diffusers

diffusers copied to clipboard

ControlNet LoRA support more broadly across models

1

**Is your feature request related to a problem? Please describe.** In SimpleTuner, I'm currently extending ControlNet training to all implemented model architectures. So far, Control LoRA seems to be the...

bghira

Add VidTok AutoEncoders

5

We add VidTok, a versatile and state-of-the-art video tokenizer, as an autoencoder model to diffusers. Paper: https://arxiv.org/pdf/2412.13061 Code: https://github.com/microsoft/VidTok Model: https://huggingface.co/microsoft/VidTok

annitang1997

Dreambooth finetune FLUX dev CLIPTextModel

5

### Describe the bug ValueError: Sequence length must be less than max_position_embeddings (got `sequence length`: 77 and max_position_embeddings: 0 I used four A100 to full amount of fine-tuning Flux. 1...

Wuyiche

[torch.compile] Make HiDream torch.compile ready

10

# What does this PR do? Part of https://github.com/huggingface/diffusers/issues/11430 Trying to make the HiDream model fully compatible with `torch.compile()` but it fails with: https://pastebin.com/EbCFqBvw To reproduce run the following from...

sayakpaul

### 📄 7% (0.07x) speedup for ***`Kandinsky3ConditionalGroupNorm.forward` in `src/diffusers/models/unets/unet_kandinsky3.py`*** ⏱️ Runtime : **`2.16 milliseconds`** **→** **`2.02 milliseconds`** (best of `332` runs) ### 📝 Explanation and details Certainly! Here are the...

misrasaurabh1

[NewModel] Adds SEVA ( Stable Virtual Camera) model

4

# What does this PR do? Fixes #11151 ## Who can review? @a-r-r-o-w @yiyixuxu

ishan-modi

[Refactor] Initialize class members inside init method

2

# What does this PR do? Fixes #11002 ## Who can review? @DN6 @yiyixuxu

ishan-modi

[Feature] MultiControlNet support for SD3Impainting

1

# What does this PR do? Fixes #11208 ## Who can review? @yiyixuxu

ishan-modi

[FR] Please support MagCache: Fast Video Generation with Magnitude-Aware Cache

3

**Is your feature request related to a problem? Please describe.** No problem. The results are good and less-latency. Following models are supported [Wan2.1](https://github.com/Wan-Video/Wan2.1) [HunyuanVideo](https://github.com/Tencent/HunyuanVideo) [FLUX](https://github.com/black-forest-labs/flux), **Describe the solution you'd like.**...

nitinmukesh

Update pipeline_flux_fill.py

2

# What does this PR do? This PR adds the load_ip_adapter attribute to the FluxFillPipeline object. Fixes #11561 ## Before submitting - [ ] This PR fixes a typo or...

tanuj-rai