Does your code support resuming training from where it left off after an interruption?

[Shuweis]

How to resume training from where it left off after an interruption?

[sculmh]

The configuration file contains checkpointing_steps, which saves the model every n steps. After an interruption, you can change the pretrained_model_path to the checkpoint from the most recent step.

[Shuweis]

Ok，thank you for your reply! And does your code on SVD model support the text condition and motion strength control?

[sculmh]

1. The original img2video-based SVD model does not support text conditions.
2. SVD allows control over motion speed using motion_bucket_id.
3. Our version only fine-tunes with motion_mask on top of SVD, so it also does not support text conditions.