# Diffusers changelog

> State-of-the-art diffusion models for image and audio.

- Vendor: Hugging Face
- Category: AI
- Official site: https://huggingface.co/docs/diffusers
- Tracked by: What's New (https://whatsnew.fyi/product/diffusers)
- Harvested from: GitHub (huggingface/diffusers)
- Entries below: 10 (newest first)

What's New is an index, not a publisher: every entry below links to the vendor's own release notes, which are the authoritative source. Entries are labelled where they are hand-curated sample data, pre-releases, or drawn from a secondary source such as a developer blog.

## Releases

### v0.39.0 — Diffusers 0.39.0: New image and video pipelines, core library improvements, and more

- Date: 2026-07-03
- Version: v0.39.0
- Original notes: https://github.com/huggingface/diffusers/releases/tag/v0.39.0
- Permalink: https://whatsnew.fyi/product/diffusers/releases/v0.39.0

- **added** — Cosmos 3 pipeline, NVIDIA's unified world foundation model with Mixture-of-Transformers architecture for world generation, physical reasoning, and action generation, including video-to-video and action-conditioned generation
- **added** — Ideogram 4 pipeline, a flow-matching text-to-image model with multimodal text encoder and asymmetric classifier-free guidance
- **added** — Krea 2 pipeline, a flow-matching text-to-image model with single-stream MMDiT and Qwen3-VL text encoder, supporting both base and distilled checkpoints
- **added** — DreamLite pipeline for text-to-image and image-editing from ByteDance, featuring DreamLiteUNetModel and DreamLiteMobilePipeline for on-device generation
- **added** — PRXPixel pipeline for pixel-space text-to-image generation using a ~7B PRXTransformer2DModel without VAE
- **added** — Motif-Video pipeline, a 2B parameter diffusion transformer for text-to-video and image-to-video generation with three-stage architecture
- **added** — AnyFlow pipeline, the first any-step video diffusion framework supporting arbitrary inference budgets with bidirectional and causal variants

##### New Pipelines

###### Cosmos 3

[**Cosmos 3**](https://huggingface.co/docs/diffusers/main/api/pipelines/cosmos3) 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.

- PR: [https://github.com/huggingface/diffusers/pull/13818](https://github.com/huggingface/diffusers/pull/13818)
- Docs: [https://huggingface.co/docs/diffusers/main/api/pipelines/cosmos3](https://huggingface.co/docs/diffusers/main/api/pipelines/cosmos3)

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

###### Ideogram 4

[**Ideogram 4**](https://huggingface.co/docs/diffusers/main/api/pipelines/ideogram4) 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.

- PR: [https://github.com/huggingface/diffusers/pull/13859](https://github.com/huggingface/diffusers/pull/13859)
- Docs: [https://huggingface.co/docs/diffusers/main/api/pipelines/ideogram4](https://huggingface.co/docs/diffusers/main/api/pipelines/ideogram4)

Thanks to @JinLiIdeogram for the contribution.

###### Krea 2

[**Krea 2 (K2)**](https://huggingface.co/docs/diffusers/main/api/pipelines/krea2) 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.

- PR: [https://github.com/huggingface/diffusers/pull/14045](https://github.com/huggingface/diffusers/pull/14045)
- Docs: [https://huggingface.co/docs/diffusers/main/api/pipelines/krea2](https://huggingface.co/docs/diffusers/main/api/pipelines/krea2)

Thanks to @EleaZhong and @Abhinay1997 for the contribution.

###### DreamLite

[**DreamLite**](https://huggingface.co/docs/diffusers/main/api/pipelines/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.

- PR: [https://github.com/huggingface/diffusers/pull/13815](https://github.com/huggingface/diffusers/pull/13815)
- Docs: [https://huggingface.co/docs/diffusers/main/api/pipelines/dreamlite](https://huggingface.co/docs/diffusers/main/api/pipelines/dreamlite)

Thanks to @Carlofkl for the contribution.

###### PRX Pixel

[**PRXPixel**](https://huggingface.co/docs/diffusers/main/api/pipelines/prx_pixel) 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).

