Spaces:
Running
Image2image
Is it possible to add a feature that can do image2image. That will be pretty awesome.
Not for all models, each model needs to be converted to an img2img version of diffusers, or alternatively, using CPU inference with a version that requires low step count such as ones merged with HyperSD1.5 (though, the results are different) - OR, alternatively, having a version that takes 15 minutes for image instead of 10 seconds, but at that point that exists in a space like this: https://huggingface.co/spaces/Yntec/Anything7.0-Webui-CPU which has img2img and other features like controlnet, we'd need to create spaces like this for the models that we want to use.
The problem is we are relying on txt2img technology that can't do img2img but if you see a space that is not running on ZeroGPU and is allowing img2img with these speeds I could copy their code and have it implemented today!