GeezerVison

MiniMax-H3-Image

MiniMax-H3-Image supports text-to-image generation and editing with one or more reference images. The integration includes a dedicated image transformer and autoencoder, with a fixed six-step sampling schedule and distilled guidance. A separate prompt-enhancement workflow expands short requests and suggests an aspect ratio.

Load it through ModularPipeline.from_pretrained(...), then call load_components(...). The MiniMax-H3-Image documentation covers reference-image prompts, output sizes, prompt enhancement, and component offloading.

Pika H3 Flex

Pika H3 Flex builds on MiniMax-H3 to generate video and synchronized stereo audio in six steps. It supports text prompts, first and/or last frames, and multimodal references through separate modular workflows. The integration includes LoRA adapter loading and a guide to sharing already-loaded MiniMax-H3 components between the two pipelines.

See the Pika H3 Flex documentation for workflow selection, adapter loading, and component reuse.

P-H3-Image

P-H3-Image from Pruna AI brings text-to-image generation and instruction-based editing with up to eight input images. Its modular pipeline includes a prompt upsampler that turns requests into detailed generation or editing prompts, an image autoencoder, and a text connector. The checkpoint uses eight sampling steps with distilled guidance.

See the P-H3-Image documentation for image sizing, reference-image editing, prompt upsampling, and offloading. Thanks to @nifleisch for the integration in #15171.

Kandinsky S2V 1.0

Kandinsky S2V 1.0 animates a person from a reference image and a speech recording, including full-body movement aligned with the speech. It generates video in five-second chunks, carrying FramePack memory between chunks. The supplied speech provides the soundtrack; the model generates the video.

The modular integration supports Lite and Pro checkpoints, including distilled variants that use PiflowScheduler with a default of 16 steps. See the Kandinsky S2V documentation for audio preparation, checkpoint settings, and memory requirements. Thanks to @nvsm for the integration in #15164.

PAN 2

PAN 2 simulates scenes from prompts describing actions and camera movements over time. This release integrates its 29B generative simulator through text-to-video and image-to-video modular workflows. Context parallelism and FirstBlockCache, MagCache, and TaylorSeer caching are supported. Prompt and video safety checks are enabled by default and require the separate pan2-guardrail package.

See the PAN 2 documentation for loading, guardrails, and distributed inference. The integration covers the simulator; the separate reasoning model that writes its prompts is linked from the model documentation. Thanks to @wu-qing-157 for the integration in #15202.

Other highlights

• MiniMax-H3 tensor parallelism: shard the transformer across devices, including its large modulation projections. Replicated parameters also retain their intended dtype during tensor-parallel loading. See: #14609, #15147.
• AMD attention support: the new aiter_sage_hub backend brings a Hub-hosted SageAttention kernel to RDNA4 gfx1201 GPUs, with native SDPA fallback for unsupported input shapes. See #15060.
• Attention backend fixes: choosing a backend on one model no longer changes the backend of unrelated models. Individual attention modules now check requirements and download Hub kernels when selecting a backend, and the FlashAttention-2 integration works with newer Hub kernel versions. See #14335, #15000.
• LoRA fixes: component-weight dictionaries now broadcast correctly across multiple adapters, and Kohya text-encoder LoRAs retain their per-module alpha values. Adapter-weight documentation and type hints also clarify supported formats. See: #15128, #15157.
• Checkpoint saving: save_pretrained() preserves other variants' shards and removes obsolete indexes when replacing a sharded checkpoint with a single-file checkpoint. See #14844.
• Modular caching: stateful cache hooks reset after each pipeline call, including calls that raise an exception, so subsequent generations start with fresh cache state. See #15212.
• Training examples: fix gradient clipping in the text-to-image LoRA and SDXL LCM LoRA scripts, update FLUX quantized embedding preparation for current Transformers, and remove expired deprecation blocks that crashed training with --revision. See: #15184, #15144, #15118.
• Documentation: reorganized distributed-inference guides, updated AWS Trainium examples for native TorchNeuron, and corrected Cosmos3 FP8 checkpoint locations. See: #15172, #15138, #15236.

Deprecations and compatibility

• Dependency checks: availability checks now enforce supported dependency versions. An installed package below the supported minimum is treated as unavailable, with an upgrade message. For example, the existing minimums of torch>=2.6 and transformers>=4.41.2 are now enforced by these checks. (#15148)
• Relocated model components: latent upscalers and condition embedders move under diffusers.models. Existing checkpoints remain loadable, while old pipeline-module imports are deprecated for removal in 1.0.0. Use public imports such as from diffusers import LTX2TextConnectors. See: #15111.
• Modular input defaults: explicitly passing None preserves that value. Omitted inputs receive the default of the first block that reads them. An InputParam declared both required and with a non-None default is treated as optional and emits a warning. See: #15135, #14521.
• Training behavior: --max_grad_norm now takes effect in the affected LoRA examples. Runs with gradients above the clipping threshold can therefore behave differently from earlier versions. See: #15184.
• Component loading: repository configuration no longer causes arbitrary component-library imports. Custom component modules outside the supported library set must already be imported by the caller, and modular custom-block loading checks remote-code trust before processing requirements. See: 3076605.

Thanks to everyone who contributed code, documentation, reviews, and issue reports!

All commits

• Remove expired non_ema_revision deprecation blocks from training examples by @A-Raphie in #15118.
• Extend CLI compile and attention-backend tests to other devices by @sywangyi in #15039.
• Fix shared LoRA component weight broadcasting by @gorkadabo1 in #15128.
• [agent docs] prefer named arguments over nested dicts by @yiyixuxu in #15129.
• when an InputParam is both required and has a default: set required = False and warn by @yiyixuxu in #14521.
• Restrict repository-specified component imports and check custom-block trust before requirements by @LinZiyuu, with @DN6, in 3076605.
• Fix the Marigold intrinsics integration test by @sywangyi in #15086.
• [core] move latent upscalers and condition embedders out of pipelines by @sayakpaul in #15111.
• [CI] Use HF Trusted Publisher OIDC tokens by @DN6 in #15106.
• [docs] AWS Trainium: rewrite guide for native TorchNeuron support by @JingyaHuang in #15138.
• Add aiter_sage_hub attention backend for AMD RDNA4 (gfx1201) by @Shan2L in #15060.
• fix: scope save_pretrained shard cleanup to the saved variant (avoid deleting another variant's shards) by @AloysJehwin in #14844.
• Pika h3 flex by @yiyixuxu in #15150.
• [docs] Pika H3 Flex: add a section on reusing minimax h3 components by @yiyixuxu in #15153.
• [compon...