- PR: [https://github.com/huggingface/diffusers/pull/13928](https://github.com/huggingface/diffusers/pull/13928)
- Docs: [https://huggingface.co/d

_[Truncated at 4000 characters — full notes: https://github.com/huggingface/diffusers/releases/tag/v0.39.0]_

### v0.38.0 — Diffusers 0.38.0: New image and audio pipelines, Core library improvements, and more

- Date: 2026-05-01
- Version: v0.38.0
- Original notes: https://github.com/huggingface/diffusers/releases/tag/v0.38.0
- Permalink: https://whatsnew.fyi/product/diffusers/releases/v0.38.0

- **added** — Add LLaDA2 pipeline for discrete diffusion language models that generate text through block-wise iterative refinement
- **added** — Add NucleusMoE-Image pipeline, a 2B active 17B parameter sparse MoE model for image generation
- **added** — Add ERNIE-Image pipeline, an 8B parameter image generation model
- **added** — Add LongCat-AudioDiT pipeline for text-to-audio diffusion
- **added** — Add ACE-Step 1.5 pipeline for generating variable-length stereo audio from text prompts and optional lyrics
- **added** — Add support for Flux.2 small decoder model
- **added** — Add modular pipeline support for LTX-2 and Hunyuan 1.5
- **added** — Add Flash Attention 4 backend support
- **added** — Add FlashPack loading support
- **added** — Add group offloading with TorchAO support
- **added** — Add ring_anything as a new context parallelism backend
- **added** — Add pipeline profiling capabilities to Diffusers
- **added** — Add training configuration for Flux.2 series LoRA
- **fixed** — Fix Ulysses SP backward pass with SDPA
- **fixed** — Fix MotionConv2d to cast blur_kernel to input dtype instead of reverse
- **fixed** — Ensure consistent dtype and eval mode in pipeline save and load tests
- **fixed** — Fix single file path validation logic
- **fixed** — Validate eta is in [0, 1] range in DDIMPipeline
- **fixed** — Fix Dynamo lru_cache warnings during torch.compile

##### New Pipelines

###### LLaDA2

[**LLaDA2**](https://huggingface.co/collections/inclusionAI/llada21) is a family of discrete diffusion language models that generate text through block-wise iterative refinement. Instead of autoregressive token-by-token generation, LLaDA2 starts with a fully masked sequence and progressively unmasks tokens by confidence over multiple refinement steps. 

- PR: https://github.com/huggingface/diffusers/pull/13226
- Docs: [https://huggingface.co/docs/diffusers/main/api/pipelines/llada2](https://huggingface.co/docs/diffusers/main/api/pipelines/llada2)

###### Nucleus-MoE

NucleusMoE-Image is a 2B active 17B parameter model trained with efficiency at its core. Our novel architecture highlights the scalability of a sparse MoE architecture for Image generation. 

- PR: https://github.com/huggingface/diffusers/pull/13317
- Docs: [https://huggingface.co/docs/diffusers/main/api/pipelines/nucleusmoe_image](https://huggingface.co/docs/diffusers/main/api/pipelines/nucleusmoe_image)

Thanks to @sippycoder for the contribution.

###### Ernie-Image

ERNIE-Image is a powerful and highly efficient image generation model with 8B parameters. 

- PR: https://github.com/huggingface/diffusers/pull/13432
- Docs: [https://huggingface.co/docs/diffusers/main/api/pipelines/ernie_image](https://huggingface.co/docs/diffusers/main/api/pipelines/ernie_image)

Thanks to @HsiaWinter for the contribution.

###### LongCat-AudioDiT

LongCat-AudioDiT is a text-to-audio diffusion model from Meituan LongCat. 

- PR: https://github.com/huggingface/diffusers/pull/13483
- Docs: [https://huggingface.co/docs/diffusers/main/api/pipelines/longcat_audio_dit](https://huggingface.co/docs/diffusers/main/api/pipelines/longcat_audio_dit)

Thanks to @RuixiangMa for the contribution.

