ComfyUI

This release includes a minor tweak to the ltxv vae memory estimation.

This release introduces support for Gemma 12B quantized weights and includes several optimizations for LTXAV text encoders, alongside dependency bumps and workflow fixes.

This release introduces support for sage attention 3, exposes several components in the public API, and includes numerous bug fixes, especially around model loading, memory management, and API nodes integration.

This release introduces new nodes like ManualSigmas and Kling Motion Control, enables async offload for AMD, and enforces a minimum PyTorch version of 2.4. It also deprecates the "seededit" feature.

This release introduces significant refactoring, including resolution bucketing and Trainer updates, alongside new API endpoints and expanded model support for Qwen and Gemma architectures. Several bug fixes address issues related to batching, mask handling, and specific API nodes.

This release introduces support for GPT-Image-1.5 in API Nodes and resolves several bugs related to V3 node processing, SA-Solver accuracy, and torch initialization.

This release focuses heavily on memory optimization, especially around Z Image and VAE loading, alongside introducing several new video and image generation nodes (Kling, Tripo 3.0, new samplers). Several old API nodes (Kling, Pika) were dropped, and internal API type handling was cleaned up.

This release introduces the Ovis image model, significant VRAM reductions via temporal rolling VAE, and extensive migration to the V3 Schema for several node groups. It also includes numerous bug fixes focused on memory management, model loading, and compatibility across various models like Kandinsky and Qwen.

This release introduces significant feature enhancements, including cheap latent preview for flux 2, new API nodes, and improved LoRA training capabilities. It also includes several performance optimizations related to VRAM usage and bug fixes across various components.

This release introduces the new Z Image model and includes fixes for LoRA functionality in mixed fp8 environments and Flux2 memory estimation.

This release focuses on stability and optimization, including a fix for a crash and reduced VRAM usage for the flux 2 text encoder. Workflow templates have also been updated.

This release introduces support for Flux 2, updates workflow templates with new video variants, and includes several fixes related to API nodes and quantized tensor handling.

This release introduces new API nodes (Topaz, Nano Banana Pro), updates the HunyuanVideo model, and includes several bug fixes related to image batching and audio conversion.

This release introduces support for NVIDIA cu126 workflows and adds new Gemini models via api-nodes, alongside several bug fixes for EasyCache and hunyuan 3d.

This release focuses heavily on memory optimization across various models (Qwen, Flux) by refining tensor pinning logic and reducing VRAM usage. It also includes several bug fixes for controlnets, quantized operations, and API node updates.

This release introduces a new Mixed Precision Quantization System and significant performance improvements, especially around memory offloading and torch.compile optimizations for fp8 operations. Several API nodes were updated to use a new client structure, and various stability fixes were applied to caching and execution flow.

This release introduces significant updates to API nodes, including V3 schema conversions and network client improvements, alongside enhanced dependency caching and hardware-specific optimizations.

This release focuses heavily on performance improvements, including faster workflow cancelling and optimizations for chroma radiance, alongside addressing several GPU-specific memory and compilation issues. It also introduces Python 3.14 instructions and deprecates older API usage.

This release focuses heavily on migrating nodes to the V3 schema, introducing new features for GeminiImage API nodes, and implementing performance and stability fixes, especially for AMD users.

This release introduces Gemma 3 text encoder support and a new Sora2 API node, alongside several bug fixes and schema conversions for existing nodes.

This release focuses heavily on migrating existing nodes to the V3 schema, introduces support for the hunyuan vae, and includes several bug fixes related to VRAM leaks and API nodes. The 'soundfile' dependency has also been removed.

This release focuses on fixing workflow permissions and resolving a missing parameter issue in Rodin3D-Gen2, alongside adding support for the Seedance Pro model in a specific node.

This release introduces new API nodes for Rodin3D and audio processing, alongside significant internal updates converting many nodes to the V3 schema. Key improvements include memory leak fixes and enhancements to the stable release automation.

This release heavily focuses on integrating and refining support for the Hunyuan image model, alongside significant updates to API node schemas (V3 conversion) and the introduction of several new model supports across image, audio, and video generation.

This release introduces the new ByteDanceSeedream (4.0) node and updates the template version to 0.1.78.

This release introduces support for Hunyuan 3D 2.1 and Hunyuan image 2.1 models, alongside several bug fixes related to lowVRAM, AMD compatibility, and progress updates. Several new API nodes for video and audio processing were also added.

This release focuses heavily on migrating various nodes, including Stable Cascade, Video, and Runway API nodes, to the V3 schema. New features include USO subject identity lora, Convolution AutoTuning, and new API nodes from ByteDance and Ideogram.

This release focuses on performance improvements, specifically reducing RAM usage on Windows systems.

This release introduces new nodes for video extension and latent cutting, alongside support for the Wan2.2 5B models. Several minor bugs related to s2v extension and audio trimming have also been resolved.

This release includes an update to the nodes_model_patch.py file via a pull request.

This release introduces new features like the Gemini Image API node and wav2vec2 audio encoder, alongside several bug fixes related to caching, S2V performance, and ControlNet compatibility.

This release focuses heavily on V3 schema conversions for various nodes (String, Google Veo, Ideogram) and introduces significant support for Qwen Diffsynth Controlnets. It also includes several performance and usability improvements like EasyCache implementation and bug fixes for workflow saving and latent conditioning.

This release focuses heavily on compatibility fixes, especially for older PyTorch versions, alongside numerous updates to API nodes and improvements to Qwen model handling. Key additions include context window support and new video/image generation nodes.

This release introduces significant support for Qwen Image models, including merging nodes and LoRA support, alongside various stability fixes and asynchronous API enhancements. It also updates internal templates and testing workflows.

