11 Pull requests merged by 7 people

• Wan VACE

  #11582 merged Jun 6, 2025

• [tests] add test for torch.compile + group offloading

  #11670 merged Jun 6, 2025

• use deterministic to get stable result

  #11663 merged Jun 6, 2025

• [examples] flux-control: use num_training_steps_for_scheduler

  #11662 merged Jun 5, 2025

• [chore] bring PipelineQuantizationConfig at the top of the import chain.

  #11656 merged Jun 5, 2025

• [CI] Some improvements to Nightly reports summaries

  #11166 merged Jun 5, 2025

• [docs] Model cards

  #11112 merged Jun 2, 2025

• [docs] Caching methods

  #11625 merged Jun 2, 2025

• [tests] chore: rename lora model-level tests.

  #11481 merged Jun 2, 2025

• [chore] misc changes in the bnb tests for consistency.

  #11355 merged Jun 2, 2025

• Use float32 RoPE freqs in Wan with MPS backends

  #11643 merged Jun 2, 2025

16 Pull requests opened by 11 people

• Set _torch_version to N/A if torch is disabled.

  #11645 opened Jun 2, 2025

• Add Finegrained FP8

  #11647 opened Jun 3, 2025

• Use real-valued instead of complex tensors in Wan2.1 RoPE

  #11649 opened Jun 3, 2025

• Allow remote code repo names to contain "."

  #11652 opened Jun 3, 2025

• enable torchao test cases on XPU and switch to device agnostic APIs for test cases

  #11654 opened Jun 4, 2025

• [LoRA] support Flux Control LoRA with bnb 8bit.

  #11655 opened Jun 4, 2025

• Update pipeline_flux_inpaint.py to fix padding_mask_crop returning only the inpainted area

  #11658 opened Jun 4, 2025

• [WIP] [LoRA] support omi hidream lora.

  #11660 opened Jun 5, 2025

• Bump torch from 2.2.0 to 2.7.1 in /examples/research_projects/realfill

  #11664 opened Jun 5, 2025

• ⚡️ Speed up method `AutoencoderKLWan.clear_cache` by 886%

  #11665 opened Jun 5, 2025

• ⚡️ Speed up method `BlipImageProcessor.postprocess` by 51%

  #11666 opened Jun 5, 2025

• ⚡️ Speed up method `Kandinsky3ConditionalGroupNorm.forward` by 7%

  #11667 opened Jun 5, 2025

• Fix wrong param types, docs, and handles noise=None in scale_noise of FlowMatching schedulers

  #11669 opened Jun 6, 2025

• enable cpu offloading of new pipelines on XPU & use device agnostic empty to make pipelines work on XPU

  #11671 opened Jun 6, 2025

• [wip][tests] start adding tests for compilation + quantization

  #11672 opened Jun 6, 2025

• Support Expert loss for HiDream

  #11673 opened Jun 6, 2025

6 Issues closed by 6 people

• [BUG]: Using args.max_train_steps even if it is None in diffusers/examples/flux-control

  #11661 closed Jun 5, 2025

• Diffusers Transformer Pipeline Produces ComplexDouble Tensors on MPS, Causing Conversion Error

  #10986 closed Jun 3, 2025

• CompVis/stable-diffusion-v1-4 is missing fp16 files

  #11460 closed Jun 3, 2025

• import error

  #11560 closed Jun 3, 2025

• DreamO/Flux group offload: Expected all tensors to be on the same device, but found at least two devices

  #11628 closed Jun 2, 2025

• StableDiffusion3Pipeline with InstantX/SD3.5-Large-IP-Adapter

  #11627 closed Jun 2, 2025

12 Issues opened by 12 people

• HunyuanVideoImageToVideoPipeline memory leak

  #11676 opened Jun 7, 2025

• Error in loading the pretrained lora weights

  #11675 opened Jun 6, 2025

• [FR] Please support ref image and multiple control videos in Wan VACE

  #11674 opened Jun 6, 2025

• LoRA load issue

  #11659 opened Jun 4, 2025

• Custom Wan diffusion Lora runs without error but doesn't apply effect and gives warning: No LoRA keys associated to WanTransformer3DModel found with the prefix='transformer'.