###### Ace-Step 1.5

ACE-Step 1.5 generates variable-length stereo audio at 48 kHz (10 seconds to 10 minutes) from text prompts and optional lyrics. The full system pairs a Language Model planner with a Diffusion Transformer (DiT) synthesizer; this pipeline wraps the DiT half of that stack, and consists of three components: an [**AutoencoderOobleck**](https://huggingface.co/docs/diffusers/main/en/api/models/autoencoder_oobleck#diffusers.AutoencoderOobleck) VAE that compresses waveforms into 25 Hz stereo latents, a Qwen3-based text encoder for prompt and lyric conditioning, and an [**AceStepTransformer1DModel**](https://huggingface.co/docs/diffusers/main/en/api/models/ace_step_transformer#diffusers.AceStepTransformer1DModel) DiT that operates in the VAE latent space using flow matching.

- PR: [https://github.com/huggingface/diffusers/pull/13095](https://github.com/huggingface/diffusers/pull/13095)
- Docs: [https://huggingface.co/docs/diffusers/main/api/pipelines/ace_step](https://huggingface.co/docs/diffusers/main/api/pipelines/ace_step)

Thanks to @[ChuxiJ](https://github.com/ChuxiJ) for the contribution.

###### Flux.2 Small Decoder

Make your Flux.2 decoding faster with this new small decoder model from the Black Forest Labs. You can check it out [here](https://huggingface.co/black-forest-labs/FLUX.2-small-decoder). It was contributed by @huemin-art in [this PR](https://github.com/huggingface/diffusers/pull/13428).

##### Modular Pipeline Support

We added modular support for LTX-2 and Hunyuan 1.5. 

##### Core Library

- [Flash Attention 4 backend](https://github.com/huggingface/diffusers/issues/13280)
- [FlashPack loading](https://github.com/huggingface/diffusers/issues/12700)
- [Group offloading + TorchAO](https://github.com/huggingface/diffusers/pull/13276)
- [`ring_anything` as a new CP backend](https://github.com/huggingface/diffusers/pull/13545)
- [Profiling pipelines in Diffusers](https://github.com/huggingface/diffusers/pull/13356)

##### All commits

* [Discrete Diffusion] Add LLaDA2 pipeline  by @kashif in #13226
* [LLADA2] documentation fixes  by @kashif in #13333
* [ci] claude in ci.  b

_[Truncated at 4000 characters — full notes: https://github.com/huggingface/diffusers/releases/tag/v0.38.0]_

### v0.37.1 — Fixes for AutoModel type hints in Modular Pipelines and Flux Klein LoRA loading

- Date: 2026-03-25
- Version: v0.37.1
- Original notes: https://github.com/huggingface/diffusers/releases/tag/v0.37.1
- Permalink: https://whatsnew.fyi/product/diffusers/releases/v0.37.1

- **fixed** — Fix loading ModularPipelines with AutoModel type hints in their modular_model_index.json
- **fixed** — Fix Flux Klein LoRA loading
- **fixed** — Fix unguarded torchvision import in Cosmos Predict 2.5

-  Fix for loading `ModularPipelines` with `AutoModel` type hints in their `modular_model_index.json` #13271
-  Fix Flux Klein LoRA loading #13313
-  Fix unguarded `torchvision` import in Cosmos Predict 2.5 #13321

### v0.37.0 — Diffusers 0.37.0: Modular Diffusers, New image and video pipelines, multiple core library improvements, and more 🔥

- Date: 2026-03-05
- Version: v0.37.0
- Original notes: https://github.com/huggingface/diffusers/releases/tag/v0.37.0
- Permalink: https://whatsnew.fyi/product/diffusers/releases/v0.37.0

