I can't test it, can you tell me the specific test command?

[Valentina-tinder]

Looking forward to your reply

[Arlaz]

You need to launch the inference/generate.py file inside an appropriate environment.
Feel free to replace the input.mp4 video with yours

[Valentina-tinder]

Thanks for your reply. What a great work, an advancement in the field~ I run it successfully. Can I add depth controlnet (lllyasviel/control_v11f1p_sd15_depth) directly to the test without training the onlyflow model? Will this increase the stability of video generation?

4.29.3.-1.mp4

https://github.com/user-attachments/assets/e3180b4f-e0f7-4a83-9da3-b09f8966b7a7

1.mp4

Can you tell me if retraining this model with more channels using optical flow + depth concatenate input will be better? Thank you very much