One character, forty shots, zero drift
A motion designer ran a brand mascot through a 40-shot social campaign — same character, consistent across angles, lighting, and motion — using a Flux character LoRA feeding Kling for animation. Six months ago this needed a 3D pipeline and a render farm.
Consistent characters were the wall between AI video and real campaign work. The wall just moved.
The how
- Character sheet first: 60 renders of the mascot across angles/expressions, generated then hand-curated to 25 for LoRA training.
- Flux LoRA locks identity; every campaign still is generated against it with shot-specific prompts from the boards.
- Stills become Kling image-to-video shots, 4–8 seconds each; motion prompts kept minimal to avoid identity drift.
- Drift QA: every tenth frame auto-compared against the character sheet; failing shots regenerate with tighter conditioning.
- Edit assembled conventionally — the AI replaces the render pipeline, not the editor.
Run it back
Character sheet discipline is everything: curate training images like you’re casting. Budget two days for the LoRA loop, then shots are minutes each.