36 Pull requests merged by 20 people

• [docs] minor updates to bitsandbytes docs.

  #11509 merged May 6, 2025

• [lora_conversion] Enhance key handling for OneTrainer components in LORA conversion utility (#11441)

  #11487 merged May 6, 2025

• enable lora cases on XPU

  #11506 merged May 6, 2025

• Hunyuan Video Framepack

  #11428 merged May 6, 2025

• [LoRA] use removeprefix to preserve sanity.

  #11493 merged May 6, 2025

• update dep table.

  #11504 merged May 6, 2025

• Update setup.py to pin min version of peft

  #11502 merged May 6, 2025

• Fix torchao docs typo for fp8 granular quantization

  #11473 merged May 6, 2025

• Update training script for txt to img sdxl with lora supp with new interpolation.

  #11496 merged May 5, 2025

• Set LANCZOS as the default interpolation method for image resizing.

  #11492 merged May 5, 2025

• [training] feat: enable quantization for hidream lora training.

  #11494 merged May 5, 2025

• [train_dreambooth_lora_lumina2] Add LANCZOS as the default interpolation mode for image resizing

  #11491 merged May 5, 2025

• [train_dreambooth_lora_sdxl] Add --image_interpolation_mode option for image resizing (default to lanczos)

  #11490 merged May 5, 2025

• [train_text_to_image_sdxl]Add LANCZOS as default interpolation mode for image resizing

  #11455 merged May 5, 2025

• [Feature] Implement tiled VAE encoding/decoding for Wan model.

  #11414 merged May 5, 2025

• enable semantic diffusion and stable diffusion panorama cases on XPU

  #11459 merged May 5, 2025

• [train_dreambooth_lora_flux_advanced] Add LANCZOS as the default interpolation mode for image resizing

  #11472 merged May 2, 2025

• [train_dreambooth_lora_sdxl_advanced] Add LANCZOS as the default interpolation mode for image resizing

  #11471 merged May 2, 2025

• [docs] Adapters

  #11331 merged May 2, 2025

• [docs] Memory optims

  #11385 merged May 1, 2025

• cache packages_distributions

  #11453 merged May 1, 2025

• [tests] xfail recent pipeline tests for specific methods.

  #11469 merged May 1, 2025

• Fix typos in docs and comments

  #11416 merged May 1, 2025

• [WAN] fix recompilation issues

  #11475 merged May 1, 2025

• [modular diffusers] introducing ModularLoader

  #11462 merged Apr 30, 2025

• make autoencoders. controlnet_flux and wan_transformer3d_single_file pass on xpu

  #11461 merged Apr 30, 2025

• Add LANCZOS as default interplotation mode.

  #11463 merged Apr 30, 2025

• [test_models_transformer_hunyuan_video] help us test torch.compile() for impactful models

  #11431 merged Apr 30, 2025

• make safe diffusion test cases pass on XPU and A100

  #11458 merged Apr 30, 2025

• Add StableDiffusion3InstructPix2PixPipeline

  #11378 merged Apr 30, 2025

• Add generic support for Intel Gaudi accelerator (hpu device)

  #11328 merged Apr 30, 2025

• enable unidiffuser test cases on xpu

  #11444 merged Apr 30, 2025

• enable consistency test cases on XPU, all passed

  #11446 merged Apr 30, 2025

• torch.compile fullgraph compatibility for Hunyuan Video

  #11457 merged Apr 30, 2025

• enable marigold_intrinsics cases on XPU

  #11445 merged Apr 30, 2025

• Raise warning instead of error for block offloading with streams

  #11425 merged Apr 30, 2025

16 Pull requests opened by 11 people

• [LoRA] make lora alpha and dropout configurable

  #11467 opened Apr 30, 2025

• Fix typos in strings and comments

  #11476 opened May 1, 2025

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

  #11477 opened May 1, 2025

• [tests] Enable testing for HiDream transformer

  #11478 opened May 1, 2025

• Update Intel Gaudi doc

  #11479 opened May 1, 2025

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

  #11481 opened May 2, 2025

• [Modular diffusers] more refactors

  #11484 opened May 2, 2025

• [feat] Add ostris / Flex.2-preview pipeline

  #11485 opened May 3, 2025

• Fix huggingface-hub package name.

