Additional experiments about controls

[justinday123]

Thank you for your nice work and code released. I want to know if your team have a plan to make some results adding controlnet such as pose or depth?

[SHYuanBest]

Thanks for your interest! We have this plan, but our computing resources are currently being rebuilt, so I'm not sure if we'll be able to implement this in the end, stay tuned.

[justinday123]

Thank you for your response. I want to know, if i want to use additional control such as pose or depth, is additional training is required?

[SHYuanBest]

If you're using CogVideoX-based ControlNet and the DiT parameters were frozen during its training, I think it can be directly integrated with ConsisID without additional training. Otherwise, training would still be necessary. Of course, if you're employing some zero-shot control strategies, training isn't required either.