","category-ia"],"blocksy_meta":[],"uagb_featured_image_src":{"full":["https:\/\/laurentchani.fr\/wp-content\/uploads\/@Big-Daddy-100-P.jpg",784,512,false],"thumbnail":["https:\/\/laurentchani.fr\/wp-content\/uploads\/@Big-Daddy-100-P-150x150.jpg",150,150,true],"medium":["https:\/\/laurentchani.fr\/wp-content\/uploads\/@Big-Daddy-100-P-300x196.jpg",300,196,true],"medium_large":["https:\/\/laurentchani.fr\/wp-content\/uploads\/@Big-Daddy-100-P-768x502.jpg",768,502,true],"large":["https:\/\/laurentchani.fr\/wp-content\/uploads\/@Big-Daddy-100-P.jpg",784,512,false],"1536x1536":["https:\/\/laurentchani.fr\/wp-content\/uploads\/@Big-Daddy-100-P.jpg",784,512,false],"2048x2048":["https:\/\/laurentchani.fr\/wp-content\/uploads\/@Big-Daddy-100-P.jpg",784,512,false],"trp-custom-language-flag":["https:\/\/laurentchani.fr\/wp-content\/uploads\/@Big-Daddy-100-P-18x12.jpg",18,12,true],"mailpoet_newsletter_max":["https:\/\/laurentchani.fr\/wp-content\/uploads\/@Big-Daddy-100-P.jpg",784,512,false]},"uagb_author_info":{"display_name":"admin5261","author_link":"https:\/\/laurentchani.fr\/en\/author\/admin5261\/"},"uagb_comment_info":0,"uagb_excerpt":"Quand l\u2019IA \u201cmiroir\u201d un d\u00e9tail\u2026 et casse la coh\u00e9rence d\u2019un personnageJ\u2019ai men\u00e9 un test simple, mais tr\u00e8s r\u00e9v\u00e9lateur, sur la consistance des personnages en g\u00e9n\u00e9ration d\u2019images.Cas : un personnage non sym\u00e9trique (ex. sac port\u00e9 \u00e0 l\u2019\u00e9paule gauche du personnage). On part d\u2019une image de r\u00e9f\u00e9rence (profil gauche), puis on demande le profil oppos\u00e9\u2026 et le&hellip;","_links":{"self":[{"href":"https:\/\/laurentchani.fr\/en\/wp-json\/wp\/v2\/posts\/4365","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/laurentchani.fr\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/laurentchani.fr\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/laurentchani.fr\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/laurentchani.fr\/en\/wp-json\/wp\/v2\/comments?post=4365"}],"version-history":[{"count":1,"href":"https:\/\/laurentchani.fr\/en\/wp-json\/wp\/v2\/posts\/4365\/revisions"}],"predecessor-version":[{"id":4367,"href":"https:\/\/laurentchani.fr\/en\/wp-json\/wp\/v2\/posts\/4365\/revisions\/4367"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/laurentchani.fr\/en\/wp-json\/wp\/v2\/media\/4366"}],"wp:attachment":[{"href":"https:\/\/laurentchani.fr\/en\/wp-json\/wp\/v2\/media?parent=4365"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/laurentchani.fr\/en\/wp-json\/wp\/v2\/categories?post=4365"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/laurentchani.fr\/en\/wp-json\/wp\/v2\/tags?post=4365"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}