Contributors

This is a security release. See GHSA-4prv-5hj7-c2xr for details.

This patch release fixes how a component's declared library name is resolved when loading with from_pretrained.

Pipelines and models declare each component as a ["<library>", "<class>"] pair in model_index.json / config.json. This release changes how that library name is resolved across the DiffusionPipeline, AutoModel, and ModularPipeline loading paths, and adds a clearer error when the declared class is not found in the resolved module.

All users on 0.42.0 are encouraged to upgrade:

pip install diffusers==0.42.1

FLUX.3 Image Klein

FLUX.3 Image Klein from Black Forest Labs supports text-to-image generation and editing with one or more reference images. Both the base and step-distilled checkpoints are available through ModularPipeline, with an integrated prompt upsampler that turns short requests into detailed captions and supports layout instructions with bounding boxes. The integration also includes single-file transformer loading and BFL FP8r and NVFP4 checkpoints.

Load the pipeline with ModularPipeline.from_pretrained(...), then load its weights with load_components(...). See the FLUX.3 documentation for examples and guidance on prompt upsampling, editing, and memory use.

Bria Fibo 2.0

Bria Fibo 2.0 supports image generation from structured JSON captions and editing with up to five reference images through BriaFibo2Pipeline. The integration includes the dev checkpoint and the edit-turbo checkpoint, which runs in four steps without classifier-free guidance. See the Fibo 2.0 documentation for the supported prompt formats.

LoRA loading and DreamBooth LoRA training examples are included for text-to-image and image editing. Thanks to @muhammadali-afridi for the model integration (#14991).

Kandinsky 6

Kandinsky 6 generates video with synchronized audio from text or a reference image through Kandinsky6TI2VAPipeline. A separate Kandinsky6SRPipeline provides video super-resolution. The integration supports Pro and Lite generation checkpoints, including distilled variants, and introduces PiflowScheduler for the distilled models. See the Kandinsky 6 documentation for checkpoint-specific settings and super-resolution requirements. Thanks to @Lefff for the model integration (#14949).

SANA-WM

SANA-WM generates video from a starting image, a text prompt, and a camera trajectory. Control the camera with explicit poses or a WASD/IJKL action string, then optionally refine the output with the separate LTX-2 refiner. See the SANA-WM documentation for the two-stage workflow.

Thanks to @lawrence-cj for the integration (#13881).

RFD4-Proteína

RFD4-Proteína from the Baker Lab and NVIDIA brings biomolecular design to Diffusers. It jointly generates protein sequences and all-atom structures while conditioning on molecular context such as binding targets, motifs, or ligands. The integration includes the pipeline, transformer, autoencoder, schedulers, and LoRA loading. Preparing model inputs and writing CIF files uses the separate rfd4-proteina-prep package; see the RFD4-Proteína documentation.

Thanks to @kierandidi for the integration (#15088).

Other highlights

• Qwen-Image 2.1: new modular pipelines and tensor parallelism support. (#14976, #14865)
• More hardware support: TorchTPU integration covers eager execution, torch.compile, and tensor parallelism. Tensor parallelism also gains Intel XPU support. (#14039, #15013)
• LoKr adapters: load LyCORIS Kronecker-product adapters for Z-Image and FLUX.2, including Klein. (#14163)
• LTX-2 motion control: motion_speed controls the positional time base independently of playback frame rate. Use a compatible adapter for speeds far from the default, such as the LTX-2.5 slow-motion control LoRA. (#14967)
• KVAE-Audio: AutoencoderKLKVAEAudio adds encoding and decoding for 48 kHz audio. (#14384)
• Caching: MagCache now supports FLUX.2. (#14756)
• Adapter fixes: Z-Image honors the LoRA scale in joint_attention_kwargs, Qwen-Image applies .alpha scaling to lora_A/lora_B checkpoints, and text encoders correctly report active adapters. (#15015, #15016, #14961)
• Pipeline fixes: preserve explicit timesteps in FlowMatchEulerDiscreteScheduler, fix Anima text encoder exports and standalone core denoising, and preserve the three-value return from InpaintProcessor.preprocess when no mask is supplied. (#14956, #14943, #14481)

Deprecations and compatibility

• Python 3.10 support is dropped. Upgrade to Python 3.11 or newer before installing this release. (#15030)
• Modular latent outputs: FLUX, Krea 2, and LTX-Video now unpack latents in their core denoise groups. Custom workflows should pass unpacked latents to their decode blocks. The fallback for packed latents is deprecated for removal in 0.47.0. (#14737)
• Hub compatibility: the supported huggingface_hub range is now >=1.32.0,<3.0. (#14863)
• TorchAO imports: having a TorchAO version older than 0.15 installed no longer breaks model imports. TorchAO quantization still requires version 0.15 or newer. (#15095)

Thanks to everyone who contributed to this release!

All commits

• fix #14461 by @yiyixuxu in #14956
• Kandinsky6 by @sayakpaul in #14949
• Bump doc-builder workflow pin by @paulinebm in #14957
• [docs] Jobs guide by @stevhliu in #14926
• [docs] Community methods by @stevhliu in #14973
• Fix active adapter detection for text encoders by @Subodh-17 in #14961
• fix(image_processor): maintain 3-tuple return contract in InpaintProcessor.preprocess when mask is None by @patrickswedish in #14481
• [ci] consolidate labeling on PRs. by @sayakpaul in #14911
• [LoRA] add LoKr adapter support (Z-Image, Flux2/Klein) by @christopher5106 in #14163
• [TPU] TorchTPU backend integration - eager / torch.compile / tp by @JingyaHuang in #14039
• [cli] detect composite block classes in custom_blocks by @yiyixuxu in #15001
• Fix Anima text encoder exports and standalone core denoising by @lucasruan1618 in #14943
• Bump huggingface_hub upper bound to <3.0 (diffusers) by @Wauplin in #14863
• Make Stable Diffusion device-map comparisons deterministic by @sywangyi in #15010
• xpu supports TP as well by @sywangyi in #15013
• Add SANA-WM camera-controlled image-to-video pipeline by @lawrence-cj in #13881
• [Z-Image] Apply the LoRA scale passed via joint_attention_kwargs by @Sergasgr in #15015
• [utils] preserve method metadata in apply_forward_hook by @javierdejesusda in #13803
• Update FlashAttention Hub integration and guard CUDA-only tests by @sywangyi in #15024
• [hooks] Support Flux2 in MagCache by @rootonchair in #14756
• 🔒 Pin GitHub Actions to commit SHAs by @paulinebm in #15028
• [CI] Clean up test_modeling_common.py by @DN6 in #14988
• ci: move to python 3.11. by [@...

