Pedro Cuenca

Flax: trickle down `norm_num_groups`

4

`FlaxEncoder` receives it [here](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/vae_flax.py#L326), but then it's ignored. `FlaxResnetBlock2D` uses a [hardcoded value](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/vae_flax.py#L99). Discovered by @younesbelkada in https://github.com/huggingface/diffusers/pull/589#discussion_r975796548

Stable Diffusion: providing a list of positive prompts and a list of negative prompts does not work as expected

1

### Describe the bug See this forum post: https://discuss.huggingface.co/t/stable-diffusion-bs-1-uses-negative-as-prompt/24130. In short: ```Python _ = pipe(["frog"]*2, negative_prompt=["bird"]*2) ``` Reaches this condition https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py#L247. As expected, `negative_prompt` is a list with the same...

CI testing on `mps`

1

We should add a couple of simple tests that use the `mps` device to detect issues like #760. I'm not familiar with how the CI system is set up or...

Handle `kwargs` and not serializable values as configuration attributes

1

Context: #655. A `kwargs` argument was added to schedulers in https://github.com/huggingface/diffusers/commit/d7dcba4a130a2eeecd46fc1d2ed244f54fd2a8f6 just to display a deprecation warning. `kwargs` was interpreted as a configurable attribute, but it could not be serialized....

JAX / Flax: avoid recompilation on params change: guidance scale, size, number of steps

10

This is required to improve the demo(s), we currently have the sliders disabled. Some users are also requesting it in discord and the forums. Our demo notebook is very fast...

Potential regression in deterministic outputs

19

### Describe the bug I've started noticing different outputs ~~in the latest version of diffusers~~ starting from diffusers `0.4.0` when compared against `0.3.0`. This is my test code (extracted from...

Implement `add_noise` in iPNDMScheduler

1

This method is missing from the recently-added IPNDMScheduler. It's not required for inference, but I think we should add it for consistency with all the others. TODO: - [ ]...

Implement v-prediction in all schedulers

3

This would be nice to guarantee scheduler interoperability in Stable Diffusion 2. DDIM, DDPM and EulerDiscrete already support v-prediction, the following ones remain (from the list of compatible stable diffusion...