{"id":13622,"date":"2026-11-17T11:02:28","date_gmt":"2026-11-17T02:02:28","guid":{"rendered":"https:\/\/localmodelwatch.tsuchitsuchi.com\/?p=13622"},"modified":"2026-11-17T11:11:39","modified_gmt":"2026-11-17T02:11:39","slug":"comfyui-v0-43-0-released","status":"publish","type":"post","link":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/comfyui-v0-43-0-released\/","title":{"rendered":"ComfyUI v0.43.0 Released: MOSS-Image and Fibo 2.0 Support"},"content":{"rendered":"<p><!-- lmw:facts --><\/p>\n<h2>At a Glance<\/h2>\n<div class=\"lmw-table-scroll\" tabindex=\"0\" style=\"overflow-x:auto;-webkit-overflow-scrolling:touch;max-width:100%;\">\n<table style=\"width:max-content;min-width:100%;border-collapse:collapse;\">\n<thead>\n<tr>\n<th>Item<\/th>\n<th>Value<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Repository<\/td>\n<td><a href=\"https:\/\/github.com\/Comfy-Org\/ComfyUI\">Comfy-Org\/ComfyUI<\/a><\/td>\n<\/tr>\n<tr>\n<td>Version<\/td>\n<td><a href=\"https:\/\/github.com\/Comfy-Org\/ComfyUI\/releases\/tag\/v0.43.0\">v0.43.0<\/a><\/td>\n<\/tr>\n<tr>\n<td>Published<\/td>\n<td>2026-10-27<\/td>\n<\/tr>\n<tr>\n<td>License<\/td>\n<td>GPL-3.0<\/td>\n<\/tr>\n<tr>\n<td>Source type<\/td>\n<td>Primary source (the publisher itself)<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<p><em>Values determined by this site&#8217;s code when the information was collected. Dates are JST.<\/em><\/p>\n<p><!-- \/lmw:facts --><\/p>\n<h2>Overview<\/h2>\n<p>ComfyUI&#8217;s latest version &#8220;v0.43.0&#8221; has been released. ComfyUI is a powerful Python-based, GPL-3.0 licensed GUI, API, and inference backend that allows users to run diffusion models and build pipelines using a node-based graph interface.<\/p>\n<p>This version introduces support for and functional enhancements to the MOSS-Image and Fibo 2.0 models, improved inference performance for the Qwen 3.x series, and new nodes such as the Pad Image and Branch nodes. This release is particularly beneficial for users leveraging models like MOSS-Image, Fibo 2.0, and Qwen 3.x in their local environments.<\/p>\n<h2>Key Changes<\/h2>\n<h3>Multifaceted Enhancements and Memory Optimization for MOSS-Image Models<\/h3>\n<p>Support for the MOSS-Image model has been significantly strengthened. This includes support for RoPE scaling, color depth control, and the loading of MOSS-Image edit LoRAs. Additionally, attention bias conditioning, image strength settings during editing, and a dedicated MOSS model merge node have been added. Adjustments have also been made to reduce peak memory usage.<\/p>\n<p>This directly impacts users who want to generate images, edit, or merge models using the MOSS-Image model. To utilize the latest features, workflows will need to be rebuilt after the update using new MOSS-related nodes (such as the dedicated merge node).<\/p>\n<h3>Fibo 2.0 Model Support and Parameter Adjustments<\/h3>\n<p>Support for the Fibo 2.0 model has been added. Furthermore, the RoPE scaling configuration in Fibo 2.0 has been changed to &#8220;NTK aware,&#8221; and a bug related to negative prompt processing during Text-to-Image (T2I) execution has been fixed.<\/p>\n<p>This affects users generating images with Fibo 2.0. If negative prompts were not reflecting properly during T2I generation with Fibo 2.0 in previous versions, or if you were affected by the RoPE settings, updating to this version will restore normal operation and appropriate scaling results.<\/p>\n<h3>Inference Performance Improvements and Fixes for Qwen 3.x and Gemma 4 31B<\/h3>\n<p>As part of performance improvements for Qwen 3.x, &#8220;Adaptive tree MTP&#8221; has been introduced. Issues occurring in the Gemma 4 31B model regarding global attention (k_eq_v and four global KV heads) have also been fixed. In addition, a function to apply single-head LoRA to the PDD head banks of MiniMax-H3 has been added.