Contributors

Qwen-Image 2.1

Qwen-Image 2.1 unifies image generation and editing in one model. Its visual generation component has 7B parameters, and it supports multiple reference images and native RGBA output for transparent images. Check out the docs for more details.

Thanks to @naykun for the model integration (#14804).

Other highlights

• Tensor-parallel checkpoint loading: each rank reads its own slice of sharded weights directly, reducing the memory needed during loading. (#14544)
• LTX-2.5 DFR: new pipelines support generation with keyframe slots and composable spatial and temporal refinement. (#14567)
• Cosmos 3: SeaCache support and mixed W8A8/W8A16 denoising for ModelOpt FP8 checkpoints. (#14663, #14664)
• Modular Diffusers: Wan 2.2 VACE gains modular blocks, and loading from existing local paths avoids unnecessary weight downloads. (#14432, #14797)
• More single-file loaders: support for Krea 2 and MiniMax-H3. (#14914, #14839)
• Tests: We removed the monolithic test_pipelines_common.py test suite to establish more modularity and transparent implementation of tests. Refer to #14669 for more details.

Deprecations and compatibility

🚨 ONNX support is deprecated; use Optimum for ONNX Runtime workflows. (#14658)

Previously deprecated APIs have been removed, including the LuminaText2ImgPipeline and Lumina2Text2ImgPipeline aliases. Use LuminaPipeline and Lumina2Pipeline instead. Older integrations should review the removed code paths before upgrading.

Thanks to everyone who contributed to this release!

All commits

• [docs] Clean up links by @stevhliu in #14535
• docs: fix 14 typos in docs, comments and docstrings by @iridescentWen in #14541
• [docs] LTX-2.5 Modular Pipeline Docs by @dg845 in #14450
• [agent docs] add lora test instructions. by @sayakpaul in #14549
• refactor bria pipeline tests to the new mixin structure by @akshan-main in #14547
• feat: improve issue reminder workflow and introduce conditional autoclose by @sayakpaul in #14477
• [tests] AOT compilation failure fixes by @sayakpaul in #14553
• [tests] skip hunyuan framepack group offloading tests by @sayakpaul in #14551
• [tests] remove autoencodertestermixin and rename new one. by @sayakpaul in #14555
• tests: fix CUDA OOM in VidTok slicing/tiling tests by @sayakpaul in #14554
• refactor bria fibo edit pipeline tests to the new mixin structure by @akshan-main in #14577
• refactor bria fibo pipeline tests to the new mixin structure by @akshan-main in #14570
• refactor deepfloyd if pipeline tests to the new mixin structure by @akshan-main in #14580
• Deprecate 'fp16_safetensors' command in CLI documentation by @sayakpaul in #14583
• Fix duplicated words and a misspelling in messages and docstrings by @Sreekant13 in #14287
• Fibo Edit: multi-reference conditioning and batching by @ShivamShrirao in #14566
• [tests] Add Config to Exclude Modules from Leaf-Level Group Offloading by @dg845 in #14564
• [tests] refactor flux control lora tests by @sayakpaul in #14557
• [tests] migrate test_pipelines.py to use pytest. by @sayakpaul in #14222
• [core] pass diffusers library name to kernels calls by @sayakpaul in #14441
• [tests] hunyuan, ltx, ltx2 by @sayakpaul in #14587
• make examples in docs device agnostic by @kaixuanliu in #14414
• Move LTX2Guidance Guider to guiders/ Folder by @dg845 in #14558
• refactor deepfloyd if img2img pipeline tests to the new mixin structure by @akshan-main in #14596
• refactor deepfloyd if inpainting pipeline tests to the new mixin structure by @akshan-main in #14602
• refactor deepfloyd if superresolution pipeline tests to the new mixin structure by @akshan-main in #14603
• refactor deepfloyd if img2img superresolution pipeline tests to the new mixin structure by @akshan-main in #14604
• docs: correct the enable_gqa / cuDNN explanation by @yiyixuxu in #14561
• refactor deepfloyd if inpainting superresolution pipeline tests to the new mixin structure by @akshan-main in #14606
• [tests] migrate acestep, auraflow, cogview4, and helios tests by @sayakpaul in #14582
• remove dead deepfloyd if shared test mixin by @akshan-main in #14611
• [agent docs]add reporting guide by @yiyixuxu in #14531
• migrate cogvideox lora tests to the pipeline-level mixins by @akshan-main in #14615
• [tests] migrate flux2 lora. by @sayakpaul in #14610
• migrate wan vace lora tests to the pipeline-level mixins by @akshan-main in #14614
• post release 0.40.0 by @sayakpaul in #14545
• migrate wan lora tests to the pipeline-level mixins by @akshan-main in #14613
• tests: add ideogram4 tests for pipelines. by @sayakpaul in #14599
• [tests] fix remote decoding tests and disable encode tests. by @sayakpaul in #14601
• [tests] migrate remaining flux pipeline tests to use new mixins by @sayakpaul in #14560
• [tests] refactor zimage and krea2 pipeline tests by @sayakpaul in #14563
• [tests] refactor sdxl tests. by @sayakpaul in #14619
• [tests] remove quanto tests. by @sayakpaul in #14254
• tests: migrate tests from tests/lora by @sayakpaul in #14559
• ci: fix invocation for compile tests. by @sayakpaul in #14552
• Fix Ideogram 4 Callback Handling and Tests by @dg845 in #14621
• Fix callback tensor inputs that are never bound in the denoising loop by @LuShadowX in #14416
• [CI] Repair broken links by @stevhliu in #14627
• migrate ideogram4 lora tests to the pipeline-level mixins by @akshan-main in #14628
• refactor lumina2 pipeline tests to the new mixin structure by @akshan-main in #14629
• [tests] refactor d series pipeline tests. by @sayakpaul in #14630
• [tests] refactor a* pipeline tests. by @sayakpaul in #14622
• [tests] refactor b* pipeline tests by @sayakpaul in #14625
• [tests] refactor e and g series pipeline tests by @sayakpaul in #14631
• [tests] refactor h and i series pipelines by @sayakpaul in #14633
• [tests] refactor j series pipeline tests by @sayakpaul in #14632
• [Download] Resolve the revision once at the beginning of from_pretrained by @Wauplin in #14340
• Add LTX-2.5 DFR pipeline (keyframe slots, spatial detailing, tiled temporal rounds) by @alexanderar in #14567
• skip these case for no-cuda device since the modeling contains cuda s… by @sywangyi in #14656
• [tests] refactor remaining s series pipelines by @sayakpaul in #14654
• [tests] refactor k-series pipeline tests by @sayakpaul in #14636
• [tests] refactor v-series pipeline tests by @sayakpaul in #14653
• [tests] refactor m and n series pipeline tests by @sayakpaul in #14638
• [tests] refactor l-series pipeline tests by @sayakpaul in #14637
• [tests] refactor c* pipeline tests. by @sayakpaul in #14626
• [tests] refactor o and p series pipeline tests by @sayakpaul in #14646
• [tests] refactor ltx2 dfr pipeline tests by @sayakpaul in #14655
• [core] deprecate onnx by @sayakpaul in #14658
• docs: correct stated defaults in Chroma pipeline docstrings by @iridescentWen in #14578
• [tests] fix SVD tests by @sayakpaul in #14687
• [tests] Split Pipeline Group Offloading Block-Level and Leaf-Level Tests by @dg845 in #14635
• [tests] fix sana bf16 tests on ci. by @sayakpaul in #14688
• Don't materialise the PRX attention mask by @DavidBert in #14677
• [Cosmos 3] Fix Cosmos 3 Group Offloading Tests by @dg845 in #14693
• Fix PriorTransformer Group Offloading Bug by @dg845 in #14695
• [train_text_to_image_decoder.py] Fix the LR scheduler when num_train_epochs is passed in a distributed training env by @shoutoutuoadi325 in #14542
• ModularPipeline.save_pretrained: change the default overwrite_modular_index to be True by @yiyixuxu in #14659
• add modular blockset for wan 2.2 vace by @akshan-main in #14432
• Invalidate HookRegistry child-registries cache on enable/disable cache by @SuryanshSS1011 in #14093
• Set scheduler begin index in remaining FlowMatch pipelines to avoid DtoH sync by @shoemoney in #14576
• tests: add lora tester mixin for modular and migrate minimax h3. by @sayakpaul in #14605
• agent doc: add a requirement for PR communications by @yiyixuxu in #14709
• [Community] Add CachedSearch script: cheaper best-of-N for video pipelines by @shreshthsaini in #14735
• [tests] partially fix the sd2 upscale tests by @sayakpaul in #14689
• [httpx migration] Import httpx from huggingface_hub by @Wauplin in #14753
• [docs] Improve navigation by @stevhliu in #14537
• Route non-interactive library messages through loggers by @wunianze666-netizen in #14682
• fix modular pipeline class resolution for families outside the auto-pipeline mappings by @akshan-main in #14736
• docs: fix two docstrings that...

