search

thu-ml / controlvideo

Official implementation for "ControlVideo: Adding Conditional Control for One Shot Text-to-Video Editing"
Apache License 2.0
214 stars 15 forks source link

ID Consistency #9

Open Carlyx opened 1 year ago

Carlyx commented 1 year ago

Hi, awesome work!

In fact, I use a dancing video as input, and try to change hair by "+ with red hair". However, I just get a different result in regard to identity and background (style). Do you have any suggestion for this result?

https://github.com/thu-ml/controlvideo/assets/34021717/c094c4aa-9f58-4b83-bb5a-ac45dc6ee7ed

gracezhao1997 commented 1 year ago

Hi, you can change pose control to Canny or HED boundary. These two controls contain detailed information of the source video and thus can preserve the source video content faithfully. You can also try to train the model on the source video using more steps, which allows a trade-off between faithfulness and editability.