This release introduces new features like the Recently Used Items API and initial support for the qwen image model, alongside several bug fixes related to device placement and ControlNet functionality.

This release introduces backend partial execution support and the initial V3 Node Schema Definition. It also includes several optimizations and a fix for the WanFirstLastFrameToVideo node.

This release focuses on performance and memory optimization, particularly for the wan2.2 VAE and 5B i2v model, alongside a Windows-specific VRAM reservation improvement.

This release introduces significant hardware support updates, including Iluvatar CoreX and WAN 2.2, alongside improvements to XPU functionality and training node features. It also replaces torchaudio.load with pyav for audio loading.

This release introduces several new features, including support for async node functions and a reconstructed SA-Solver, alongside numerous bug fixes addressing issues in training, API polling, and various node operations. The release also cleans up old code and improves compatibility.

This release introduces several new nodes (TCFG, ER-SDE migration, SLG, ImageRotate/Flip, Moonvalley) and enhances custom node handling with whitelisting options. It also includes performance optimizations and dependency cleanups, notably removing a deprecated torchaudio function.

This release updates the readme with new models and includes bug fixes related to VAE encoding noise and memory estimation for kontext.

This release introduces major new image editing models, Flux Kontext and Omnigen2, alongside several bug fixes and dependency updates. Key changes include improved error handling for OOM situations and updates to model naming conventions.

This release introduces significant feature enhancements, including support for Cosmos predict2 models, LoRA training via weight adapters, and sqlite database support. It also includes various bug fixes, especially concerning fp16 operations and hardware-specific optimizations.

This release introduces several new utility nodes like regex replace and image stitching, improves API handling, and fixes bugs including a potential XSS vulnerability and deprecated Pillow API usage.

This release introduces support for HiDream SimpleTuner loras and adds BFL Kontext API Nodes. It also includes a fix for a unique_id issue in Pika API node imports.

This release introduces an update to memory estimation to account for conditioning inputs and fixes a frontend issue where API nodes were unnecessarily re-executing.

This release introduces several new API nodes for integration with services like Gemini and OpenAI, alongside hardware-specific performance improvements and bug fixes.

This release focuses on performance improvements, especially for queued prompts and torch.compile compatibility, alongside adding new features like stackable VACE conditionings and fp8 compute support.

This release introduces significant feature enhancements, including new audio output formats (opus, mp3), string operation nodes, APG guidance implementation, and improved API client robustness. Several bug fixes address issues related to RMSNorm, APG momentum, and API request retries.

This release introduces support for streaming text to the node UI and Comfy API keys, alongside several bug fixes for samplers, fp8 operations, and torch warnings.

This release introduces initial support for the ACE-Step model, improves memory estimation for LTXV models, and resolves several bugs related to ACE functionality and character encoding. Torchaudio is no longer a hard dependency.

This release adds more API nodes and introduces an option to disable them, alongside several updates to normalization layers across various models and a fix for video saving errors.

This release introduces experimental asynchronous offloading, new nodes like T5TokenizerOptions, and significant fixes for ltxv and serialization issues. It also bumps the minimum required version for the pyav library.

This release introduces significant feature enhancements, including the Unified Weight Adapter system, OFT/BOFT algorithms, API Node support, and updates for HiDream and WAN VACE models. Several bug fixes address frontend issues, video encoding, and specific model implementations.

This release introduces significant new model support, including HiDream and Wan variants, alongside performance optimizations like MLU memory optimization and Dependency Aware Node Caching. It also deprecates an old input option and enhances node isolation security.

No release notes provided.

This release introduces support for the Hunyuan3Dv2 model and Wan Control Loras, alongside several bug fixes and updates to the frontend package.

This release introduces support for new model types, optimizes compute precision for WAN 2.1 models, and includes several bug fixes related to frontend messages and VAE stability.

This release introduces support for new video and model formats, sets fp16 as the default compute dtype for WAN 2.1 models, and includes several bug fixes related to frontend messages and VAE stability.

This release primarily addresses a regression related to the use of incompatible embeddings.

This release introduces support for the HunyuanVideo Image to Video model and includes several internal improvements to NodeDef handling and argument parsing.

This release primarily addresses a bug related to ltxv crop guides functionality with zero keyframes.

This release primarily addresses a bug in the LTXV 0.9.5 img2vid workflow related to image resolution handling.

This release introduces support for LTXV 0.9.5 and improves the guide message displayed when the frontend is missing.

This release introduces Temporal Area Composition and optimizes fp16 accumulation loading. It also updates frontend dependency management via a PyPI package.

This release introduces performance improvements for Wan via batching and adds support for Cambricon MLU devices.

This minor release addresses a specific bug related to artifact generation during inference with fp8 weights.

This release introduces initial support for the new Wan 2.1 video models and updates installation documentation.

This release brings various bug fixes across attention mechanisms, hardware acceleration (ROCm, DirectML, Ascend NPU), and dependency updates. Key feature additions include support for ancestral samplers and Lumina2 system prompt integration.

This release introduces support for Lumina Image 2.0 and adds a new node for 3D animation preview. It also includes improvements to response body compression handling and better guidance for sageattention.

This release introduces support for Nvidia Blackwell/50 series hardware and updates the underlying PyTorch version. It also removes support for Python 3.8 and includes several bug fixes and new nodes.

This release introduces support for Nvidia Cosmos 7B/14B models and the new res_multistep sampler, alongside numerous bug fixes and internal improvements, including better handling of file loading on PyTorch 2.4+.

This release introduces support for Nvidia Cosmos 7B/14B video models, adds the res_multistep sampler, and includes several stability fixes, particularly around model loading and sampler determinism.