Add Stable Diffusion Interpolation Example

[nateraw]

Adds interpolate_stable_diffusion.py, which is a pipeline that lets you generate images as you interpolate between different prompts/seeds.

Its __call__ fn is the same as StableDiffusionPipeline, but with the added text_embeddings kwarg. The walk function is where the logic happens. Fine with renaming these if need be so that walk becomes __call__.

Here's how I have been using it (from within examples/community dir)

from interpolate_stable_diffusion import StableDiffusionWalkPipeline
import torch

pipe = StableDiffusionWalkPipeline.from_pretrained(
    "CompVis/stable-diffusion-v1-4",
    revision='fp16',
    torch_dtype=torch.float16,
    safety_checker=None,  # Very important for videos...lots of false positives while interpolating
).to('cuda')

frame_filepaths = pipe.walk(
    prompts=['a dog', 'a cat', 'a horse'],
    seeds=[42, 1337, 1234],
    num_interpolation_steps=16,
    output_dir='./dreams',
    batch_size=24,
    height=512,
    width=512,
    guidance_scale=8.5,
    num_inference_steps=50,
)

Though it seems (I think) that once it's merged, you'll be able to use community='interpolate_stable_diffusion' in the from_pretrained fn instead of having to be in the examples/community dir.

[nateraw]

Few notes

[nateraw]

btw fixed this table here... @patrickvonplaten your name was under the Description column.

[nateraw]

will add notebook once this is merged

[nateraw]

prompt becomes optional, as you can either pass that or text_embeddings

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[patrickvonplaten]

Thanks a lot @nateraw !