23 Pull requests merged by 10 people

• Modular custom config object serialization

  #11868 merged Jul 5, 2025

• Modular Guider ConfigMixin

  #11862 merged Jul 5, 2025

• Modular PAG Guider

  #11860 merged Jul 4, 2025

• Fix Wan AccVideo/CausVid fuse_lora

  #11856 merged Jul 4, 2025

• FIX set_lora_device when target layers differ

  #11844 merged Jul 4, 2025

• [benchmarks] overhaul benchmarks

  #11565 merged Jul 4, 2025

• [docs] Add a note of _keep_in_fp32_modules

  #11851 merged Jul 2, 2025

• [Flux Kontext] Support Fal Kontext LoRA

  #11823 merged Jul 2, 2025

• update: FluxKontextInpaintPipeline support

  #11820 merged Jul 2, 2025

• [lora] tests for exclude_modules with Wan VACE

  #11843 merged Jul 2, 2025

• [From Single File] support from_single_file method for WanVACE3DTransformer

  #11807 merged Jul 2, 2025

• fix norm not training in train_control_lora_flux.py

  #11832 merged Jul 2, 2025

• [docs] Deprecated pipelines

  #11838 merged Jul 2, 2025

• [docs] Batch generation

  #11841 merged Jul 2, 2025

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

  #11649 merged Jul 1, 2025

• [docs] fix single_file example.

  #11847 merged Jul 1, 2025

• [single file] Cosmos

  #11801 merged Jul 1, 2025

• [tests] Fix failing float16 cuda tests

  #11835 merged Jul 1, 2025

• reset deterministic in tearDownClass

  #11785 merged Jul 1, 2025

• [tests] add test for hotswapping + compilation on resolution changes

  #11825 merged Jul 1, 2025

• Remove print statement in SCM Scheduler

  #11836 merged Jun 30, 2025

• ENH: Improve speed of function expanding LoRA scales

  #11834 merged Jun 30, 2025

• [lora]feat: use exclude modules to loraconfig.

  #11806 merged Jun 30, 2025

10 Pull requests opened by 7 people

• [docs] torch.compile blog post

  #11837 opened Jun 30, 2025

• [tests] unbloat `tests/lora/utils.py`

  #11845 opened Jul 1, 2025

• [docs] LoRA metadata

  #11848 opened Jul 1, 2025

• Propose to refactor output normalization in several transformers

  #11850 opened Jul 2, 2025

• [docs] fix references in flux pipelines.

  #11857 opened Jul 4, 2025

• [training ] add Kontext i2i training

  #11858 opened Jul 4, 2025

• [fix] VaeImageProcessor.resize method can take list as input

  #11863 opened Jul 4, 2025

• Add cuda kernel support for GGUF inference

  #11869 opened Jul 5, 2025

• Add STORK scheduler

  #11871 opened Jul 6, 2025

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

  #11873 opened Jul 6, 2025

16 Issues closed by 5 people

• Multi-ID customization with CogvideoX 5B

  #10462 closed Jul 5, 2025

• When using gradient accumulation, optimizer.step() runs every single step. Is it as expected？

  #10855 closed Jul 5, 2025

• Time series diffusion

  #10886 closed Jul 5, 2025

• train_dreambooth_lora_flux crash on batch size greater than 1

  #10994 closed Jul 5, 2025

• support SageAttention

  #11121 closed Jul 5, 2025

• How to multi-GPU WAN inference

  #11339 closed Jul 5, 2025

• convert_hunyuan_video_to_diffusers.py not working - invalid load key, '\xd8'.

  #11370 closed Jul 5, 2025

• Bug: set adapters when model is compiled

  #11766 closed Jul 5, 2025

• support for MeanFlow

  #11781 closed Jul 5, 2025

• [FR] Please support WanVACETransformer3DModel.from_single_file

  #11630 closed Jul 5, 2025

• set_lora_device reports KeyError even though lora is loaded

  #11833 closed Jul 4, 2025

• Conflix size when load controlnet+IP-Adapter when load stable-diffusion-xl-base-1.0

  #11853 closed Jul 3, 2025

• framepack pipeline return pure black output when using torch.float16

  #11830 closed Jul 3, 2025

• Single File and GGUF support of Cosmos-Predict2

  #11798 closed Jul 1, 2025

• flux_control support flux kontext dev

  #11817 closed Jul 1, 2025

• set_adapters performance degrades with the number of inactive adapters

  #11816 closed Jun 30, 2025

11 Issues opened by 11 people

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

  #11872 opened Jul 6, 2025

• Question about control sd3

  #11867 opened Jul 5, 2025

• [Docs] Visualizing the codebase for diffusers

  #11866 opened Jul 5, 2025

• AttributeError: type object 'CosmosTransformer3DModel' has no attribute 'from_single_file'

  #11865 opened Jul 5, 2025

• AutoencoderDC.encode fails with torch.compile(fullgraph=True) - "name 'torch' is not defined"

  #11864 opened Jul 5, 2025

• ControlnetUnion not working correctly

  #11861 opened Jul 4, 2025

• STORK Scheduler

  #11859 opened Jul 4, 2025

• Can not load fusionx_lora into original wan2.1-14b

  #11849 opened Jul 2, 2025

• Finetuning DreamBoot Inpainting

  #11846 opened Jul 1, 2025

• finetune flux kontext on custom dataset

  #11840 opened Jul 1, 2025

• Try load gguf for FluxKontextPipeline but invalid input size

  #11839 opened Jul 1, 2025

33 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.

