Diffusers 0.39.0

Original release notes

New Pipelines

Cosmos 3

Cosmos 3 is NVIDIA's unified world foundation model (WFM) for Physical AI - a single omni-model built on a Mixture-of-Transformers (MoT) architecture that combines world generation, physical reasoning, and action generation, replacing the separate Predict, Reason, and Transfer models from earlier Cosmos releases. A single Cosmos3OmniTransformer runs a Qwen-style language model in parallel with a diffusion generation pathway, joined by a 3D multimodal RoPE. This release also lands video-to-video and action-conditioned generation, and a sound encoder.

Thanks to @atharvajoshi10, @yzhautouskay, and @MaciejBalaNV for the contributions.

Ideogram 4

Ideogram 4 is a flow-matching text-to-image model that uses a multimodal text encoder and an asymmetric classifier-free guidance scheme: a dedicated unconditional_transformer produces the negative branch with zeroed text features, while the main transformer consumes the full packed text + image sequence. The pipeline ships with structured prompt upsampling and LoRA loading support.

Thanks to @JinLiIdeogram for the contribution.

Krea 2

Krea 2 (K2) is a flow-matching text-to-image model built around a single-stream MMDiT with grouped-query attention. A Qwen3-VL text encoder provides the conditioning - hidden states from twelve decoder layers are tapped per token and fused inside the transformer by a small text-fusion stage - and images are decoded with the Qwen-Image VAE. Both the base (midtrain) and TDM (distilled, few-step) checkpoints are supported, alongside a LoRA DreamBooth trainer.

Thanks to @EleaZhong and @Abhinay1997 for the contribution.

DreamLite

DreamLite is a text-to-image and image-editing model from ByteDance. It pairs a custom 2D U-Net (DreamLiteUNetModel) with the Qwen3-VL multimodal encoder as its prompt / image-instruction encoder, and uses an AutoencoderTiny (TAESD-style) VAE for fast latent encode/decode. A distilled DreamLiteMobilePipeline targets on-device, low-latency generation.

Thanks to @Carlofkl for the contribution.

PRX Pixel

PRXPixel is a pixel-space text-to-image generation model by Photoroom. A ~7B PRXTransformer2DModel denoises raw RGB images directly - no VAE is needed. The model is conditioned on a Qwen3-VL text encoder and uses flow matching where the transformer predicts the clean image at each step (x-prediction).

Thanks to @DavidBert for the contribution.

Motif-Video

Motif-Video is a 2B parameter diffusion transformer for text-to-video and image-to-video generation. It features a three-stage architecture (12 dual-stream + 16 single-stream + 8 DDT decoder layers), Shared Cross-Attention for stable text-video alignment over long sequences, a T5Gemma2 text encoder, and rectified flow matching for velocity prediction.

Thanks to @waitingcheung for the contribution.

AnyFlow

AnyFlow from NVIDIA, NUS, and MIT is the first any-step video diffusion framework built on flow maps, enabling a single model (bidirectional or causal) to adapt to arbitrary inference budgets. It ships both bidirectional and FAR causal pipelines built on Wan2.1 backbones, covering text-to-video, image-to-video, and video-to-video.

Thanks to @Enderfga for the contribution.

JoyAI-Image-Edit

JoyAI-Image is a unified multimodal foundation model for image understanding, text-to-image generation, and instruction-guided image editing. It combines an 8B Multimodal LLM with a 16B Multimodal Diffusion Transformer (MMDiT). JoyImageEditPipeline supports general image editing as well as spatial editing capabilities including object move, object rotation, and camera control.

Thanks to @Moran232 for the contribution.

DiffusionGemma

DiffusionGemma is a block-diffusion encoder-decoder language model. A causal encoder reads the clean prompt (and any previously generated blocks) into a KV cache, and a bidirectional decoder denoises a fixed-size "canvas" of tokens by cross-attending to that cache, committing the most confident tokens via the new BlockRefinementScheduler. The released checkpoint is google/diffusiongemma-26B-A4B-it.

Anima

Anima is a 2 billion parameter text-to-image model created via a collaboration between CircleStone Labs and Comfy Org. It is focused mainly on anime concepts, characters, and styles, but is also capable of generating a wide variety of other non-photorealistic content.

It reuses the CosmosTransformer3DModel with a Qwen3 text encoder, a T5-token text conditioner, and the AutoencoderKLQwenImage VAE.

Thanks to @rmatif for the contribution.

LTX-2.X IC LoRA and HDR Pipelines

New LTX2InContextPipeline (in-context LoRA) and LTX2HDRPipeline extend the LTX-2 family with in-context conditioning and HDR video generation.

