huggingface/diffusers v0.43.0

Diffusers 0.43.0: MiniMax-H3-Image, Pika H3 Flex, P-H3-Image, Kandinsky S2V, and PAN 2

on GitHub

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.
• [components manager] when a component is registered again: warn -> debug by @yiyixuxu in #15152.
• Fix/flux lora quantization compute embeddings by @kahlun in #15144.
• Add Kandinsky S2V 1.0 (speech-to-video) as a modular pipeline by @nvsm in #15164.
• Add P-H3-Image by @nifleisch in #15171.
• Fix attention backend leaking across models by @raghulchandramouli in #14335.
• Docs: build English only, other languages are machine-translated by @mishig25 in #15149.
• [core] download hub kernel in AttentionModuleMixin.set_attention_backend by @deep1blue in #15075.
• [docs] Fix 22 dead external links in docs/source/en by @mlzerotoheroyoutube in #15174.
• Clarify LoRA adapter weight documentation and type hints by @gorkadabo1 in #15157.
• Fix replicated parameter dtypes during tensor-parallel loading by @yzhautouskay in #15147.
• [examples] Fix gradient clipping in train_text_to_image_lora.py and train_lcm_distill_lora_sdxl.py by @gorkemgul in #15184.
• [feat] Tensor parallelism for MiniMax-H3 by @JingyaHuang in #14609.
• Fix stale links to moved and renamed files by @sarankumar1325 in #15151.
• agents: add a section on _tp_plan in modeling reference. by @sayakpaul in #15194.
• Style changes by @sayakpaul in #15198.
• [docs/CI] contribution guide update, adding a context item on issue template, update issue-link reminder by @yiyixuxu in #15179.
• Replace the remaining forum links with GitHub Discussions by @yiyixuxu in #15200.
• [modular] fill a missing input from the first block that reads it, keep a passed None by @yiyixuxu in #15135.
• Pin hf-doc-builder to 0.6.0 by @yiyixuxu in #15201.
• [ci]: solve protobuf problem for the training e examples problem. by @sayakpaul in #15206.
• Reset stateful cache hooks after every ModularPipeline call by @yiyixuxu in #15212.
• [docs] discourage cross repo references for modular repos. by @sayakpaul in #15081.
• [Modular] Add PAN 2 by @wu-qing-157 in #15202.
• fix: prefer latest flash attn2 by @drbh in #15000.
• MiniMax H3 Image by @sayakpaul in #15230.
• [docs] Distributed inference by @stevhliu in #15172.
• Update dead Kaggle Cats vs Dogs download URL in create_dataset doc by @Hisernberg in #15231.
• Bump doc-builder to render math in docstring descriptions by @mishig25 in #15223.
• Fix the PAN 2 context-parallelism documentation link by @yiyixuxu in #15238.
• Document Cosmos3 Nano, Super, and Edge FP8 as standalone repos by @ConstBob in #15236.
• [core] Enforce minimum supported versions in availability checks by @DN6 in #15148.
• fix: derive text-encoder alpha_pattern keys from the LoRA module name by @akhil9tiet in #14972.
• [chore] remove smoke test file. by @sayakpaul in #15160.
• [chore] update return type annotations for models. by @sayakpaul in #15089.
• post release 0.42.0. by @sayakpaul in #15243.
• Release: v0.43.0-release by @sayakpaul (release commit).

Significant community contributions

• @nifleisch: P-H3-Image generation, editing, and prompt upsampling. (#15171)
• @nvsm: Kandinsky S2V 1.0 speech-driven video generation. (#15164)
• @wu-qing-157: PAN 2 modular video generation and world simulation. (#15202)
• @Shan2L: AMD RDNA4 SageAttention backend. (#15060)
• @raghulchandramouli, @deep1blue, and @drbh: attention backend isolation, Hub kernel loading, and FlashAttention-2 compatibility. (#14335, #15075, #15000)
• @gorkadabo1 and @akhil9tiet: LoRA weight broadcasting, documentation, and text-encoder alpha handling. (#15128, #15157, #14972)
• @AloysJehwin and @yzhautouskay: checkpoint variant preservation and tensor-parallel loading precision. (#14844, #15147)
• @gorkemgul, @kahlun, and @A-Raphie: gradient clipping, quantized embedding preparation, and training startup fixes. (#15184, #15144, #15118)

Thanks to our issue reporters

These authors opened issues resolved by PRs merged into this release. Thank you for helping improve Diffusers!

• @andre-menutole: text-encoder LoRA alpha handling — #14970 → #14972.
• @gorkadabo1: LoRA weight broadcasting and documentation — #15127 → #15128; #15146 → #15157.
• @gorkemgul: ineffective gradient clipping — #15178 → #15184.
• @kahlun: FLUX quantized embedding preparation — #15145 → #15144.
• @Marius-Graml: missing Hub kernel downloads — #13284 → #15075.
• @Pdevadiga45: checkpoint variant shards being deleted — #14719 → #14844.
• @pratikgx: broken documentation links — #14886 → #15174.
• @SeverinVisionary: training crashes with --revision — #15115 → #15118.
• @Shan2L: RDNA4 SageAttention support — #15014 → #15060.

Full changelog: v0.42.0 to the v0.43.0 release commit