Generating consistent features

fudan-generative-vision / hallo

Hallo: Hierarchical Audio-Driven Visual Synthesis for Portrait Image Animation

https://fudan-generative-vision.github.io/hallo/

MIT License

9.5k stars 1.3k forks source link

Generating consistent features #60

Open piyushK52 opened 5 months ago

piyushK52 commented 5 months ago

I have noticed that teeth in the generated videos are always a hit or a miss, is it because of the nature of base ad? (that it finds it difficult to maintain consistency in smaller elements). Also, do you guys know how this can be avoided/minimized ?

https://github.com/fudan-generative-vision/hallo/assets/34690994/b0c4291a-452b-4e7c-bae5-4c8a12ae2d4c

subazinga commented 5 months ago

Yes, that a known problem. We are still fixing it. Currently, maybe gfpgan can help. It is also used by SadTalker.

piyushK52 commented 5 months ago

Nice! Looking forward to it