<\/p>\n<p>This affects users running inference and fine-tuning (applying LoRAs) on these models (Qwen 3.x, Gemma 4 31B, and MiniMax-H3) in ComfyUI. Users running Qwen 3.x can expect more efficient inference simply by updating, while those facing precision issues with Gemma 4 31B will see them resolved by updating to this version without requiring re-conversion.<\/p>\n<h3>Addition of Image Processing and Utility Nodes<\/h3>\n<p>New nodes include the &#8220;Pad Image&#8221; node, which performs mirror padding or padding to multiple sizes, and the &#8220;Branch&#8221; node, which branches processing flows. Furthermore, ACEScg color space support has been added to the Convert Image Color Space and Save Image (Advanced) EXR nodes.<\/p>\n<p>This affects users performing edge processing on images, dynamic workflow control, and color management. After updating, these new nodes will be selectable from the standard node list, helping to organize existing workflows and build strict color pipelines.<\/p>\n<h3>PyTorch Attention Optimization and VRAM Management for AMD GPUs<\/h3>\n<p>PyTorch attention in VAEs has been enabled on AMD environments (particularly gfx11 and similar). However, adjustments have been made to avoid using PyTorch attention on gfx11 when head dimensions exceed 256. Stability and optimization around VRAM and memory have also been implemented, including the introduction of Dynamic VRAM KV caches (Comfy-Aimdo 0.6.0), synchronization of offload streams, and graph cleanup synchronization in model_prefetch.<\/p>\n<p>This affects users running AMD Radeon GPUs or operating graphics cards in environments with limited VRAM. While no special configuration changes are required, improved operating speed and stability can be expected in AMD environments.<\/p>\n<h2>Supported Models and Hardware<\/h2>\n<h3>New and Enhanced Model Architectures<\/h3>\n<p>This version adds support for and enhances the following model architectures. Users can utilize these latest models directly in ComfyUI or through dedicated nodes.<\/p>\n<ul>\n<li><strong>MOSS-Image<\/strong>: Support for the MOSS-Image model has been substantially expanded. In addition to RoPE scaling support, color depth control, application of MOSS-Image edit LoRAs, attention bias conditioning, and image strength adjustments are now possible. A dedicated MOSS model merge node has also been added to improve model customizability.<\/li>\n<li><strong>Fibo 2.0<\/strong>: The Fibo 2.0 model is newly supported. RoPE scaling is set to NTK aware, and bugs concerning negative prompt processing in Text-to-Image (T2I) have been fixed.<\/li>\n<li><strong>Qwen 3.x<\/strong>: Inference performance has been improved in the Qwen 3.x series through the introduction of Adaptive tree MTP (Multi-Token Prediction). This allows Qwen 3.x-based models to run faster.<\/li>\n<li><strong>Gemma 4 31B<\/strong>: Added support for the global attention structure specific to the Gemma 4 31B model (k_eq_v, 4 global KV heads). This enables accurate inference of 31B parameter Gemma 4 models.<\/li>\n<li><strong>MiniMax-H3<\/strong>: Single-head LoRAs can now be applied to the PDD head banks of MiniMax-H3.<\/li>\n<li><strong>Flux 2 VAE<\/strong>: Fusion and stripping of the Flux 2 VAE are supported through comfy-kitchen, facilitating VAE optimization.<\/li>\n<\/ul>\n<h3>External Model and Service Integration via Partner Nodes<\/h3>\n<p>Through API nodes provided by partners, the following latest models and features are now available. These rely on cloud service integration rather than local inference.<\/p>\n<ul>\n<li><strong>Black Forest Labs (BFL)<\/strong>: The FLUX 3 Video node now supports 2K and 4K resolution generation. A FLUX 3 Max node has also been added.<\/li>\n<li><strong>Kling AI<\/strong>: The Kling 4.0 video generation model node has been added.<\/li>\n<li><strong>PixVerse<\/strong>: Added support for the PixVerse V6-Max model.