Contributors

New Pipelines

MiniMax-H3

MiniMax-H3 generates video and its soundtrack together. A single transformer denoises one packed sequence containing the text conditioning, the conditioning media, and the target video and audio latents — there is no separate vocoder and no post-hoc audio pass. Its conditioner is a Qwen3VLForConditionalGeneration whose unnormalized 50th-decoder-layer hidden state is read instead of the last one.

MiniMax-H3 is integrated as Modular Diffusers blocks only — MiniMaxH3Blocks and their MiniMaxH3ModularPipeline are the whole integration. The conversion ships both checkpoint partitions in one repository and exposes three workflows (t2va, fl2va, ref2va) that can be pruned at from_pretrained time so only that task's components are declared and downloaded.

MiniMax Music 3

MiniMax Music 3 produces complete songs up to five minutes long from lyrics and a music description, with expressive vocals and long-range structure. It is a hybrid of an autoregressive and a diffusion stage: an 8B Qwen3-based global language model predicts one semantic audio token per frame while a small depth decoder fills in seven residual RVQ codebooks, and their fused hidden states condition a 2.4B flow-matching transformer that produces Flow-VAE latents in overlapping chunks. A DAC-style decoder turns the latents into 44.1 kHz stereo audio.

Stable Audio 3

Stable Audio 3 is a text-to-audio model from Stability AI that generates high-quality stereo audio at 44.1 kHz. It uses a rectified-flow DiT conditioned on a frozen T5Gemma text encoder (via cross-attention) and on duration (a float embedded by StableAudio3DurationEmbedder and used for adaptive layer norm), and decodes with the SAME (Semantically-Aligned Music Encoder) autoencoder, AutoencoderSAME.

Three pipelines ship: StableAudio3Pipeline, StableAudio3AudioToAudioPipeline, and StableAudio3InpaintPipeline.

Thanks to @buffett0323 for the contribution (#14119).

LTX-2.5

LTX-2.5 reuses the existing LTX2Pipeline / LTX2VideoTransformer3DModel / AutoencoderKLLTX2Video classes — there is no separate pipeline class. The user-visible difference is the text encoder: LTX-2.5 is paired with a Gemma 4 (gemma4_unified) checkpoint, loaded automatically from a converted LTX-2.5 repo.

Lightricks/LTX-2.5-Diffusers ships both the distilled DiT (transformer/) and the full/SFT DiT (transformer_full/), plus everything two-stage generation needs. Alongside the checkpoint support, this release adds:

• LTX2VideoDiffusionDecoderModel and LTX2VideoDiffusionDecodePipeline — a second video decoder over the same latent space, so latents are interchangeable between decoders.
• A duration_head that predicts shot length from the text-connector output, so num_frames is auto-predicted by default when the loaded pipeline has one.
• Prompt enhancement through a separate off-the-shelf google/gemma-4-E2B-it checkpoint (enable_prompt_enhancement=True).
• LTX25AutoBlocks for Modular Diffusers (#14453).

