ComfyUI Releases

Added dependency and tool parameters to guideline and observation creation, while deprecating the legacy `attach_tool()` method.

Feb 17, 2026

Major release introducing Node Replacement API, training with proper offloading, improved dynamic_vram handling, and new API nodes for Magnific Upscalers and Tencent 3D tools.

Feb 17, 2026

{ "summary": "This release focuses heavily on improving ACE Step 1.5 model support with numerous bug fixes, performance enhancements, and LLM-free operation capabilities. It also introduces a new Create List node, adds LTX2 support to EasyCache, and expands fp16 precision support for Cosmos-Predict2 and Anima models.", "keyHighlights": [ "Comprehensive ACE Step 1.5 improvements including low VRAM optimizations, text encoding enhancements, LLM-free operation, and reference audio padding",

Feb 10, 2026

This release introduces session label propagation, lifecycle hooks, and streaming output to a guideline-based system. It also adds per-agent configuration and fixes various bugs regarding websocket warnings and guideline matching.

Feb 9, 2026

This release focuses on stability improvements for ACE Step 1.5, including fixes for NaN issues, text encoder slowness, and enhanced reference audio support. It also introduces new 3D file handling capabilities and audio processing features, alongside critical fixes for safetensors loading and checkpoint saving.

Feb 5, 2026

Adds MCPB support for single-click Claude Desktop installation, introduces verbose output options, and adds metadata support for tasks to store custom JSON data.

Feb 4, 2026

This release introduces support for the 4B Ace Step 1.5 language model while fixing a VRAM leak in the dynamic VRAM management system and resolving crash issues related to the new model.

Feb 4, 2026

This release enhances support for ACE Step models, particularly version 1.5, with added LoRA compatibility, tiled VAE fixes, and a new progress bar indicator. It also introduces basic 3D model file type support in the Comfy API and addresses several macOS stability issues.

Feb 3, 2026

Version 0.12.0 introduces significant memory management improvements with dynamic VRAM optimizations that reduce RAM usage and fix VRAM OOM issues, alongside expanded API node integrations for external AI services including Recraft, Vidu, and HitPaw. The release also adds support for new models such as Ace Step 1.5 and Qwen 3, and implements KV caching for Llama-based text generation.

Feb 3, 2026

Adds support for dev-only nodes and Grok Imagine nodes while updating the manager version and Python 3.14 compatibility notes.

Jan 29, 2026

Adds support for new AI models (LTX2, Flux2 Klein, Qwen 3), 3D formats (PLY, 3DGS), and audio processing improvements.

Jan 27, 2026

Adds ByteDance nodes and fixes AMD GPU audio handling issues while improving workflow templates and widget support.

Jan 21, 2026

Fixes LiteLLM passthrough support for all providers and resolves Windows file path compatibility issues while introducing a single-model Ollama configuration.

Jan 17, 2026

Enhances task filtering capabilities to identify parallelizable tasks and resolves concurrency issues with file storage and task execution.

Jan 15, 2026

Adds Gemini safety exceptions, Z-Image Controlnet support, and Meshy 3D nodes while improving VAELoader and image resizing.

Jan 15, 2026

Minor bump to adjust memory estimation for LTX AV models.

Jan 13, 2026

Adds audio channel joining, WAN VAE support, Siglip 2 Naflex model support, and enables PyTorch Attention on gfx1200 GPUs.

Jan 13, 2026

This minor release introduces an automated task execution loop with preset workflows, enhances LLM tool understanding via annotations, and fixes compatibility issues with Zod schemas and AI SDKs.

Jan 11, 2026

A minor update focusing on optimizing memory estimation for the LTXV VAE model.

Jan 8, 2026

Fixes stable release workflow issues and introduces support for Gemma 12B quantized weights while improving LTXV text encoder efficiency.

Jan 8, 2026