Overview

16 Pull requests merged by12 people

• Fix kontext finetune issue when batch size >1

  #11921 mergedJul 18, 2025

• [docs] include bp link.

  #11952 mergedJul 18, 2025

• [docs] Update toctree

  #11936 mergedJul 18, 2025

• [Modular] Updates for Custom Pipeline Blocks

  #11940 mergedJul 18, 2025

• [docs] clarify the mapping betweenTransformer2DModel and finegrained variants.

  #11947 mergedJul 18, 2025

• [refactor] Flux/Chroma single file implementation + Attention Dispatcher

  #11916 mergedJul 17, 2025

• Add SkyReels V2: Infinite-Length Film Generative Model

  #11518 mergedJul 16, 2025

• [training] add an offload utility that can be used as a context manager.

  #11775 mergedJul 16, 2025

• enable flux pipeline compatible with unipc and dpm-solver

  #11908 mergedJul 16, 2025

• [ControlnetUnion] Propagate #11888 to img2img

  #11929 mergedJul 16, 2025

• Fixed bug: Uncontrolled recursive calls that caused an infinite loop when loading certain pipelines containing Transformer2DModel

  #11923 mergedJul 16, 2025

• Remove forced float64 from onnx stable diffusion pipelines

  #11054 mergedJul 16, 2025

• Remove device synchronization when loading weights

  #11927 mergedJul 15, 2025

• [tests] Improve Flux tests

  #11919 mergedJul 15, 2025

• Bump aiohttp from 3.10.10 to 3.12.14 in /examples/server

  #11924 mergedJul 15, 2025

• Fix: Align VAE processing in ControlNet SD3 training with inference

  #11909 mergedJul 14, 2025

14 Pull requests opened by7 people

• [modular diffusers] Wan

  #11913 openedJul 12, 2025

• [refactor] Wan single file implementation

  #11918 openedJul 14, 2025

• [tests] Add test slices for Wan

  #11920 openedJul 14, 2025

• [WIP] Speed up PR CI

  #11926 openedJul 15, 2025

• [docs] Modular diffusers

  #11931 openedJul 15, 2025

• [docs] update `guidance_scale` docstring for guidance_distilled models.

  #11935 openedJul 15, 2025

• [Modular] Fast Tests

  #11937 openedJul 16, 2025

• Context Parallel w/ Ring & Ulysses & Unified Attention

  #11941 openedJul 16, 2025

• [Modular] support standard repo

  #11944 openedJul 17, 2025

• Teacache implemented into hunyuan video framepack transformer model

  #11949 openedJul 17, 2025

• [WIP] SVDQuant

  #11950 openedJul 17, 2025

• [tests] Add fast test slices for HiDream-Image

  #11953 openedJul 18, 2025

• [tests] Add test slices for Hunyuan Video

  #11954 openedJul 18, 2025

• [tests] Add test slices for Cosmos

  #11955 openedJul 18, 2025

12 Issues closed by5 people

• Implement caching on LTX and WAN video models

  #11134 closedJul 18, 2025

• Kontext model loading quantization problem

  #11951 closedJul 18, 2025

• [Feature request] Integrate SkyReels-V2 support in diffusers

  #11374 closedJul 16, 2025

• flux pipeline not work with DPMSolverMultistepScheduler and UniPCMultistepScheduler

  #11907 closedJul 16, 2025

• [Docs] Visualizing the codebase for diffusers

  #11866 closedJul 15, 2025

• Support for Large attn_bias via Sparse Tensors or On‑The‑Fly Construction (seq_len ≈ 12 288)

  #11928 closedJul 15, 2025

• Loading multiple LoRAs to 1 pipeline in parallel, 1 LoRA to 2-pipelines on 2-GPUs

  #11914 closedJul 15, 2025

• refactors pipelines inheriting from `StableDiffusionPipeline`

  #6984 closedJul 15, 2025

• Question about control sd3

  #11867 closedJul 14, 2025

• SD3ControlNet

  #11917 closedJul 14, 2025

• WanVACETransformer3DModel with GGUF not working for 1.3B model

  #11878 closedJul 12, 2025

• Issue with FluxKontextPipeline

  #11886 closedJul 12, 2025

12 Issues opened by12 people

• Flash/Sage varlen does not work with torch.compile

  #11957 openedJul 18, 2025

• Frequency-Decoupled Guidance (FDG) for diffusion models

  #11956 openedJul 18, 2025

• Impossible to load WanTransformer3DModel when offline using the 'from_pretrained' function.