Wan-Animate-2

Wan-Animate-2 by the Alibaba Wan Team animates a reference character image with the motion of a driving video. The driving video is processed in fixed-length segments: each segment runs a reference-extraction pass that caches the driving segment's K/V in every transformer layer, denoises against that cache, and is decoded inside the loop, because the next segment conditions on the previous segment's decoded tail frames.

Two presets are available — the base checkpoint samples with classifier-free guidance, and the distilled checkpoint samples in few steps without it. Guidance is owned by the pipeline's guider component, so there is no guidance_scale argument.

Thanks to @kelseyee for authoring the integration (#14413).

JoyAI-Image-Edit-Plus

JoyAI-Image-Edit-Plus extends the JoyAI-Image family (an 8B MLLM paired with a 16B MMDiT) to multi-image instruction-guided editing. It accepts 1–5 reference images plus a text instruction and composes elements from the references into a new image.

Thanks to @tangyanf for the contribution (#14032).

Cosmos 3 follow-ups

Cosmos 3 landed in 0.39.0 and gets substantially more coverage in this release:

• A Cosmos 3 Modular pipeline (Cosmos3OmniModularPipeline), with Transfer support for precomputed control videos (edge, blur, depth, segmentation, world-scenario maps) generated autoregressively in chunks and stitched automatically.
• Edge support.
• Distilled (few-step) text-to-image and image-to-video via Cosmos3DistilledModularPipeline.
• Model parallelism.

Thanks to @yzhautouskay and @atharvajoshi10 for the contributions.

Modular Pipeline Support

Modular Diffusers is no longer marked experimental (#14525) — the API warning has been dropped.

• Three of this release's new integrations — MiniMax-H3, MiniMax Music 3, and Wan-Animate-2 — ship as Modular blocks only, with no DiffusionPipeline half.
• Cosmos 3 (#14110, #14150, #14177) and Krea 2 (#14083, thanks to @Cedric-Perauer) gained modular pipelines, and Anima gained img2img blocks (#13929, thanks to @PreethamNoelP).
• Branch-specific input defaults (#14234): when sibling blocks of a ConditionalPipelineBlocks declare different defaults for the same input, combine_inputs now merges the default to None and records the per-block defaults in a new InputParam.defaults_by_block field. get_block_state falls back to the block's own declared default, so each branch resolves its own default when it actually runs. Docstrings render conflicted defaults as e.g. "defaults to None or 189, depending on the workflow".
• Group offloading now works under auto offloading, and a group-offload device mismatch was fixed.
• Unused intermediate_inputs were cleaned up, stale auto-docstrings are now detected in CI, and Mellon custom-block required-input handling was fixed.

Core Library

Tensor parallelism

Tensor-parallel inference is now supported for model inference on CUDA and AWS Neuron (Trainium/Inferentia), exposed through the same public API already used for context parallelism:

from diffusers import TensorParallelConfig

pipe.transformer.enable_parallelism(config=TensorParallelConfig(mesh=tp_mesh))

Sharding is model-agnostic and driven from a flat _tp_plan, which has been added to the FLUX.1, FLUX.2 and Qwen-Image DiTs. Check out the docs for more details.

New quantization backends

• SDNQ — a training-free backend supporting int8 down to 2-bit, FP8 and other low-bit float formats, with optional SVD correction, Hadamard rotation, and quantized INT8/FP8 matmul. Runs on CUDA, ROCm, XPU, MPS, and CPU. SDNQ MiniMax-H3 loading landed too. Thanks to @asomoza.
• Nunchaku Lite for loading prequantized checkpoints, with fused-kernel performance documented. Thanks to @rootonchair.

Attention, kernels and offloading

• Kernels are downloaded only when users request them
• ROCm FlashAttention now uses the kernels-community/aiter-flash-attn-ck Hub kernel, dropping the aiter dependency. Thanks to @Abdennacer-Badaoui.
• [Sequence-parallel support for the `_f...

Contributors

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.

• PR: #13818
• Docs: https://huggingface.co/docs/diffusers/main/api/pipelines/cosmos3

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.

• PR: #13859
• Docs: https://huggingface.co/docs/diffusers/main/api/pipelines/ideogram4

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.

• PR: #14045
• Docs: https://huggingface.co/docs/diffusers/main/api/pipelines/krea2

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.

• PR: #13815
• Docs: https://huggingface.co/docs/diffusers/main/api/pipelines/dreamlite

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).

• PR: #13928
• Docs: https://huggingface.co/docs/diffusers/main/api/pipelines/prx_pixel

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.

• PR: #13551
• Docs: https://huggingface.co/docs/diffusers/main/api/pipelines/motif_video

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.

• PR: #13745
• Docs: https://huggingface.co/docs/diffusers/main/api/pipelines/anyflow

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.

• PR: #13444
• Docs: https://huggingface.co/docs/diffusers/main/api/pipelines/joyimage_edit

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.

• PR: #13986
• Docs: https://huggingface.co/docs/diffusers/main/api/pipelines/diffusion_gemma

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.

• PR: #13732
• Docs: https://huggingface.co/docs/diffusers/main/api/pipelines/anima

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.

• PR: #13572
• Docs: https://huggingface.co/docs/diffusers/main/api/pipelines/ltx2

Modular Pipeline Support

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

Core Library

• AutoRound quantization integration
• safetensors support in the TorchAO backend and _dequantize for the TorchAO quantizer
• BitsAndBytes quantization on MPS
• AutoPipelineForText2Audio
• AWS Neuron (Trainium/Inferentia) as an officially supported device with [torch.compile compatibility](#1348...

Contributors

New Pipelines

LLaDA2

LLaDA2 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: #13226
• Docs: 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: #13317
• Docs: 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: #13432
• Docs: 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: #13483
• Docs: 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 VAE that compresses waveforms into 25 Hz stereo latents, a Qwen3-based text encoder for prompt and lyric conditioning, and an AceStepTransformer1DModel DiT that operates in the VAE latent space using flow matching.

• PR: #13095
• Docs: https://huggingface.co/docs/diffusers/main/api/pipelines/ace_step

Thanks to @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. It was contributed by @huemin-art in this PR.