• The Modular Diffusers

  #9672 commented on Jul 6, 2025 • 66 new comments

• Add SkyReels V2: Infinite-Length Film Generative Model

  #11518 commented on Jul 5, 2025 • 23 new comments

• First Block Cache

  #11180 commented on Jul 3, 2025 • 2 new comments

• Tensor-parallel like FeedForward to lower memory requirements

  #10623 commented on Jul 2, 2025 • 0 new comments

• Support for control-lora

  #10686 commented on Jul 5, 2025 • 0 new comments

• Add gradient checkpointing support for AutoencoderKLWan

  #11105 commented on Jul 5, 2025 • 0 new comments

• Add VidTok AutoEncoders

  #11261 commented on Jul 2, 2025 • 0 new comments

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

  #11690 commented on Jul 3, 2025 • 0 new comments

• Add MAGI-1: Autoregressive Video Generation at Scale

  #11713 commented on Jul 6, 2025 • 0 new comments

• [tests] add compile + offload tests for GGUF.

  #11740 commented on Jul 2, 2025 • 0 new comments

• Avoid creating tensor in CosmosAttnProcessor2_0 (#11761)

  #11763 commented on Jul 5, 2025 • 0 new comments

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

  #11765 commented on Jun 30, 2025 • 0 new comments

• Fix unique memory address when doing group-offloading with disk

  #11767 commented on Jul 4, 2025 • 0 new comments

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

  #11775 commented on Jul 4, 2025 • 0 new comments

• Add --lora_alpha and metadata handling to train_dreambooth_lora_sd3.py

  #11783 commented on Jun 30, 2025 • 0 new comments

• [wip] Higgs quantization

  #11797 commented on Jul 2, 2025 • 0 new comments

• [Refactor] TensorRTStableDiffusionPipeline - Remove StableDiffusionReferencePipeline inheriting from DiffusionPipeline

  #11829 commented on Jun 30, 2025 • 0 new comments

• [Fix Issue #1197 since 2022] Support pre-trained openai/guided-diffusion (ADM) with minimal code change

  #6730 commented on Jul 4, 2025 • 0 new comments

• Error when I'm trying to train a Flux lora with train_dreambooth_lora_flux_advanced

  #11489 commented on Jul 6, 2025 • 0 new comments

• Error in init from pretrained for LTXConditionPipeline

  #11644 commented on Jul 6, 2025 • 0 new comments

• StableDiffusionXLControlNetUnionInpaintPipeline requires a float for controlnet_conditioning_scale & control_guidance_end

  #11828 commented on Jul 6, 2025 • 0 new comments

• `StableAudioDiTModel` throws `[rank0]: RuntimeError: DDP...` on `accelerator.prepare` when training on 4 GPUs

  #10160 commented on Jul 5, 2025 • 0 new comments

• Comments bug in FluxPriorReduxPipeline

  #10457 commented on Jul 5, 2025 • 0 new comments

• WanImageToVideoPipeline - swap out a limited number of blocks

  #10999 commented on Jul 5, 2025 • 0 new comments

• Multi-gpus Context Parallel training support?

  #11020 commented on Jul 5, 2025 • 0 new comments

• set_adapters not support for compiled model

  #11408 commented on Jul 5, 2025 • 0 new comments

• FlowMatchEulerDiscreteScheduler Timesteps have Floating Point Errors

  #11749 commented on Jul 5, 2025 • 0 new comments

• WanVACEPipeline: Passing a list of prompts is not yet supported. This may be supported in the future.

  #11821 commented on Jul 5, 2025 • 0 new comments

• Run FluxKontextPipeline with multi-images input, But An error occurred

  #11824 commented on Jul 4, 2025 • 0 new comments

• Dreambooth finetune FLUX dev CLIPTextModel

  #10925 commented on Jul 3, 2025 • 0 new comments

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

  #11700 commented on Jul 1, 2025 • 0 new comments

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

  #11790 commented on Jun 30, 2025 • 0 new comments

• CogvideoX VAE decoder consumes significantly more memory in the latest version

  #10091 commented on Jun 30, 2025 • 0 new comments