AnimeBackgroundGAN (CartoonGAN by Chen et. al.)
- Makoto Shinkai (新海誠) pre-trained model from CartoonGAN
[Chen et al., CVPR18]
. - This model can transform real-life photos into Japanese-animation-like backgrounds, following the style of movies such as Kimi no Na wa with a photorealistic painting style.
- The implementation is in PyTorch (see source code here).
- Check out the demo here:
Other pre-trained model versions
The other versions were also trained from movies of the different Japanese animation directors.
Mamoru Hosoda(細田守)
- director of Wolf Children, with a distinct mild and cool background style
- Director Profile
- Model Repository: https://huggingface.co/akiyamasho/AnimeBackgroundGAN-Hosoda
Satoshi Kon(今敏)
- director of Paprika with a distinct high contrast, reddish hue style
- Director Profile
- Model Repository: https://huggingface.co/akiyamasho/AnimeBackgroundGAN-Kon
Hayao Miyazaki(宮崎駿)
- director of Howl's Moving Castle with a relatively soft and painterly style
- Director Profile
- Model Repository: https://huggingface.co/akiyamasho/AnimeBackgroundGAN-Miyazaki
Credits
- Paper at CartoonGAN: Generative Adversarial Networks for Photo Cartoonization
[Chen et al., CVPR18]
- Original PyTorch implementation was created by Yijun Li
- Spaces/Models re-packaging and implementation by Shō Akiyama.
Special Thanks
Inference API (serverless) does not yet support pytorch models for this pipeline type.