Img-2-Img generation with diffuser pipeline "Inpainting conditioning mask strength 0"

#18
by Parsley3000 - opened

Hey I'm working on a project & want to change the art style of images but keep them close to their original form
Similar to this:

image.png

Video link: https://www.youtube.com/watch?v=VdDhGjkKD1A&ab_channel=Aitrepreneur

Want to perform this using diffusers import StableDiffusionImg2ImgPipeline
Not sure if this feature is available in the pipeline but want "Inpainting conditioning mask strength set to 0"

Wondering what arguments I need in the pipeline to achieve this
Currently using (*snippet):

model_path = "runwayml/stable-diffusion-inpainting"
pipe = StableDiffusionImg2ImgPipeline.from_pretrained(
model_path,
revision="fp16",
torch_dtype=torch.float16,
use_auth_token=auth_token
)

images = pipe(prompt=prompt, image=face_img, strength=0.75, guidance_scale=7.5).images

image.png

I have the same question. How to set this hyper parameter in the pipeline (outside the webui)?

@Parsley3000 were you able to fix this issue? I am struggling with the same problem, the sd-v1-5-inpainting.ckpt has a different input tensor shape then the normal ckpt and the pipeline for img2img does not work with it. I checked AUTOMATIC111's code and it seems he is manually loading the weights and using it as img2img without any mask. I think some modifications to the StableDiffusionImg2ImgPipeline might work but I am no expert to do those changes.

No still haven't found a solution

Sign up or log in to comment