Modular Pipeline Support

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

Core Library

• Flash Attention 4 backend
• FlashPack loading
• Group offloading + TorchAO
• ring_anything as a new CP backend
• Profiling pipelines in Diffusers

All commits

• [Discrete Diffusion] Add LLaDA2 pipeline by @kashif in #13226
• [LLADA2] documentation fixes by @kashif in #13333
• [ci] claude in ci. by @sayakpaul in #13297
• [docs] kernels by @stevhliu in #13139
• [tests] Tests for conditional pipeline blocks by @sayakpaul in #13247
• avoid hardcode device in flux-control example by @kaixuanliu in #13336
• fix claude workflow to include id-token with write. by @sayakpaul in #13338
• Update LTX-2 Docs to Cover LTX-2.3 Models by @dg845 in #13337
• remove str option for quantization config in torchao by @howardzhang-cv in #13291
• [ci] include checkout step in claude review workflow by @sayakpaul in #13352
• change minimum version guard for torchao to 0.15.0 by @howardzhang-cv in #13355
• [ci] move to assert instead of self.Assert* by @sayakpaul in #13366
• [docs] refactor model skill by @stevhliu in #13334
• Fix Ulysses SP backward with SDPA by @zhtmike in #13328
• Add train flux2 series lora config by @tcaimm in #13011
• [docs] Add NeMo Automodel training guide by @pthombre in #13306
• Fix: ensure consistent dtype and eval mode in pipeline save/load tests by @YangKai0616 in #13339
• [ci] support claude reviewing on forks. by @sayakpaul in #13365
• Fix MotionConv2d to cast blur_kernel to input dtype instead of reverse by @YangKai0616 in #13364
• chore: update claude_review.yml by @hf-security-analysis[bot] in #13374
• corrects single file path validation logic by @andrew-w-ross in #13363
• [docs] deprecate pipelines by @stevhliu in #13157
• 🔒 Pin GitHub Actions to commit SHAs by @paulinebm in #13385
• [docs] add auto docstring and parameter templates documentation for m… by @yiyixuxu in #13382
• Fix typos and grammar errors in documentation by @GalacticAvenger in #13391
• fix(ddim): validate eta is in [0, 1] in DDIMPipeline by @NIK-TIGER-BILL in #13367
• Fix Dynamo lru_cache warnings during torch.compile by @jiqing-feng in #13384
• [tests] refactor wan autoencoder tests by @sayakpaul in #13371
• NucleusMoE-Image by @sippycoder in #13317
• Add examples on how to profile a pipeline by @sayakpaul in #13356
• Update README.md of the profiling guide by @sayakpaul in #13400
• [CI] Refactor Cosmos Transformer Tests by @DN6 in #13335
• [tests] refactor autoencoderdc tests by @sayakpaul in #13369
• [CI] Hunyuan Transformer Tests Refactor by @DN6 in #13342
• Fix VAE offload encode device mismatch in DreamBooth scripts by @azolotenkov in #13417
• Remove references to torchao's AffineQuantizedTensor by @andrewor14 in #13405
• [tests] fix autoencoderdc tests by @sayakpaul in #13424
• [core] fix group offloading when using torchao by @sayakpaul in #13276
• Fix IndexError in HunyuanVideo I2V pipeline by @kaixuanliu in #13244
• improve Claude CI by @yiyixuxu in #13397
• FLUX.2 small decoder by @huemin-art in #13428
• [CI] Add PR/Issue Auto Labeler by @DN6 in #13380
• [CI] Add GLM Image Transformer Model Tests by @DN6 in #13344
• [CI] Use finegrained token for Issue Labeler by @DN6 in #13433
• Handle prompt embedding concat in Qwen dreambooth example by @chenyangzhu1 in #13387
• fix(qwen-image dreambooth): correct prompt embed repeats when using --with_prior_preservation by @chenyangzhu1 in #13396
• Cache RoPE freqs on device to avoid repeated CPU-GPU copy in QwenImage by @akshan-main in #13406
• [tests] tighten dependency testing. by @sayakpaul in #13332
• Fix grammar in LoRA documentation by @Xyc2016 in #13423
• Fix HunyuanVideo 1.5 I2V by preprocessing image at pixel resolution i… by @akshan-main in #13440
• [modular] Add LTX Video modular pipeline by @akshan-main in #13378
• Add ernie image by @HsiaWinter in #13432
• [core] fix fa4 integration by @sayakpaul in #13443
• FlashPack by @hlky in #12700
• [ptxla] fix pytorch xla inference on TPUs. by @entrpn in #13463
• fix some dtype issue for gguf / some gpu backends by @HsiaWinter in #13464
• Fix Qwen Image DreamBooth prior-preservation batch ordering by @azolotenkov in #13441
• [tests] fix deprecated attention processor testing. by @sayakpaul in #13469
• [tests] xfail clip related issues. by @sayakpaul in #13454
• [agent] add modular doc by @yiyixuxu in #13410
• [tests] fix training tests by @sayakpaul in #13442
• fix(profiling): preserve instance isolation when decorating methods by @Akash504-ai in #13471
• [Feat] Adds LongCat-AudioDiT pipeline by @RuixiangMa in #13390
• Fix Flux2 DreamBooth prior preservation prompt repeats by @azolotenkov in #13415
• chore: bump doc-builder SHA for PR upload workflow by @rtrompier in #13476
• Remove compile bottlenecks from ZImage pipeline by @hitchhiker3010 in #13461
• [chore] Add diffusers-format example to LongCatAudioDiTPipeline by @RuixiangMa in #13483
• [core] fix autoencoderkl qwenimage for xla by @sayakpaul in #13480
• add PR fork workable by @paulinebm in #13438
• Add modular pipeline for HunyuanVideo 1.5 by @akshan-main in #13389
• [agents docs] add float64 gotcha by @yiyixuxu in #13472
• fix(ernie-image): avoid locals() comprehension scope issue in callback kwargs by @songh11 in #13478
• [Bugfix] Fix shape mismatch in LongCatAudioDiTTransformer conversion by @RuixiangMa in #13494
• feat: bump safetensors to 0.8.0-rc.0 by @McPatate in #13470
• fix(qwen): fix CFG failing when passing neg prompt embeds with none mask by @Sunhill666 in #13379
• add an example of spmd for flux on v5e-8 by @sayakpaul in #13474
• Add FLUX.2 Klein Inpaint Pipeline by @adi776borate in #13050
• [docs] add a mention of torchao and other backends in speed memory docs. by @sayakpaul in #13499
• Fix Flux2 non-diffusers guidance LoRA conversion by @yadferhad in #13486
• add _native_npu_attention support mask shape like [B,1,1,S] by @chang-zhijie in #13490
• fix(freeu): run FFT in float32 for float16 inputs to avoid ComplexHalf by @Ricardo-M-L in #13503
• Fix non-deterministic T5 outputs in HiDream pipeline tests by @kaixuanliu in #13534
• Fix AuraFlow attn processors applying norm_added_q to key projection by @Ricardo-M-L in #13533
• add _repeated_blocks for ErnieImageTransformer2DModel by @ka...

