Testing AI character consistency across a 20-image travel sequence
SiennaVaire · reddit · 2026-09-23
The author is building a recurring fictional AI character, Sienna Vaire, and stress-testing identity preservation over a 20-image "Leaving London" set: same outfit, luggage, station and weather, while changing pose, framing and action.
The four published frames focus on three checks: face consistency, clothing continuity, and whether the sequence reads as one continuous scene rather than separate generations — the core difficulty of multi-image character generation.
More from Multimodal
- Given a 3-hour budget and a one-line prompt, Opus 5.5 produced a full Kowloon horror short itself — rainbird · 2026-09-23
- Image-gen prompt sharer techhalla posts 'safer highways' generations with prompts — techhalla · 2026-09-23
- Opus 5.5 recreates early-2000s stick figure fight YouTube animations — JasonBotterill · 2026-09-23
- 'Ordinary': An AI-generated Kowloon-set horror short turns heads on Reddit — rainbird · 2026-09-23
- Screenshot any character, one prompt recreates it — iamaliveix · 2026-09-23
- Google open-sources Noto 3D emoji — sculpting revealed their poodle only ever had three legs — GoogleDesign · 2026-09-23