  #11498 opened May 5, 2025

• clean up the __Init__ for stable_diffusion

  #11500 opened May 5, 2025

• Change timestep device to cpu for xla

  #11501 opened May 5, 2025

• enable 7 cases on XPU

  #11503 opened May 6, 2025

• docs: fix invalid links

  #11505 opened May 6, 2025

• enable print_env on xpu

  #11507 opened May 6, 2025

• [tests] Changes to the `torch.compile()` CI and tests

  #11508 opened May 6, 2025

• Add FluxPAGPipeline with support for PAG

  #11510 opened May 6, 2025

15 Issues closed by 9 people

• Unable to load Flux LoRA trained with OneTrainer – NotImplementedError in `_convert_mixture_state_dict_to_diffusers`

  #11441 closed May 6, 2025

• Unable to load FLUX.1-Canny-dev-lora into FluxControlPipeline

  #11464 closed May 6, 2025

• TypeError: BnB4BitDiffusersQuantizer.create_quantized_param() got an unexpected keyword argument 'dtype'

  #11149 closed May 6, 2025

• [quant] about float8_e4m3_tensor

  #11468 closed May 6, 2025

• Failed to import diffusers.schedulers.scheduling_ddim because of the following error (look up to see its traceback): numpy.dtype size changed, may indicate binary incompatibility. Expected 96 from C header, got 88 from PyObject

  #11169 closed May 5, 2025

• Better image interpolation in training scripts follow up

  #11390 closed May 5, 2025

• The module 'HunyuanVideoTransformer3DModel' has been loaded in `bitsandbytes` 8bit and moving it to cpu via `.to()` is not supported. Module is still on cuda:0

  #10653 closed May 4, 2025

• Is there a way to generate a single image using multiple GPUs?

  #11108 closed May 2, 2025

• StableDiffusion3Pipeline RuntimeError: c10::Half != c10::BFloat16

  #11117 closed May 2, 2025

• IndexError: index 999 is out of bounds for dimension 0 with size 51

  #11474 closed May 2, 2025

• Can't load MistoLine rank_256 ControlNet with from_single_file

  #11178 closed Apr 30, 2025

• a

  #11177 closed Apr 30, 2025

• what is the difference between from diffusion import *** and from diffusers import ***?

  #11272 closed Apr 30, 2025

• Wan2.1 Out of Memory for WanI2V-720P-14B model

  #11289 closed Apr 30, 2025

• Why Wan i2v video processor always float32 datatype?

  #11351 closed Apr 30, 2025

9 Issues opened by 9 people

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

  #11499 opened May 5, 2025

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

  #11497 opened May 5, 2025

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

  #11489 opened May 4, 2025

• Sincerely Request The Support for Flux PAG Pipeline

  #11488 opened May 4, 2025

• enable_xformers_memory_efficient_attention in the training script

  #11486 opened May 3, 2025

• [New Project] Phantom: Subject-Consistent Video Generation via Cross-Modal Alignment

  #11480 opened May 2, 2025

• Train_controlnet_sdxl.py and tensorboard log images

  #11470 opened Apr 30, 2025

• Finetuning of flux or scratch training

  #11466 opened Apr 30, 2025

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

  #11460 opened Apr 30, 2025

48 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.

• [Feature] AutoModel can load components using model_index.json

  #11401 commented on May 6, 2025 • 19 new comments

• [LoRA] parse metadata from LoRA and save metadata

  #11324 commented on May 5, 2025 • 17 new comments

• [Quantization] Add TRT-ModelOpt as a Backend

  #11173 commented on May 6, 2025 • 17 new comments

• Add VisualCloze

  #11377 commented on May 6, 2025 • 10 new comments

• Cosmos

  #10660 commented on May 5, 2025 • 8 new comments

• First Block Cache

  #11180 commented on May 5, 2025 • 7 new comments

• Attention Dispatcher

  #11368 commented on May 5, 2025 • 2 new comments

• Add basic implementation of AuraFlowImg2ImgPipeline

  #11340 commented on May 2, 2025 • 2 new comments

• Pipeline deprecations

  #11354 commented on May 6, 2025 • 1 new comment

• AttributeError: _hf_hook caused by delattr in hooks.remove_hook_from_module()

  #10729 commented on May 6, 2025 • 0 new comments

• TypeError: UVit2DModel.forward.<locals>.layer_() got an unexpected keyword argument 'encoder_hidden_states'

