diffusers
diffusers copied to clipboard
huggingface
🤗 Diffusers: State-of-the-art diffusion models for image and audio generation in PyTorch and FLAX.
### Describe the bug Textual inversion uses a deprecated import for the scaling methods in Pillow DeprecationWarning: BICUBIC is deprecated and will be removed in Pillow 10 (2023-07-01). Use Resampling.BICUBIC...
Like in the Imagen Video paper. According to Katherine Crowson it's more stable and faster: https://twitter.com/rivershavewings/status/1578193039423852544?s=21&t=oIToMoHbtU1U8VH3q4Mzng
Make sure `init_images` and `prompts` behavior matches.
Case in point: attempting to load a pipeline that uses the k-LMS scheduler using `from_pretrained` without `scipy` installed fails [here](https://github.com/huggingface/diffusers/blob/v0.4.1/src/diffusers/pipeline_utils.py#L457) because `load_method_name` is `None`. The error the user sees is:...
The Stable Diffusion pipeline accepts latents as an optional input for reproducibility purposes, see #247. In conversation with @tmabraham, he suggested extending this feature to other pipelines too. In addition...
Now that Pipelines return a well-defined PipelineOutput object, I suggest something along the lines of ```.py class StableDiffusionPipelineOutput: def images(self) -> List[PIL.Image.Image]: # […] def as_array(self) -> np.ndarray: # […]...
### Describe the bug When multiple images are passed to the `init_image` argument as a tensor, `init_latents` will have the same dimension 0 as `init_images`. But this line is further...
This fixes PIL.Image Resampler warnings, fixing bug https://github.com/huggingface/diffusers/issues/784
### Model/Pipeline/Scheduler description Some of the samplers (e.g. DDIM) are fully reversible. A simple API addition of `reverse_step()` or a mixin can be implemented. This will allow interpolating between images...
Metadata
Owner
Metadata
🤗 Diffusers: State-of-the-art diffusion models for image and audio generation in PyTorch and FLAX.