Modular Pipeline Support

We added a modular pipeline for Stable Diffusion 3 (SD3) in https://github.com/huggingface/diffusers/pull/13324 (thanks to @AlanPonnachan).
We added a modular pipeline for Anima in https://github.com/huggingface/diffusers/pull/13732 (thanks to @rmatif).
LoRA loading is now enabled on ErnieImageModularPipeline (#13948) and Ideogram4ModularPipeline (#13980), thanks to @SamuelTallet.

Core Library

All commits

[CI] Update all workflows with permissions by @DN6 in #13672
[agents docs] update models.md with class attributes and attention mask by @yiyixuxu in #13665
Fix ignored generator in FlowMatchEulerDiscreteScheduler by @RobbinMarcus in #13678
[core] remove txt_seq_lens from qwen transformer. by @sayakpaul in #13674
[tests] fix lora tests involving clip. by @sayakpaul in #13675
post release 0.38.0 by @sayakpaul in #13670
Fix NameError in ZImageOmniPipeline when guidance_scale=0 by @Ricardo-M-L in #13527
Enable TorchAO int4wo quantization tests on XPU by @jiqing-feng in #13537
[CI] QOL improvement for PR size labeler by @DN6 in #13554
Fix BucketBatchSampler cache alignment in DreamBooth scripts by @azolotenkov in #13353
chore: update pr_labeler.yml by @hf-security-analysis[bot] in #13685
Address ernie-image review findings #13577 by @akshan-main in #13663
feat: Add Modular Pipeline for Stable Diffusion 3 (SD3) by @AlanPonnachan in #13324
Update attention_backends.md to update FA3 minimum support to Ampere by @sayakpaul in #13283
[CI] Bump style-bot SHA + switch to GitHub App by @paulinebm in #13690
[feat] JoyAI-JoyImage-Edit support by @Moran232 in #13444
Add LoRA support for Cosmos Predict 2.5 and fix pipeline to match official Cosmos repo by @terarachang in #13664
Eliminate GPU sync overhead and CPUGPU transfers across LTX2 pipeline by @ViktoriiaRomanova in #13564
Gate deep imports from torch.distributed by @hlky in #13673
Bump diffusers from 0.20.1 to 0.38.0 in /examples/research_projects/realfill by @dependabot[bot] in #13692
Reduce WanAnimate TorchAO test input sizes to prevent OOM by @jiqing-feng in #13541
add SP support for flash_varlen_hub backend by @zhtmike in #13479
[ci] allow claude to open PRs for certain instructions. by @sayakpaul in #13536
[ci] remove compel. by @sayakpaul in #13715
styling fix. by @sayakpaul (direct commit on v0.39.0-release)
better usage of UV_PRERELEASE=allow by @sayakpaul in #13716
[docs] add magcache to caching api listing by @sayakpaul in #13714
[tests] refactor autoencoderkl tests by @sayakpaul in #13368
[docs] add docs for JoyAI-Image-Edit by @feice-huang in #13726
[tests] add attention backend tests. by @sayakpaul in #13174
Install transformers from main for doc and staging by @sayakpaul in #13723
Update Flax removal version by @DN6 in #13729
examples/dreambooth: fix LR scheduler step count for multi-GPU in train_dreambooth_lora_sd3.py by @Dev-X25874 in #13731
Serge reviewer by @sayakpaul in #13735
[ci] switch to a more unique name by @sayakpaul in #13738
fix autoencoder memory tests by @sayakpaul in #13734
Fix GGUF to Work Better with modules_to_not_convert / keep_in_fp32_modules by @dg845 in #13697
[tests] refactor ltx2 autoencoder tests to use latest mixins by @sayakpaul in #13739
feat: Add Motif-Video model and pipelines by @waitingcheung in #13551
Update contribution guidelines by @DN6 in #13753
[agents] add a section on tests in the ai skill and integration guides. by @sayakpaul in #13752
Add LTX-2.X IC LoRA and HDR Pipelines by @dg845 in #13572
[tests] Fix controlnet tests by @sayakpaul in #13736
[tests] fix bitsandbytes compile tests for flux. by @sayakpaul in #13750
[core] minimum torch version is 2.6 by @sayakpaul in #13725
[tests] fix lora checkpoint serialization issues by @sayakpaul in #13676
fix(randn_tensor): compare device.type, not torch.device, when suppressing MPS info log by @Ricardo-M-L in #13508
[LLADA2] Fix llada2 review #13598 by @kashif in #13698
fix lfs pointer rejection problems for hub tests by @sayakpaul in #13733
Fix training gradient underflow in quantization tests by @jiqing-feng in #13539
examples/dreambooth: fix missing weighting chunk when using prior preservation in Flux and SD3 LoRA training by @Dev-X25874 in #13743