7 months, 240 generations: measuring how far an AI character's face actually drifted
No_Issue_8224 · reddit · 2026-09-03
A creator running a weekly AI picture story measured face drift across 240 generations of the same fictional character (Jan–Jul). Using 38 'anchor' images and 20 fresh generations from an untouched January prompt — all normalized to 500px pupil distance and compared pixel-wise in GIMP — they found: mouth width grew 11% (236→261px), hairline rose 29px, jaw narrowed 26px, while nose width stayed within their self-measured 8px noise floor. Most telling: the character's signature mole below the right eye, present in 214/240 historical images, appeared in only 6 of the 20 new ones — proving the drift is real and not a change in the author's taste. A rare longitudinal, quantitative look at character consistency in AI image generation.
More from Multimodal
- ComfyUI to announce MiniMax H3 sync sound challenge winners in special livestream — MiniMax_AI · 2026-09-03
- Wan 2.2 Image-to-Video Tested in ComfyUI With Two-Stage High-to-Low Noise KSampler Pass — VictorVisuals · 2026-09-03
- AI influencers reshape content creation as Seedance 2.5 targets influencer vlogs — aftahi_ai · 2026-09-03
- Seedance 2.5 tested for AI influencer vlogs: 30-second clips with consistent identity — aftahi_ai · 2026-09-03
- A crowned, sneering AI portrait in fluorescent pink and yellow — misovalko · 2026-09-03
- Oxford & NUS Propose V-RAE: Replacing Pixel Reconstruction with Pretrained Visual Representations for Video Generation — jiqizhixin · 2026-09-03