- **added** — Modular Diffusers introduces a new way to build diffusion pipelines by composing reusable blocks instead of writing entire pipelines from scratch
- **added** — Z Image Omni Base pipeline for image generation with good quality, robust generative diversity, broad stylistic coverage, and precise prompt adherence
- **added** — Flux2 Klein pipeline for unified generation and editing in a compact architecture with state-of-the-art quality and end-to-end inference under a second
- **added** — Qwen Image Layered model capable of decomposing an image into multiple RGBA layers for independent layer manipulation
- **added** — FIBO Edit image-to-image model that operates on JSON inputs paired with source images for deterministic and repeatable editing workflows
- **added** — Cosmos Predict2.5 model specialized for simulating and predicting the future state of the world
- **added** — Cosmos Transfer2.5 conditional world generation model with adaptive multimodal control supporting edges, blurred video, segmentation maps, and depth maps
- **added** — GLM-Image generation model with hybrid autoregressive and diffusion decoder architecture for visual fidelity and fine-grained details
- **added** — RAE (Representation Autoencoders) as an alternative to traditional VAEs for latent-space diffusion models of image generation
- **added** — LTX-2 audio-conditioned text-to-video generation model that can generate videos with synced audio, supporting full and distilled model inference and two-stage inference with spatial sampling
- **added** — Helios 14B video generation model that runs at 17 FPS on a single NVIDIA H100 GPU and supports minute-scale generation
- **added** — MagCache caching method for improved performance
- **added** — TaylorSeer caching method for improved performance
- **added** — Unified Sequence Parallel attention context-parallelism backend
- **added** — Ulysses Anything Attention context-parallelism backend

##### Modular Diffusers

Modular Diffusers introduces a new way to build diffusion pipelines by composing reusable blocks. Instead of writing entire pipelines from scratch, you can now mix and match building blocks to create custom workflows tailored to your specific needs! This complements the existing `DiffusionPipeline` class, providing a more flexible way to create custom diffusion pipelines.

Find more details on how to get started with Modular Diffusers [here](https://huggingface.co/docs/diffusers/en/modular_diffusers/quickstart), and also check out the [announcement post](https://huggingface.co/blog/modular-diffusers).

##### New Pipelines and Models

###### Image 🌆

- [Z Image Omni Base](https://huggingface.co/docs/diffusers/en/api/pipelines/z_image): Z-Image is the foundation model of the Z-Image family, engineered for good quality, robust generative diversity, broad stylistic coverage, and precise prompt adherence. While Z-Image-Turbo is built for speed, Z-Image is a full-capacity, undistilled transformer designed to be the backbone for creators, researchers, and developers who require the highest level of creative freedom. Thanks to @RuoyiDufor for contributing this in #12857.
- [Flux2 Klein](https://huggingface.co/docs/diffusers/main/en/api/pipelines/flux2#diffusers.Flux2KleinPipeline):FLUX.2 [Klein] unifies generation and editing in a single compact architecture, delivering state-of-the-art quality with end-to-end inference in as low as under a second. Built for applications that require real-time image generation without sacrificing quality, and runs on consumer hardware, with as little as 13GB VRAM.
- [Qwen Image Layered](https://huggingface.co/Qwen/Qwen-Image-Layered): Qwen-Image-Layered is a model capable of decomposing an image into multiple RGBA layers. This layered representation unlocks inherent editability: each layer can be independently manipulated without affecting other content. Thanks to @naykun for contributing this in #12853.
- [FIBO Edit](https://huggingface.co/docs/diffusers/main/en/api/pipelines/bria_fibo_edit): Fibo Edit is an 8B parameter image-to-image model that introduces a new paradigm of structured control, operating on JSON inputs paired with source images to enable deterministic and repeatable editing workflows. Featuring native masking for granular precision, it moves beyond simple prompt-based diffusion to offer explicit, interpretable control optimized for production environments. Its lightweight architecture is designed for deep customization, empowering researchers to build specialized “Edit” models for domain-specific tasks while delivering top-tier aesthetic quality. Thanks galbria for contributing it in [https://github.com/huggingface/diffusers/pull/12930](https://github.com/huggingface/diffusers/pull/12930).
- [Cosmos Predict2.5](https://huggingface.co/docs/diffusers/main/en/api/pipelines/cosmos): Cosmos-Predict2.5, the latest version of the Cosmos World Foundation Models (WFMs) family, specialized for simulating and predicting the future state of the world. Thanks to @miguelmartin75 for contributing it in #12852.
- [Cosmos Transfer2.5](https://huggingface.co/docs/diffusers/main/en/api/pipelines/cosmos): Cosmos-Transfer2.5 is a conditional world generation model with adaptive multimodal control, that produces high-quality world simulations conditioned on multiple control inputs. These inputs can take different modalities—including edges, blurred video, segmentation maps, and depth maps. Thanks to @miguelmartin75 for contributing it in #13066.
- [GLM-Image](https://huggingface.co/docs/diffusers/main/en/api/pipelines/glm_image): GLM-Image is an image generation model adopts a hybrid autoregressive + diffusion decoder architecture, effectively pushing the upper bound of visual fidelity and fine-grained details. In general image generation quality, it aligns with industry-standard LDM-based approaches, while demonstrating significant advantages in knowledge-intensi

