JoJoGAN VS AnimeGANv3

I've tried style transfer and some GANs like this one: https://github.com/mchong6/JoJoGAN

🔗 https://github.com/mchong6/JoJoGAN

And I quite liked and wanted to train this model with my own dataset, and always fell into the same error CUDA Out of Memory again and again. After searching across the internet for answers ended up finding it -> here.

Code for https://arxiv.org/abs/2112.11641 found: https://github.com/mchong6/JoJoGAN

Since you want to use multiple images for style, it reminded me of this pipeline. In it you can take a dataset of style images and finetune a pretrained model for 500-1000 iterations to achieve the style from these images on new ones. I am not sure if that pipeline is what you need or not though, because it is built for faces in particular, but maybe you can take inspiration from their approach for a generic method.

Here are some models that do that with face images that I have tried out: AnimeGANv3 - this one just came out ArcaneGAN - for faces JoJoGAN - for faces

github: https://github.com/mchong6/JoJoGAN

[Not For Commercial Use, Unless With Author's Consent] AnimeGAN Demo, by Long (Tony) Lian. You may need a fairly strong computer for this one, or the website may otherwise crash. The original GitHub repository was made by Asher Chan, who you need to contact in the case you want to use these materials commercially.

(2021.12.25) AnimeGANv3 has been released.🎄

Here are some models that do that with face images that I have tried out: AnimeGANv3 - this one just came out ArcaneGAN - for faces JoJoGAN - for faces