  #11214 commented on May 6, 2025 • 0 new comments

• Stable Virtual Camera

  #11151 commented on May 6, 2025 • 0 new comments

• The Modular Diffusers

  #9672 commented on May 1, 2025 • 0 new comments

• feat: pipeline-level quantization config

  #11130 commented on May 6, 2025 • 0 new comments

• Added unload_lora_weights to StableDiffusionPipeline

  #11172 commented on May 5, 2025 • 0 new comments

• Add SDXL implementation for PyTorch/XLA training

  #11204 commented on May 4, 2025 • 0 new comments

• [modular diffusers] more refactor

  #11235 commented on May 2, 2025 • 0 new comments

• [Feature] MultiControlNet support for SD3Impainting

  #11251 commented on May 2, 2025 • 0 new comments

• Add VidTok AutoEncoders

  #11261 commented on May 3, 2025 • 0 new comments

• [LoRA] improve LoRA fusion tests

  #11274 commented on May 3, 2025 • 0 new comments

• [Tests] Enable more general testing for `torch.compile()` with LoRA hotswapping

  #11322 commented on May 3, 2025 • 0 new comments

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

  #11355 commented on May 3, 2025 • 0 new comments

• [LoRA training] add aspect ratio bucketing

  #11438 commented on May 5, 2025 • 0 new comments

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

  #11452 commented on May 3, 2025 • 0 new comments

• Controlnet Inference example, CUDA OOM

  #11363 commented on Apr 30, 2025 • 0 new comments

• diffusers中的cogvideo image2video中建议的分辨率尺寸与cogvideo原项目中建议的尺寸不一致？

  #11336 commented on Apr 30, 2025 • 0 new comments

• How to convert the hidream lora trained by diffusers to a format that comfyui can load?

  #11396 commented on Apr 30, 2025 • 0 new comments

• scheduler UniPCMultistepScheduler get_velocity()

  #11213 commented on Apr 30, 2025 • 0 new comments

• How to use attention_mask and encoder_attention_mask or apply prompts to specific areas in the image?

  #11176 commented on Apr 30, 2025 • 0 new comments

• Sage Attention for diffuser library

  #11168 commented on Apr 30, 2025 • 0 new comments

• partial models spec in model_index.json

  #10413 commented on Apr 30, 2025 • 0 new comments

• Support custom_pipeline argument for Diffuser T2V Community Pipelines

  #11183 commented on May 1, 2025 • 0 new comments

• [Feature request] LTX-Video v0.9.6 15x faster inference than non-distilled model.

  #11359 commented on May 1, 2025 • 0 new comments

• `.from_pretrained` `torch_dtype="auto"` argument not working a expected

  #11432 commented on May 1, 2025 • 0 new comments

• [core] introduce auto classes for T2V and I2V

  #10486 commented on May 2, 2025 • 0 new comments

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

  #11374 commented on May 2, 2025 • 0 new comments

• FlaxUNet2DConditionModel is not initialized with correct dtypes

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

• Another comfy-ui compatible FLUX1.dev LoRA fails to load

  #10954 commented on May 2, 2025 • 0 new comments

• Failed to import diffusers.pipelines.onnx_utils

  #11199 commented on May 3, 2025 • 0 new comments

• enable_attention_slicing give NaN results for SDXL on MPS

  #11229 commented on May 3, 2025 • 0 new comments

• MultiControlNetModel is not supported for SD3ControlNetInpaintingPipeline

  #11208 commented on May 4, 2025 • 0 new comments

• controlnet pipeline and training script for sana model

  #11019 commented on May 4, 2025 • 0 new comments

• [tests] help us test `torch.compile()` for impactful models

  #11430 commented on May 5, 2025 • 0 new comments

• Error in loading Lora

  #11209 commented on May 5, 2025 • 0 new comments

• Getting CUDA out of memory error even with Colab A100 high RAM

  #11014 commented on May 5, 2025 • 0 new comments

• Hotswapping multiple LoRAs throws a peft key error.

  #11298 commented on May 5, 2025 • 0 new comments

• Mochi OutOfMemory

  #11454 commented on May 6, 2025 • 0 new comments

• RuntimeError: group_norm in VAE decode for SDXL Masked Img2Img (even with ControlNets disabled & FP32 VAE/Latents)

  #11424 commented on May 6, 2025 • 0 new comments