_[Truncated at 4000 characters — full notes: https://github.com/huggingface/diffusers/releases/tag/v0.37.0]_

### v0.36.0 — Diffusers 0.36.0: Pipelines galore, new caching method, training scripts, and more 🎄

- Date: 2025-12-08
- Version: v0.36.0
- Original notes: https://github.com/huggingface/diffusers/releases/tag/v0.36.0
- Permalink: https://whatsnew.fyi/product/diffusers/releases/v0.36.0

- **added** — Flux2 pipeline for image generation and editing that accepts multiple input images as reference
- **added** — Z-Image pipeline for image generation in the 6B parameter regime
- **added** — QwenImage Edit Plus pipeline that supports both generation and editing with multiple input images as references
- **added** — Bria FIBO pipeline trained on structured JSON captions up to 1,000+ words for precise visual parameter control
- **added** — Kandinsky Image Lite pipeline as a lightweight image generation model with 6B parameters
- **added** — ChronoEdit pipeline that reframes image editing as a video generation task for temporal consistency
- **added** — Sana-Video pipeline for fast and efficient video generation with linear attention for long video sequences
- **added** — Kandinsky 5.0 T2V Lite pipeline as a lightweight video generation model with 2B parameters
- **added** — HunyuanVideo-1.5 pipeline for state-of-the-art video generation with 8.3 billion parameters
- **added** — Wan-Animate pipeline for character animation and replacement in videos based on reference character images and driving motion
- **added** — Flash Attention 3 and Flash Attention 3 varlen kernels-powered attention backends
- **added** — Flash Attention 2 and Flash Attention 2 varlen kernels-powered attention backends
- **added** — SAGE kernels-powered attention backend
- **added** — TaylorSeer cache support delivering up to 3x speedups with negligible-to-none quality compromise
- **added** — Flux.2 LoRA fine-tuning training script with optimizations for consumer GPUs

The release features a number of new image and video pipelines, a new caching method, a new training script, new `kernels` - powered attention backends, and more. It is quite packed with a lot of new stuff, so make sure you read the release notes fully 🚀

##### New image pipelines

- [Flux2](https://huggingface.co/docs/diffusers/main/en/api/pipelines/flux2): Flux2 is the latest generation of image generation and editing model from Black Forest Labs. It’s capable of taking multiple input images as reference, making it versatile for different use cases.
- [Z-Image](https://huggingface.co/docs/diffusers/main/en/api/pipelines/z_image): Z-Image is a best-of-its-kind image generation model in the 6B param regime. Thanks to @JerryWu-code in [https://github.com/huggingface/diffusers/pull/12703](https://github.com/huggingface/diffusers/pull/12703).
- [QwenImage Edit Plus](https://huggingface.co/docs/diffusers/main/en/api/pipelines/qwenimage): It’s an upgrade of QwenImage Edit and is capable of taking multiple input images as references. It can act as both a generation and an editing model. Thanks to @naykun for contributing in https://github.com/huggingface/diffusers/issues/12357.
- [Bria FIBO:](https://huggingface.co/docs/diffusers/main/en/api/pipelines/bria_fibo) FIBO is trained on structured JSON captions up to 1,000+ words and designed to understand and control different visual parameters such as lighting, composition, color, and camera settings, enabling precise and reproducible outputs. Thanks to @galbria for contributing this in [https://github.com/huggingface/diffusers/pull/12545](https://github.com/huggingface/diffusers/pull/12545).
- [Kandinsky Image Lite](https://huggingface.co/docs/diffusers/main/en/api/pipelines/kandinsky5_image): Kandinsky 5.0 Image Lite is a lightweight image generation model (6B parameters). Thanks to @leffff for contributing this in [https://github.com/huggingface/diffusers/pull/12664](https://github.com/huggingface/diffusers/pull/12664).
- [ChronoEdit](https://huggingface.co/docs/diffusers/main/en/api/pipelines/chronoedit): ChronoEdit reframes image editing as a video generation task, using input and edited images as start/end frames to leverage pretrained video models with temporal consistency. A temporal reasoning stage introduces reasoning tokens to ensure physically plausible edits and visualize the editing trajectory. Thanks to @zhangjiewu for contributing this in https://github.com/huggingface/diffusers/pull/12593.

