Discussion about the pipelines API

Thread to collect all feedback regarding the public API of the code in src/diffusers/pipelines.

To better understand what exactly a pipeline is, please see: GitHub - huggingface/diffusers

Hello and thanks for building such an awesome and useful library! I have a question, in the example scripts you provide there is a strength parameter given to the pipeline call function. What is its purpose and what does it do? What is the apparent effect of increasing or decreasing it?

Specifically, in the official GitHub repo of the library (link) the lines:

with autocast("cuda"):
    images = pipe(prompt=prompt, init_image=init_image, mask_image=mask_image, strength=0.75).images

have the strength parameter. It would be best if you could provide the explanation in the docstring.

Hi @ehalit! Thanks a lot for your nice comments and for your question :slight_smile:

We recently added docstrings for the strength parameter here. Is there any other place where we may have missed them?

Oh I see, I was looking at the text-to-image generation docstring, as I understand this parameter is not related to this task. Maybe an indicator in the text-to-image docs would be helpful…