[Info Need]: Image2Image in Diffusers

In Playground AI, it provides

1 Filter
2 Prompt
3 Input-Image
4 Input-Image-Strength
5. Prompt-Guidance
6. Quality and Details

Does the 1, 4 and 6 support in Diffusers?

cc. @patrickvonplaten

hi @innat we support an input image strength parameter on our image variation pipelines. The quality and details section seems to indicate the number of diffusion steps the process is run for, we support this with the num_inference_steps parameter. I’m not sure how filter works. It could be anything from some text concatenated with the prompt to an auxiliary model.


My above quiers were emerged while workign with controlnet.

Could you provide some details of strenght and controlnet_conditioning_scale. Currently this trength is not documented in docstring. I think strenght params is added recently, last I checked it was available in community pipelines.

About filter, this effects the quality of diffusion model very strongly. I’m also not sure how they apply filter, is it a different model or some improvement in existing model components. What do you think aobut this? I’m not sure but if diffusers could support such feature, it would be really great.