##### New video pipelines

- [Sana-Video](https://huggingface.co/docs/diffusers/main/en/api/pipelines/sana_video): Sana-Video is a fast and efficient video generation model, equipped to handle long video sequences, thanks to its incorporation of linear attention. Thanks to @lawrence-cj for contributing this in [https://github.com/huggingface/diffusers/pull/12634](https://github.com/huggingface/diffusers/pull/12634).
- [Kandinsky 5](https://huggingface.co/docs/diffusers/main/en/api/pipelines/kandinsky5_video): Kandinsky 5.0 T2V Lite is a lightweight video generation model (2B parameters) that ranks #1 among open-source models in its class. It outperforms larger models and offers the best understanding of Russian concepts in the open-source ecosystem. Thanks to @leffff for contributing this in [https://github.com/huggingface/diffusers/pull/12478](https://github.com/huggingface/diffusers/pull/12478).
- [Hunyuan 1.5](https://huggingface.co/docs/diffusers/main/en/api/pipelines/hunyuan_video15): HunyuanVideo-1.5 is a lightweight yet powerful video generation model that achieves state-of-the-art visual quality and motion coherence with only 8.3 billion parameters, enabling efficient inference on consumer-grade GPUs.
- [Wan Animate](https://huggingface.co/docs/diffusers/main/en/api/pipelines/wan#wan-animate-unified-character-animation-and-replacement-with-holistic-replication): Wan-Animate is a state-of-the-art character animation and replacement video model based on Wan2.1.

_[Truncated at 4000 characters — full notes: https://github.com/huggingface/diffusers/releases/tag/v0.36.0]_

### v0.35.2 — 🐞 fixes for `transformers` models, imports,

- Date: 2025-10-15
- Version: v0.35.2
- Original notes: https://github.com/huggingface/diffusers/releases/tag/v0.35.2
- Permalink: https://whatsnew.fyi/product/diffusers/releases/v0.35.2

##### All commits

* Release: v0.35.1-patch by @sayakpaul (direct commit on v0.35.2-patch)
* handle offload_state_dict when initing transformers models  by @sayakpaul in #12438
* [CI] Fix TRANSFORMERS_FLAX_WEIGHTS_NAME import issue  by @DN6 in #12354
* Fix PyTorch 2.3.1 compatibility: add version guard for torch.library.…  by @Aishwarya0811 in #12206
* fix scale_shift_factor being on cpu for wan and ltx  by @vladmandic in #12347
* Release: v0.35.2-patch by @sayakpaul (direct commit on v0.35.2-patch)

### v0.35.1 — v0.35.1 for improvements in Qwen-Image Edit

- Date: 2025-08-20
- Version: v0.35.1
- Original notes: https://github.com/huggingface/diffusers/releases/tag/v0.35.1
- Permalink: https://whatsnew.fyi/product/diffusers/releases/v0.35.1

Thanks to @naykun for the following PRs that improve Qwen-Image Edit:

* https://github.com/huggingface/diffusers/pull/12188
* https://github.com/huggingface/diffusers/pull/12190

### v0.35.0 — Diffusers 0.35.0: Qwen Image pipelines, Flux Kontext, Wan 2.2, and more

- Date: 2025-08-19
- Version: v0.35.0
- Original notes: https://github.com/huggingface/diffusers/releases/tag/v0.35.0
- Permalink: https://whatsnew.fyi/product/diffusers/releases/v0.35.0

This release comes packed with new image generation and editing pipelines, a new video pipeline, new training scripts, quality-of-life improvements, and much more. Read the rest of the release notes fully to not miss out on the fun stuff.

