One character across every medium
A character should be recognizably the same person in a chat, a still image, a short clip and a full episode, even when different models produce each one.
The question
How do appearance, voice, personality, memory and world history stay consistent across conversations, images, shorts and long-form video?
Audiences notice drift immediately: a face that changes between images, a voice that does not match the video, a personality that forgets what happened last week.
Consistency across formats is what turns a set of outputs into a character people follow.
Proposed method
- 1Maintain a shared character and world representation: visual references, voice profile, personality notes, memory and timeline.
- 2Condition every generation step on that shared representation rather than on free-text descriptions.
- 3Measure drift automatically with identity and style checks, and route hero assets to human review.
What a good outcome looks like
Characters that pass identity and personality checks across every format, with drift caught before anything is published.
Current status
Early workReference-based image consistency is in use; voice, memory and video continuity are not yet unified.