Improving results with anime characters - Githubissues

fudan-generative-vision / champ

Champ: Controllable and Consistent Human Image Animation with 3D Parametric Guidance

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

MIT License

4.74k stars 598 forks source link

Improving results with anime characters #114

Open pearbender opened 6 months ago

pearbender commented 6 months ago

I tested this code on a couple of anime characters and consistently had issues with eyes. I noticed the training data includes no anime characters.

Can you please advise on how to improve results?
Should I retrain with characters?
Can the inference.yaml be tweaked to improve results?

Thanks

https://github.com/fudan-generative-vision/champ/assets/140408754/3e1e60db-890b-40a1-a00f-e4d83b53388c

anime2

https://github.com/fudan-generative-vision/champ/assets/140408754/ac967e87-1272-4736-bdf2-15588f169711

gabrielevccp commented 5 months ago

same problem, Does anyone know how to get better results?