##### New pipelines 🧨

We welcomed new pipelines in this release:

- Wan 2.2
- Flux-Kontext
- Qwen-Image
- Qwen-Image-Edit

###### Wan 2.2 📹

This update to Wan provides significant improvements in video fidelity, prompt adherence, and style. Please check out the [official doc](https://huggingface.co/docs/diffusers/main/en/api/pipelines/wan) to learn more.

###### Flux-Kontext 🎇

Flux-Kontext is a 12-billion-parameter rectified flow transformer capable of editing images based on text instructions. Please check out the [official doc](https://huggingface.co/docs/diffusers/main/en/api/pipelines/flux#kontext) to learn more about it.

###### Qwen-Image 🌅

After a successful run of delivering language models and vision-language models, the Qwen team is back with an image generation model, which is Apache-2.0 licensed! It achieves significant advances in complex text rendering and precise image editing. To learn more about this powerful model, refer to our [docs](https://huggingface.co/docs/diffusers/main/en/api/pipelines/qwenimage).

Thanks to @naykun for contributing both Qwen-Image and Qwen-Image-Edit via [this PR](https://github.com/huggingface/diffusers//issues/12055) and [this PR](https://github.com/huggingface/diffusers/pull/12164/).

##### New training scripts 🎛️

Make these newly added models your own with our training scripts:

- [Kontext trainer](https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/README_flux.md#training-kontext)
- [Qwen-Image trainer](https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/README_flux.md#training-kontext)

##### Single-file modeling implementations

Following the 🤗 Transformers’ [philosophy](https://huggingface.co/blog/transformers-design-philosophy) of single-file modeling implementations, we have started implementing modeling code in single and self-contained files. The Flux Transformer [code](https://github.com/huggingface/diffusers/blob/baa9b582f348e52aa2fc245e366611f454e1082b/src/diffusers/models/transformers/transformer_flux.py) is one example of this.

##### Attention refactor

We have massively refactored how we do attention in the models. This allows us to provide support for different attention backends (such as PyTorch native `scaled_dot_product_attention`, Flash Attention 3, SAGE attention, etc.) in the library seamlessly. 

Having attention supported this way also allows us to integrate different parallelization mechanisms, which we’re actively working on. Follow [this PR](https://github.com/huggingface/diffusers/pull/11941) if you’re interested.

Users shouldn’t be affected at all by these changes. Please open an issue if you face any problems.

##### Regional compilation

Regional compilation trims cold-start latency by only compiling the small and frequently-repeated block(s) of a model - typically a transformer layer - and enables reusing compiled artifacts for every subsequent occurrence. For many diffusion architectures, this delivers the same runtime speedups as full-graph compilation and reduces compile time by 8–10x. Refer to [this doc](https://huggingface.co/docs/diffusers/main/en/optimization/fp16#regional-compilation) to learn more.

Thanks to @anijain2305 for contributing this feature in [this PR](https://github.com/huggingface/diffusers/pull/11705).

We have also authored a number of posts that center around the use of `torch.compile`. You can check them out at the links below:

- [Presenting Flux Fast: Making Flux go brrr on H100s](https://pytorch.org/blog/presenting-flux-fast-making-flux-go-brrr-on-h100s/)
- [torch.compile and Diffusers: A Hands-On Guide to Peak Performance](https://pytorch.org/blog/torch-compile-and-diffusers-a-hand

_[Truncated at 4000 characters — full notes: https://github.com/huggingface/diffusers/releases/tag/v0.35.0]_

### v0.34.0 — Diffusers 0.34.0: New Image and Video Models, Better torch.compile Support, and more

- Date: 2025-06-24
- Version: v0.34.0
- Original notes: https://github.com/huggingface/diffusers/releases/tag/v0.34.0
- Permalink: https://whatsnew.fyi/product/diffusers/releases/v0.34.0

##### 📹 New video generation pipelines

###### Wan VACE

Wan VACE supports various generation techniques which achieve controllable video generation. It comes in two variants: a 1.3B model for fast iteration & prototyping, and a 14B for high quality generation. Some of the capabilities include:

- Control to Video (Depth, Pose, Sketch, Flow, Grayscale, Scribble, Layout, Boundary Box, etc.). Recommended library for preprocessing videos to obtain control videos: [**huggingface/controlnet_aux**](https://huggingface.co/docs/diffusers/main/en/api/pipelines/wan)
- Image/Video to Video (first frame, last frame, starting clip, ending clip, random clips)
- Inpainting and Outpainting
- Subject to Video (faces, object, characters, etc.)
- Composition to Video (reference anything, animate anything, swap anything, expand anything, move anything, etc.)

