The resource consumption of this thing is really terrifying, and the 2048×2048 picture A100 actually ran for two and a half minutes.
The first two are anime pictures, and the last two are realistic photos, but the effect is really good. It may be that the implementation issue has some collapsed textures in some detail.
Anime images are recognized as photorealistic photos, which may be a prompt issue.
The original picture is placed in the push below, so you can compare the effects before and after generation.
If you want to experience it, you can use this Colab link to try, but forget it if it’s not Pro, T4 has to run for a long time: https://colab.research.google.com/github/camenduru/DemoFusion-colab/blob/main/DemoFusion_img2img_colab.ipynb