Contributors

• 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

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, and also check out the announcement post.

New Pipelines and Models

Image 🌆

• Z Image Omni Base: 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: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: 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: 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 #12930.
• Cosmos Predict2.5: 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: 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: 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-intensive image generation scenarios. Thanks to @zRzRzRzRzRzRzR for contributing it in #12973.
• RAE: Representation Autoencoders (aka RAE) are an exciting alternative to traditional VAEs, typically used in the area of latent-space diffusion models of image generation. RAEs leverage pre-trained vision encoders and train lightweight decoders for the task of reconstruction.

Video + audio 🎥 🎼

• LTX-2: LTX-2 is an audio-conditioned text-to-video generation model that can generate videos with synced audio. Full and distilled model inference, as well as two-stage inference with spatial sampling, is supported. We also support a conditioning pipeline that allows for passing different conditions (such as images, series of images, etc.). Check out the docs to learn more!
• Helios: Helios is a 14B video generation model that runs at 17 FPS on a single NVIDIA H100 GPU and supports minute-scale generation while matching a strong baseline in quality. Thanks to @SHYuanBest for contributing this in #13208.

Improvements to Core Library

New caching methods

• MagCache — thanks to @AlanPonnachan!
• TaylorSeer — thanks to @toilaluan!

New context-parallelism (CP) backends

• Unified Sequence Parallel attention — thanks to @Bissmella!
• Ulysses Anything Attention — thanks to @DefTruth!

Misc

• Mambo-G Guidance: New guider implementation (#12862)
• Laplace Scheduler for DDPM (#11320)
• Custom Sigmas in UniPCMultistepScheduler (#12109)
• MultiControlNet support for SD3 Inpainting (#11251)
• Context parallel in native flash attention (#12829)
• NPU Ulysses Attention Support (#12919)
• Fix Wan 2.1 I2V Context Parallel Inference (#12909)
• Fix Qwen-Image Context Parallel Inference (#12970)
• Introduction to @apply_lora_scale decorator for simplifying model definitions (#12994)
• Introduction of pipeline-level “cpu” device_map (#12811)
• Enable CP for kernels-based attention backends (#12812)
• Diffusers is fully functional with Transformers V5 (#12976)

A lot of the above features/improvements came as part of the MVP program we have been running. Immense thanks to the contributors!

Bug Fixes

• Fix QwenImageEditPlus on NPU (#13017)
• Fix MT5Tokenizer → use T5Tokenizer for Transformers v5.0+ compatibility (#12877)
• Fix Wan/WanI2V patchification (#13038)
• Fix LTX-2 inference with num_videos_per_prompt > 1 and CFG (#13121)
• Fix Flux2 img2img prediction (#12855)
• Fix QwenImage txt_seq_lens handling (#12702)
• Fix prefix_token_len bug (#12845)
• Fix ftfy imports in Wan and SkyReels-V2 (#12314, #13113)
• Fix is_fsdp determination (#12960)
• Fix GLM-Image get_image_features API (#13052)
• Fix Wan 2.2 when either transformer isn't present (#13055)
• Fix guider issue (#13147)
• Fix torchao quantizer for new versions (#12901)
• Fix GGUF for unquantized types with unquantize kernels (#12498)
• Make Qwen hidden states contiguous for torchao (#13081)
• Make Flux hidden states contiguous (#13068)
• Fix Kandinsky 5 hardcoded CUDA autocast (#12814)
• Fix aiter availability check (#13059)
• Fix attention mask check for unsupported backends (#12892)
• Allow prompt and prior_token_ids simultaneously in GlmImagePipeline (#13092)
• GLM-Image batch support (#13007)
• Cosmos 2.5 Video2World frame extraction fix (#13018)
• ResNet: only use contiguous in training mode (#12977)

All commits

• [PRX] Improve model compilation by @WaterKnight1998 in #12787
• Improve docstrings and type hints in scheduling_dpmsolver_singlestep.py by @delmalih in #12798
• [Modular]z-image by @yiyixuxu in #12808
• Fix Qwen Edit Plus modular for multi-image input by @sayakpaul in #12601
• [WIP] Add Flux2 modular by @DN6 in #12763
• [docs] improve distributed inference cp docs. by @sayakpaul in #12810
• post release 0.36.0 by @sayakpaul in #12804
• Update distributed_inference.md to correct syntax by @sayakpaul in #12827
• [lora] Remove lora docs unneeded and add " # Copied from ..." by @sayakpaul in #12824
• support CP in native flash attention by @sywangyi in #12829
• [qwen-image] edit 2511 support by @naykun in #12839
• fix pytest tests/pipelines/pixart_sigma/test_pixart.py::PixArtSigmaPi… by @sywangyi in #12842
• Support for control-lora by @lavinal712 in #10686
• Add support for LongCat-Image by @junqiangwu in #12828
• fix the prefix_token_len bug by @junqiangwu in #12845
• extend TorchAoTest::test_model_memory_usage to other platform by @sywangyi in #12768
• Qwen Image Layered Support by @naykun in #12853
• Z-Image-Turbo ControlNet by @hlky in #12792
• Cosmos Predict2.5 Base: inference pipeline, scheduler & chkpt conversion by @miguelmartin75 in #12852
• more update in modular by @yiyixuxu in #12560
• Feature: Add Mambo-G Guidance as Guider by @MatrixTeam-AI in #12862
• Add OvisImagePipeline in AUTO_TEXT2IMAGE_PIPELINES_MAPPING by @alvarobartt in #12876
• Cosmos Predict2.5 14b Conversion by @miguelmartin75 in #12863
• Use T5Tokenizer instead of MT5Tokenizer (removed in Transformers v5.0+) by @alvarobartt in #12877
• Add z-image-omni-base implementation by @RuoyiDu in #12857
• fix torchao quantizer for new torchao versions by @vkuzo in #12901
• fix Qwen Image Transformer s...