  #11657 opened Jun 4, 2025

• support .alpha keys in HiDream loras trained using OneTrainer

  #11653 opened Jun 3, 2025

• Can't load flux-fill-lora with FluxControl

  #11651 opened Jun 3, 2025

• max_shard_size

  #11650 opened Jun 3, 2025

• how to load lora weight with fp8 transfomer model?

  #11648 opened Jun 3, 2025

• Error in init from pretrained for LTXConditionPipeline

  #11644 opened Jun 2, 2025

• prompt_embeds_scale in FluxPriorReduxPipeline seems to have no effect.

  #11642 opened Jun 1, 2025

• [BUG] [CleanCode] Tuple[int] = (16, 56, 56) in FluxTransformer2DModel

  #11641 opened May 31, 2025

29 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.

• [LoRA] parse metadata from LoRA and save metadata

  #11324 commented on Jun 6, 2025 • 12 new comments

• [benchmarks] overhaul benchmarks

  #11565 commented on Jun 6, 2025 • 5 new comments

• Add SkyReels V2: Infinite-Length Film Generative Model

  #11518 commented on Jun 6, 2025 • 2 new comments

• Pipeline doc pr

  #11636 commented on Jun 2, 2025 • 1 new comment

• [`Wan`] Fix VAE sampling mode in `WanVideoToVideoPipeline`

  #11639 commented on Jun 2, 2025 • 0 new comments

• Add community class StableDiffusionXL_T5Pipeline

  #11626 commented on Jun 3, 2025 • 0 new comments

• Added PhotoDoodle Pipeline

  #11621 commented on Jun 5, 2025 • 0 new comments

• Introduce DeprecatedPipelineMixin to simplify pipeline deprecation process

  #11596 commented on Jun 2, 2025 • 0 new comments

• Chroma as a FLUX.1 variant

  #11566 commented on Jun 4, 2025 • 0 new comments

• [torch.compile] Make HiDream torch.compile ready

  #11477 commented on Jun 6, 2025 • 0 new comments

• [quant] add __repr__ for better printing of configs.

  #11452 commented on Jun 2, 2025 • 0 new comments

• [NewModel] Adds SEVA ( Stable Virtual Camera) model

  #11440 commented on Jun 2, 2025 • 0 new comments

• Attention Dispatcher

  #11368 commented on Jun 4, 2025 • 0 new comments

• [loaders] make the `loaders` folder structure cleaner

  #11335 commented on Jun 2, 2025 • 0 new comments

• [performance] investigating FluxPipeline for recompilations on resolution changes

  #11360 commented on Jun 7, 2025 • 0 new comments

• SD3 ControlNet Script (and others?): dataset preprocessing cache depends on unrelated arguments

  #11497 commented on Jun 5, 2025 • 0 new comments

• InstructPix2Pix training script for SD3

  #9101 commented on Jun 4, 2025 • 0 new comments

• [FR] Please support WanVACETransformer3DModel.from_single_file

  #11630 commented on Jun 3, 2025 • 0 new comments

• model_cpu_offload failed in unidiffusers pipeline

  #11443 commented on Jun 3, 2025 • 0 new comments

• onnx export failure - timestep parameter with static value

  #11456 commented on Jun 3, 2025 • 0 new comments

• enable_xformers_memory_efficient_attention in the training script

  #11486 commented on Jun 3, 2025 • 0 new comments

• OMI Format Compatibility

  #11631 commented on Jun 3, 2025 • 0 new comments

• How to load lora weight in distribution applications?

  #11637 commented on Jun 3, 2025 • 0 new comments

• [Quantization] Support TRT as a backend

  #11032 commented on Jun 2, 2025 • 0 new comments

• FlaxUNet2DConditionModel is not initialized with correct dtypes

  #11144 commented on Jun 2, 2025 • 0 new comments

• When I use deepspeed_zero3 to train the controlnet on flux, an error occured "RuntimeError: tensor does not have a device".

  #11633 commented on Jun 2, 2025 • 0 new comments

• [Models] handle initialization of new layers in a partially pre-trained model better

  #11279 commented on Jun 2, 2025 • 0 new comments

• Hotswapping multiple LoRAs throws a peft key error.

  #11298 commented on May 31, 2025 • 0 new comments

• Train_controlnet_sdxl.py and tensorboard log images

  #11470 commented on May 31, 2025 • 0 new comments