29 Pull requests merged by 16 people

• fix flux type hint

  #12089 merged Aug 7, 2025

• Implement Frequency-Decoupled Guidance (FDG) as a Guider

  #11976 merged Aug 7, 2025

• [tests] tighten compilation tests for quantization

  #12002 merged Aug 7, 2025

• Make prompt_2 optional in Flux Pipelines

  #12073 merged Aug 7, 2025

• Helper functions to return skip-layer compatible layers

  #12048 merged Aug 6, 2025

• Fix group offloading synchronization bug for parameter-only GroupModule's

  #12077 merged Aug 6, 2025

• [refactor] condense group offloading

  #11990 merged Aug 6, 2025

• fix input shape for WanGGUFTexttoVideoSingleFileTests

  #12081 merged Aug 6, 2025

• Add cuda kernel support for GGUF inference

  #11869 merged Aug 5, 2025

• optimize QwenImagePipeline to reduce unnecessary CUDA synchronization

  #12072 merged Aug 5, 2025

• [tests] deal with the failing AudioLDM2 tests

  #12069 merged Aug 5, 2025

• [tests] Fix Qwen test_inference slices

  #12070 merged Aug 5, 2025

• [docs] small corrections to the example in the Qwen docs

  #12068 merged Aug 5, 2025

• [wip] feat: support lora in qwen image and training script

  #12056 merged Aug 5, 2025

• [docs] Install

  #12026 merged Aug 4, 2025

• fix the rest for all GPUs in CI

  #12064 merged Aug 4, 2025

• enable all gpus when running ci.

  #12062 merged Aug 4, 2025

• CI fixing

  #12059 merged Aug 4, 2025

• fix(qwen-image): update vae license

  #12063 merged Aug 4, 2025

• Cross attention module to Wan Attention

  #12058 merged Aug 4, 2025

• tests + minor refactor for QwenImage

  #12057 merged Aug 4, 2025

• [wan2.2] follow-up

  #12024 merged Aug 4, 2025

• Qwen-Image

  #12055 merged Aug 3, 2025

• Update autoencoder_kl_cosmos.py

  #12045 merged Aug 2, 2025

• Fix type of force_upcast to bool

  #12046 merged Aug 2, 2025

• [LoRA] support lightx2v lora in wan

  #12040 merged Aug 2, 2025

• Allow SD pipeline to use newer schedulers, eg: FlowMatch

  #12015 merged Aug 1, 2025

• [wan2.2] fix vae patches

  #12041 merged Aug 1, 2025

• [training-scripts] Make pytorch examples UV-compatible

  #12000 merged Jul 31, 2025

14 Pull requests opened by 11 people

• Add Flix Fill ControlNet

  #12035 opened Jul 31, 2025

• Add UV script comment block

  #12036 opened Jul 31, 2025

• support `hf_quantizer` in cache warmup.

  #12043 opened Aug 1, 2025

• `UNet2DConditionModel`: add support for QK Normalization by propagating `qk_norm` value from config through to child attention modules

  #12051 opened Aug 3, 2025

• enable compilation in qwen image.

  #12061 opened Aug 4, 2025

• [WIP][modular] sdxl methods refactor

  #12067 opened Aug 5, 2025

• [Wan 2.2 LoRA] add support for 2nd transformer lora loading + wan 2.2 lightx2v lora

  #12074 opened Aug 5, 2025

• gguf new quant type support (with demo)

  #12076 opened Aug 5, 2025

• [core] add modular support for Flux I2I

  #12086 opened Aug 6, 2025

• Fix Qwen-Image long prompt dimension mismatch error (issue #12083)

  #12087 opened Aug 6, 2025

• fix(fluX image validation): convert RGB to BGR for OpenPose images

  #12090 opened Aug 7, 2025

• Add Dream 7B Diffusion Large Language Model Pipeline

  #12091 opened Aug 7, 2025

• [docs] diffusers gguf checkpoints

  #12092 opened Aug 7, 2025

• [model] LTX Video 0.9.8

  #12095 opened Aug 7, 2025

17 Issues closed by 10 people

• WanTransformer3DModel.from_single_file wont load Wan2.2 GGUF (NotImplementedError: Cannot copy out of meta tensor; no data)

  #12009 closed Aug 7, 2025

• CPU Offloading Inefficiency: UNet/Transformer Repeatedly Moved Between CPU/GPU During Loops

  #11872 closed Aug 7, 2025

• Frequency-Decoupled Guidance (FDG) for diffusion models

  #11956 closed Aug 7, 2025

• Qwen Image : Image Editing Inference

  #12080 closed Aug 6, 2025

• Groupoffloading introduce bad results

  #11981 closed Aug 6, 2025

• Fusing Lightx2v lora on Wan2.2 GGUF fails

  #12047 closed Aug 6, 2025

• Wan 2.2 VAE forward fails

  #12039 closed Aug 6, 2025

• WanVACETransformer3DModel load gguf error

  #12085 closed Aug 6, 2025

• flux kontext transformer single blocks forward behavior changed

  #12071 closed Aug 5, 2025

• Qwen Image | Image-To-Image + Editing + Inpainting

  #12065 closed Aug 4, 2025

• Invalid API call in Cosmos VAE Encoder

  #12025 closed Aug 2, 2025

• FLUX.1-Kontext-dev Support for GGUF Quantized Model

  #11962 closed Aug 2, 2025

• Support training variance learning (Improved DDPM) with VLB loss.