  #11948 openedJul 17, 2025

• docs

  #11946 openedJul 17, 2025

• Floating point exception with nightly PyTorch and CUDA

  #11945 openedJul 17, 2025

• ValueError when loading Lycoris safetensors LoRA: keys not correctly renamed

  #11943 openedJul 16, 2025

• Kontext pipeline resizing image even if it is within PREFERRED_KONTEXT_RESOLUTIONS

  #11942 openedJul 16, 2025

• train flux_controlnet code need to update

  #11939 openedJul 16, 2025

• RMSNorm's weight not registered as submodules when initializing

  #11938 openedJul 16, 2025

• how to run convert_cosmos_to_diffusers.py correctly?

  #11930 openedJul 15, 2025

• FLUX.1-Fill-dev running problem

  #11922 openedJul 14, 2025

• Create modular pipeline from existing pipeline

  #11915 openedJul 12, 2025

20 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• fix "Expected all tensors to be on the same device, but found at least two devices" error

  #11690 commented onJul 18, 2025 • 2 new comments

• [LoRA training] add aspect ratio bucketing

  #11438 commented onJul 16, 2025 • 2 new comments

• [fix] fix for prior preservation and mixed precision sampling

  #11873 commented onJul 16, 2025 • 0 new comments

• Propose to refactor output normalization in several transformers

  #11850 commented onJul 18, 2025 • 0 new comments

• [ci] enable hotswapping tests on our nightly CI.

  #11826 commented onJul 16, 2025 • 0 new comments

• Add `--lora_alpha` and metadata handling for `train_dreambooth_lora_hidream`

  #11765 commented onJul 16, 2025 • 0 new comments

• Add MAGI-1: Autoregressive Video Generation at Scale

  #11713 commented onJul 18, 2025 • 0 new comments

• Bump requests from 2.32.3 to 2.32.4 in /examples/server

  #11686 commented onJul 15, 2025 • 0 new comments

• [GGUF] feat: support loading diffusers format gguf checkpoints.

  #11684 commented onJul 16, 2025 • 0 new comments

• Support for control-lora

  #10686 commented onJul 15, 2025 • 0 new comments

• Different generation with `Diffusers` in I2V tasks for LTX-video

  #10565 commented onJul 18, 2025 • 0 new comments

• Wan Video Vae RemoteVae

  #11326 commented onJul 18, 2025 • 0 new comments

• Multi-gpus Context Parallel training support?

  #11020 commented onJul 18, 2025 • 0 new comments

• F5-TTS Integration

  #10043 commented onJul 18, 2025 • 0 new comments

• [Performance] Issue on *SanaLinearAttnProcessor2_0 family. 1.06X speedup can be reached with a simple change.

  #11499 commented onJul 17, 2025 • 0 new comments

• Add SUPIR Upscaler

  #7219 commented onJul 16, 2025 • 0 new comments

• Bug: lora trained with train_dreambooth_lora_sdxl_advanced.py, the weight cannot be loaded

  #11790 commented onJul 16, 2025 • 0 new comments

• Dreambooth finetune FLUX dev CLIPTextModel

  #10925 commented onJul 16, 2025 • 0 new comments

• Add pruna integration for loading model through diffusers.from_pretrained / pipeline

  #11700 commented onJul 15, 2025 • 0 new comments

• HunyuanVIdeo in diffusers use negative_prompt but generate wrong video

  #10901 commented onJul 15, 2025 • 0 new comments