A little variation is normal — this is the nature of generative video models. But you can tighten consistency quite a bit with these tips:
- Create the character once, reuse it. Don’t re-describe them in every prompt — select the saved character from your Character library instead.
- Generate multi-view images after creating a character (front / side / back) — this gives the model stronger anchors.
- Keep style tags stable. Switching between “realistic” and “anime” mid-video will visibly change the character.
- Minimize pose / outfit changes in the prompt if you want the character to look identical across scenes.
We’re actively working on stronger consistency in upcoming model updates. ![]()