Why are the generated videos all mid-shot？

[bobooblue]

Even though a full-length shot is specified in the prompt, the generated videos still only show half-body views.

[SHYuanBest]

Thanks for your interest! Maybe you can try adding negative_prompt to remove half-body views.

[bobooblue]

Thank you for your reply. I tried the CogVideoX-5b-I2V native model and found that the facial consistency is also very good. Will the task of Identity-Preserving Video Generation be resolved in the next upgrade of the base model, without the need to add face condition control?

[SHYuanBest]

This is a promising direction, and we are actively working towards it.