The code snippets available in [**this**](https://github.com/huggingface/diffusers/pull/11582) pull request demonstrate some examples of how videos can be generated with controllability signals.

Check out the [docs](https://huggingface.co/docs/diffusers/main/en/api/pipelines/wan#any-to-video-controllable-generation) to learn more.

###### Cosmos Predict2 Video2World

Cosmos-Predict2 is a key branch of the [Cosmos World Foundation Models](https://www.nvidia.com/en-us/ai/cosmos) (WFMs) ecosystem for Physical AI, specializing in future state prediction through advanced world modeling. It offers two powerful capabilities: text-to-image generation for creating high-quality images from text descriptions, and video-to-world generation for producing visual simulations from video inputs.

The Video2World model comes in a 2B and 14B variant. Check out the [docs](https://huggingface.co/docs/diffusers/main/en/api/pipelines/cosmos) to learn more.

###### LTX 0.9.7 and Distilled

LTX 0.9.7 and its distilled variants are the latest in the family of models released by Lightricks.

Check out the [docs](https://huggingface.co/docs/diffusers/en/api/pipelines/ltx_video) to learn more.

###### Hunyuan Video Framepack and F1

[Framepack](https://github.com/lllyasviel/FramePack) is a novel method for enabling long video generation. There are two released variants of Hunyuan Video trained using this technique. Check out the [docs](https://huggingface.co/docs/diffusers/main/en/api/pipelines/framepack#framepack) to learn more.

###### FusionX

The [FusionX family of models and LoRAs](https://huggingface.co/vrgamedevgirl84/Wan14BT2VFusioniX), built on top of Wan2.1-14B, should already be supported. To load the model, use `from_single_file()`:

```python
from diffusers import WanTransformer3DModel

transformer = WanTransformer3DModel.from_single_file(
    "https://huggingface.co/vrgamedevgirl84/Wan14BT2VFusioniX/blob/main/Wan14Bi2vFusioniX_fp16.safetensors",
    torch_dtype=torch.bfloat16
)
```

To load the LoRAs, use `load_lora_weights()`:

```python
pipe = DiffusionPipeline.from_pretrained(
    "Wan-AI/Wan2.1-T2V-14B-Diffusers",
    torch_dtype=torch.bfloat16
).to("cuda")
pipe.load_lora_weights(
    "vrgamedevgirl84/Wan14BT2VFusioniX", weight_name="FusionX_LoRa/Wan2.1_T2V_14B_FusionX_LoRA.safetensors"
)
```

###### AccVideo and CausVid (only LoRAs)

[AccVideo](https://github.com/aejion/AccVideo) and [CausVid](https://github.com/tianweiy/CausVid) are two novel distillation techniques that speed up the generation time of video diffusion models while preserving quality. Diffusers supports loading their extracted LoRAs with their respective models.

##### 🌠 New image generation pipelines

###### Cosmos Predict2 Text2Image

Text-to-image models from the Cosmos-Predict2 release. The models comes in a 2B and 14B variant. Check out the [docs](https://huggingface.co/docs/diffusers/main/en/api/pipelines/cosmos) to learn more.

###### Chroma

Chroma is a **8.9B** parameter model based on **FLUX.1-schnell.** It’s fully **Apache 2.0 licensed**, ensuring that

_[Truncated at 4000 characters — full notes: https://github.com/huggingface/diffusers/releases/tag/v0.34.0]_

### v0.33.1 — v0.33.1: fix ftfy import

- Date: 2025-04-10
- Version: v0.33.1
- Original notes: https://github.com/huggingface/diffusers/releases/tag/v0.33.1
- Permalink: https://whatsnew.fyi/product/diffusers/releases/v0.33.1

##### All commits
* fix ftfy import for wan pipelines by @yiyixuxu  in #11262