<\/li>\n<li><strong>Krea<\/strong>: &#8220;Ultra thinking mode&#8221; has been added to the Krea 3 Image node.<\/li>\n<li><strong>Bria<\/strong>: Image generation and editing nodes using FIBO 2.0 have been added.<\/li>\n<li><strong>Ideogram<\/strong>: SVG vector image generation is now possible via the Ideogram Vector node.<\/li>\n<li><strong>HeyGen<\/strong>: Instant voice cloning and text-to-speech (TTS) features have been added.<\/li>\n<li><strong>Meshy<\/strong>: The Meshy T2.5 Smart Topology node has been added, strengthening 3D model generation workflows.<\/li>\n<li><strong>VEED \/ Mirelo<\/strong>: New nodes such as Fabric 1.0, Lipsync 2.0 (VEED), and SFX generation (Mirelo) are available.<\/li>\n<\/ul>\n<h3>Hardware Support Improvements<\/h3>\n<ul>\n<li><strong>AMD GPU (ROCm)<\/strong>: PyTorch attention has been enabled in the VAE (Variational Autoencoder). However, restrictions are in place to disable PyTorch attention for compatibility when head dimensions exceed 256 on the gfx11 architecture.<\/li>\n<li><strong>Huawei Ascend NPU<\/strong>: Huawei branding has been added to the Ascend NPU section of the README, clarifying official support.<\/li>\n<\/ul>\n<h2>How to Get It<\/h2>\n<p>To update ComfyUI to the latest state, follow the steps below. Pay attention to the Python and PyTorch version requirements.<\/p>\n<h3>Recommended Environment<\/h3>\n<ul>\n<li><strong>Python<\/strong>: Python 3.13 is recommended. Python 3.14 also works, but some custom nodes may encounter issues.<\/li>\n<li><strong>PyTorch<\/strong>: PyTorch 2.7 or higher is the minimum requirement, but using the latest version is strongly recommended. If using an NVIDIA 20 series or newer GPU, PyTorch with cu130 or higher is required.<\/li>\n<\/ul>\n<h3>Setup Steps<\/h3>\n<ol>\n<li>\n<p>Update dependencies (if using the manager):<br \/>\n   <code>bash pip install -r manager_requirements.txt<\/code><\/p>\n<\/li>\n<li>\n<p>Launch ComfyUI with the manager enabled:<br \/>\n   <code>bash python main.py --enable-manager<\/code><\/p>\n<\/li>\n<\/ol>\n<p>When updating an existing environment, run <code>git pull<\/code> in the repository directory to fetch the latest source code, then update dependencies using the command above.<\/p>\n<p><!-- lmw:releases --><\/p>\n<h2>Releases Since Our Last Article<\/h2>\n<p><em>Compiled by Local Model Watch from the project&#8217;s GitHub releases: the versions between this release and the last one we covered, which did not get separate articles.<\/em> <em>Full history: <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/engine-comfyui-en\/\">release tracker<\/a>.<\/em><\/p>\n<div class=\"lmw-table-scroll\" tabindex=\"0\" style=\"overflow-x:auto;-webkit-overflow-scrolling:touch;max-width:100%;\">\n<table style=\"width:max-content;min-width:100%;border-collapse:collapse;\">\n<thead>\n<tr>\n<th>Version<\/th>\n<th>Released<\/th>\n<th>Release notes<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>v0.42.1<\/td>\n<td>2026-10-19<\/td>\n<td><a href=\"https:\/\/github.com\/Comfy-Org\/ComfyUI\/releases\/tag\/v0.42.1\">GitHub<\/a><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<p><!-- \/lmw:releases --><\/p>\n<p><!-- lmw:related --><\/p>\n<h2>Related Articles<\/h2>\n<ul>\n<li><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/comfyui-v0420-released\/\">ComfyUI v0.42.0 Released: MOSS-image Support and Qwen Optimizations<\/a><\/li>\n<li><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/comfyui-v0450-released\/\">ComfyUI v0.45.0 Released with Qwen 3.x Optimizations<\/a><\/li>\n<li><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/comfyui-v0400-released\/\">ComfyUI v0.40.0 Released: Native LoRA Stack and More<\/a><\/li>\n<li><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/comfyui-v0-39-0-released\/\">ComfyUI v0.39.0 Released with New CLI Flags and Partner Nodes<\/a><\/li>\n<\/ul>\n<p><!