  #11998 closed Aug 2, 2025

• Kontext pipeline resizing image even if it is within PREFERRED_KONTEXT_RESOLUTIONS

  #11942 closed Aug 2, 2025

• Wan 2.2 WanTransformer3DModel not compatible with Lightx2v self-forcing guidance distilled loras

  #12037 closed Aug 2, 2025

• Wan 2.2 First vs Second stage

  #12019 closed Aug 1, 2025

• Wan 2.2 5b i2v results poor quality compared to official Wan HF Space

  #12034 closed Aug 1, 2025

17 Issues opened by 16 people

• Qwen image transformers doesn't currently support from_single_file (i.e. GGUFs)

  #12098 opened Aug 7, 2025

• Wan2.2 TI2V-5B VRAM OOM at the end

  #12097 opened Aug 7, 2025

• WanVACEPipeline - doesn't work with apply_group_offloading

  #12096 opened Aug 7, 2025

• [Wan2.2] pipeline_wan miss the 'shift' parameter which used by Wan2.2-A14B-diffusers.

  #12094 opened Aug 7, 2025

• Will `cosmos-transfer1` be supported in diffusers in the future?

  #12084 opened Aug 6, 2025

• Qwen-Image long prompt will cause error

  #12083 opened Aug 6, 2025

• wan2.1 vae take more gpu memory after compile

  #12082 opened Aug 6, 2025

• API Suggestion: Expose Methods to Convert to Sample Prediction in Schedulers

  #12079 opened Aug 6, 2025

• Problem with provided example validation input in the Flux Control finetuning example

  #12078 opened Aug 5, 2025

• Qwen Image prompt encoding is not padding to max seq len

  #12075 opened Aug 5, 2025

• Qwen Image incorrect device assignment during prompt encode

  #12066 opened Aug 4, 2025

• Is there any DiT block defined in the huggingface/diffusers OR huggingface/transformers project?

  #12060 opened Aug 4, 2025

• Kontext Lora

  #12053 opened Aug 3, 2025

• Wan 2.2 with LightX2V offloading tries to multiply tensors from different devices and fails

  #12052 opened Aug 3, 2025

• `UNet2DConditionModel` : `qk_norm` setting in `config.json` is ignored

  #12050 opened Aug 3, 2025

• AttributeError: 'bool' object has no attribute '__module__'. Did you mean: '__mod__'?

  #12044 opened Aug 2, 2025

• Dataset structure for train_text_to_image_lora.py

  #12038 opened Jul 31, 2025

22 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.

• Bria 3 2 pipeline

  #12010 commented on Aug 7, 2025 • 34 new comments

• [WIP Modular] More Updates for Custom Code Loading

  #11969 commented on Aug 7, 2025 • 6 new comments

• [lora] adapt new LoRA config injection method

  #11999 commented on Aug 7, 2025 • 3 new comments

• [modular diffusers] Wan I2V/FLF2V

  #11997 commented on Aug 6, 2025 • 3 new comments

• Torchao int4 serialization

  #11591 commented on Aug 5, 2025 • 1 new comment

• try to use deepseek with an agent to auto i18n to zh

  #12032 commented on Aug 7, 2025 • 0 new comments

• Add IP-Adapter

  #12030 commented on Aug 3, 2025 • 0 new comments

• [wip][core] parallel loading of shards

  #12028 commented on Aug 7, 2025 • 0 new comments

• [lora] factor out the overlaps in `save_lora_weights()`.

  #12027 commented on Aug 5, 2025 • 0 new comments

• [core] respect `local_files_only=True` when using sharded checkpoints

  #12005 commented on Aug 7, 2025 • 0 new comments

• [chore] complete the licensing statement.

  #12001 commented on Aug 7, 2025 • 0 new comments

• Add F5 TTS pipeline

  #11958 commented on Aug 3, 2025 • 0 new comments

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

  #11873 commented on Jul 31, 2025 • 0 new comments

• [wip] Higgs quantization

  #11797 commented on Aug 7, 2025 • 0 new comments

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

  #11765 commented on Jul 31, 2025 • 0 new comments

• Add MAGI-1: Autoregressive Video Generation at Scale

  #11713 commented on Aug 7, 2025 • 0 new comments

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

  #11684 commented on Aug 7, 2025 • 0 new comments

• [Quantization] Add TRT-ModelOpt as a Backend

  #11173 commented on Aug 3, 2025 • 0 new comments

• Dream 7B

  #12017 commented on Aug 7, 2025 • 0 new comments

• _flash_attention_3 in dispatch_attention_fn is not compatible with the latest flash-atten interface.

  #12022 commented on Aug 2, 2025 • 0 new comments

• Floating point exception with nightly PyTorch and CUDA

  #11945 commented on Aug 2, 2025 • 0 new comments

• AnimateDiff SparseCtrl RGB does not work as expected

  #9508 commented on Aug 1, 2025 • 0 new comments