Contributors

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: 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: Z-Image is a best-of-its-kind image generation model in the 6B param regime. Thanks to @JerryWu-code in #12703.
• QwenImage Edit Plus: 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 #12357.
• 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 #12545.
• Kandinsky Image Lite: Kandinsky 5.0 Image Lite is a lightweight image generation model (6B parameters). Thanks to @leffff for contributing this in #12664.
• 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 #12593.

New video 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 #12634.
• Kandinsky 5: 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 #12478.
• Hunyuan 1.5: 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: Wan-Animate is a state-of-the-art character animation and replacement video model based on Wan2.1. Given a reference character image and driving motion video, it can either animate the character with motion from the driving video, or replace the existing character in that video with that character.

New kernels-powered attention backends

The kernels library helps you save a lot of time by providing pre-built kernel interfaces for various environments and accelerators. This release features three new kernels-powered attention backends:

• Flash Attention 3 (+ its varlen variant)
• Flash Attention 2 (+ its varlen variant)
• SAGE

This means if any of the above backend is supported by your development environment, you should be able to skip the manual process of building the corresponding kernels and just use:

# Make sure you have `kernels` installed: `pip install kernels`.
# You can choose `flash_hub` or `sage_hub`, too.
pipe.transformer.set_attention_backend("_flash_3_hub")

For more details, check out the documentation.

TaylorSeer cache

TaylorSeer is now supported in Diffusers, delivering upto 3x speedups with negligible-to-none quality compromise. Thanks to @toilaluan for contributing this in #12648. Check out the documentation here.

New training script

Our Flux.2 integration features a LoRA fine-tuning script that you can check out here. We provide a number of optimizations to help make it run on consumer GPUs.

Misc

• Reusing AttentionMixin: Making certain compatible models subclass from the AttentionMixin class helped us get rid of 2K LoC. Going forward, users can expect more such refactorings that will help make the library leaner and simpler. Check out #12463 for more details.
• Diffusers backend in SGLang: sgl-project/sglang#14112.
• We started the Diffusers MVP program to work with talented community members who will help us improve the library across multiple fronts. Check out the link for more information.

All commits

• remove unneeded checkpoint imports. by @sayakpaul in #12488
• [tests] fix clapconfig for text backbone in audioldm2 by @sayakpaul in #12490
• ltx0.9.8 (without IC lora, autoregressive sampling) by @yiyixuxu in #12493
• [docs] Attention checks by @stevhliu in #12486
• [CI] Check links by @stevhliu in #12491
• [ci] xfail more incorrect transformer imports. by @sayakpaul in #12455
• [tests] introduce VAETesterMixin to consolidate tests for slicing and tiling by @sayakpaul in #12374
• docs: cleanup of runway model by @EazyAl in #12503
• Kandinsky 5 is finally in Diffusers! by @leffff in #12478
• Remove Qwen Image Redundant RoPE Cache by @dg845 in #12452
• Raise warning instead of error when imports are missing for custom code by @DN6 in #12513
• Fix: Use incorrect temporary variable key when replacing adapter name… by @FeiXie8 in #12502
• [docs] Organize toctree by modality by @stevhliu in #12514
• styling issues. by @sayakpaul in #12522
• Add Photon model and pipeline support by @DavidBert in #12456
• purge HF_HUB_ENABLE_HF_TRANSFER; promote Xet by @Vaibhavs10 in #12497
• Prx by @DavidBert in #12525
• [core] AutoencoderMixin to abstract common methods by @sayakpaul in #12473
• Kandinsky5 No cfg fix by @asomoza in #12527
• Fix: Add _skip_keys for AutoencoderKLWan by @yiyixuxu in #12523
• [CI] xfail the test_wuerstchen_prior test by @sayakpaul in #12530
• [tests] Test attention backends by @sayakpaul in #12388
• fix CI bug for kandinsky3_img2img case by @kaixuanliu in #12474
• Fix MPS compatibility in get_1d_sincos_pos_embed_from_grid #12432 by @Aishwarya0811 in #12449
• Handle deprecated transformer classes by @DN6 in #12517
• fix constants.py to user upper() by @sayakpaul in #12479
• HunyuanImage21 by @yiyixuxu in #12333
• Loose the criteria tolerance appropriately for Intel XPU devices by @kaixuanliu in #12460
• Deprecate Stable Cascade by @DN6 in #12537
• [chore] Move guiders experimental warning by @sayakpaul in #12543
• Fix Chroma attention padding order and update docs to use lodestones/Chroma1-HD by @josephrocca in #12508
• Add AITER attention backend by @lauri9 in #12549
• Fix small inconsistency in output dimension of "_get_t5_prompt_embeds" function in sd3 pipeline by @alirezafarashah in #12531
• Kandinsky 5 10 sec (NABLA suport) by @leffff in #12520
• Improve pos embed for Flux.1 inference on Ascend NPU by @gameofdimension in #12534
• support latest few-step wan LoRA. by @sayakpaul in #12541
• [Pipelines] Enable Wan VACE to run since single transformer by @DN6 in #12428
• fix crash if tiling mode is enabled by @sywangyi in #12521
• Fix typos in kandinsky5 docs by @Meatfucker in #12552
• [ci] don't run sana layerwise casting tests in CI. by @sayakpaul in #12551
• Bria fibo by @galbria in #12545
• Avoiding graph break by changing the way we infer dtype in vae.decoder by @ppadjinTT in #12512
• [Modular] Fix for custom block kwargs by @DN6 in #12561
• [Modular] Allow custom blocks to be saved to local_dir by @DN6 in #12381
• Fix Stable Diffusion 3.x pooled prompt embedding with multiple images by @friedrich in #12306
• Fix custom code loading in Automodel by @DN6 in #12571
• [modular] better warn message by @yiyixuxu in #12573
• [tests] add tests for flux modular (t2i, i2i, kontext) by @sayakpaul in #12566
• [modular]pass hub_kwargs to load_config by @yiyixuxu in #12577
• ulysses enabling in native attention path by @sywangyi in #12563
• Kandinsky ...

Contributors