-- \/lmw:related --><\/p>\n<p><!-- lmw:next-steps --><\/p>\n<h2>What to Read Next<\/h2>\n<ul>\n<li><strong>Follow this tool<\/strong> \u2192 <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/engine-comfyui-en\/\">ComfyUI overview and release history (163 releases tracked)<\/a><\/li>\n<li><strong>Other image, video and audio<\/strong> \u2192 <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/engine-diffusers-en\/\">Diffusers<\/a> \/ <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/engine-invokeai-en\/\">InvokeAI<\/a> \/ <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/engine-whisper-cpp-en\/\">whisper.cpp<\/a><\/li>\n<\/ul>\n<p><!-- \/lmw:next-steps --><\/p>\n<h2>Sources<\/h2>\n<ul>\n<li><a href=\"https:\/\/github.com\/Comfy-Org\/ComfyUI\/releases\/tag\/v0.43.0\">Comfy-Org\/ComfyUI v0.43.0<\/a><\/li>\n<li><a href=\"https:\/\/github.com\/Comfy-Org\/ComfyUI\/pull\/17172\">PR #17172: Add rope scaling support to MOSS-image<\/a><\/li>\n<li><a href=\"https:\/\/github.com\/Comfy-Org\/ComfyUI\/pull\/17206\">PR #17206: Support MOSS-image color-depth control<\/a><\/li>\n<li><a href=\"https:\/\/github.com\/Comfy-Org\/ComfyUI\/pull\/17296\">PR #17296: Support MOSS-Image edit LoRAs<\/a><\/li>\n<li><a href=\"https:\/\/github.com\/Comfy-Org\/ComfyUI\/pull\/17349\">PR #17349: Add model specific merge node for MOSS<\/a><\/li>\n<li><a href=\"https:\/\/github.com\/Comfy-Org\/ComfyUI\/pull\/17240\">PR #17240: feat: Add Fibo 2.0 support<\/a><\/li>\n<li><a href=\"https:\/\/github.com\/Comfy-Org\/ComfyUI\/pull\/17290\">PR #17290: Fibo 2.0: Change RoPE scaling to NTK aware<\/a><\/li>\n<li><a href=\"https:\/\/github.com\/Comfy-Org\/ComfyUI\/pull\/17273\">PR #17273: qwen 3.x performance improvements part III &#8211; Adaptive tree MTP<\/a><\/li>\n<li><a href=\"https:\/\/github.com\/Comfy-Org\/ComfyUI\/pull\/17220\">PR #17220: Fix Gemma 4 31B global attention<\/a><\/li>\n<li><a href=\"https:\/\/github.com\/Comfy-Org\/ComfyUI\/pull\/17196\">PR #17196: Fuse and strip the Flux2 VAE through comfy-kitchen<\/a><\/li>\n<li><a href=\"https:\/\/github.com\/Comfy-Org\/ComfyUI\/pull\/17271\">PR #17271: Enable pytorch attention in the VAE on AMD<\/a><\/li>\n<li><a href=\"https:\/\/github.com\/Comfy-Org\/ComfyUI\/pull\/17323\">PR #17323: Don&#8217;t use pytorch attention for head dims above 256 on gfx11<\/a><\/li>\n<li><a href=\"https:\/\/github.com\/Comfy-Org\/ComfyUI\/pull\/17359\">PR #17359: docs: add Huawei branding to Ascend NPU section in README<\/a><\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>ComfyUI v0.43.0 is out, featuring MOSS-Image and Fibo 2.0 support, Qwen 3.x performance improvements, and AMD GPU VAE attention optimizations.<\/p>\n","protected":false},"author":1,"featured_media":13621,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[105],"tags":[1990,677,3891,1032,3720,520,1547],"class_list":["post-13622","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-engines-and-tools","tag-amd-rocm-en","tag-comfyui-en","tag-fibo-2-0-en","tag-gemma-en","tag-moss-image-en","tag-qwen-en","tag-verified"],"lang":"en","translations":{"en":13622,"ja":13620},"_links":{"self":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts\/13622","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/comments?post=13622"}],"version-history":[{"count":2,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts\/13622\/revisions"}],"predecessor-version":[{"id":17946,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts\/13622\/revisions\/17946"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/media\/13621"}],"wp:attachment":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/media?parent=13622"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/categories?post=13622"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/tags?post=13622"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}