ComfyUI v0.42.0 Released: MOSS-image Support and Qwen Optimizations

At a Glance
| Item | Value |
|---|---|
| Repository | Comfy-Org/ComfyUI |
| Version | v0.42.0 |
| Published | 2026-10-19 |
| License | GPL-3.0 |
| Source type | Primary source (the publisher itself) |
Values determined by this site’s code when the information was collected. Dates are JST.
Overview
ComfyUI, the Python-based open-source GUI and inference engine that provides an intuitive node-based UI and powerful inference backend for diffusion models, has released its latest version, “v0.42.0". ComfyUI allows users to build and run various generative models, such as images and videos, as flexible workflows.
This version includes support for notable models and speedups, such as added support for the image generation model “MOSS-image" and the second wave of inference performance improvements for the Qwen 3.x series. It also includes the removal of older models in Quiver integration nodes and fixes regarding the handling of quantization formats during CPU execution, so users operating in related environments are advised to check before updating.
Breaking Changes and Deprecations
This version phases out older models in certain partner nodes and drops support for specific quantization formats in CPU environments.
| Change Item | Previous Behavior | Updated Behavior |
|---|---|---|
| Quiver Partner Node | Arrow 1.x models could be used | Arrow 1.x models removed due to retirement |
| Quantization Support on CPU Execution | fp8 / nvfp4 / mxfp8 / int8 operation specification allowed | Changed to a specification where CPU devices do not support these quantization formats |
This affects users who have created workflows utilizing Arrow 1.x models via Quiver nodes, or those running quantized models such as fp8 or int8 in CPU environments without a GPU. If you are using a CPU environment, please review your model formats and execution settings.
Key Changes
New Features
New Support for MOSS-image Model Support for the image generation model “MOSS-image" has been newly added(#17095). In addition, the application of fused situ to streamline processing has been implemented(#17099). Users who want to generate images using MOSS-image can now handle the model directly within their ComfyUI workflows by updating to this version.
Addition of Krea 3 Image Node In Partner Nodes, the “Krea 3 Image" node has been newly introduced(#17067). This affects users who want to call various Krea image generation features as nodes. It can be used by incorporating the corresponding node into the workflow.
Support for Loading Embedding Models in Qwen 3.5 A feature to load embedding models for Qwen 3.5 has been added(#17159). This is useful for those utilizing text processing with Qwen 3.5 or leveraging embeddings within workflows.
Support for PDD Baked-in Models Support has been implemented for models with PDD (technologies such as Position-Dependent Diffusion) baked in(#17157). Models with the corresponding specification can now be correctly loaded and executed on the inference pipeline.
Performance Improvements
Performance Improvements for Qwen 3.x (Part 2) Inference processing optimization has been implemented for models in the Qwen 3.x series(#17007). Improvements in processing speed can be expected in environments running Qwen 3.x large language models on ComfyUI. No special re-downloading or setting changes are required, as it is automatically applied via the update.
Bug Fixes and Adjustments
Cleanup and Think-Block Fixes in TextGenerate Node A bug where resources were not properly released after text generation processing has been fixed(#17169). Additionally, an issue where template processing for think-blocks in Qwen models was not performed correctly has been resolved(#17121). Stability when repeatedly executing the TextGenerate node has been improved.
Webcam Node Malfunction Fix An issue where the webcam node failed to operate normally in some environments has been fixed(#17144). Users incorporating real-time camera input into their workflows will have their functionality restored by updating to this version.
3D Model Remeshing Refactoring and Empty Mesh Crash Fix Structural improvements related to 3D model remeshing processing (CORE-560) have been carried out(#16805). Along with this, a bug causing the application to crash when processing empty meshes in solid mode has also been resolved(#17140). The robustness of pipelines handling 3D-related nodes has been enhanced.
Supported Models and Hardware
In this version, multiple updates have been made regarding new support for image generation and language models, behavior adjustments, and supported platforms for the execution environment.
Supported Models and Formats
- MOSS-image: Support for the image generation model “MOSS-image" has been newly added. The application of fused situ has also been carried out, allowing users to build efficient inference pipelines on ComfyUI.
- Qwen 3.5: The feature to load embedding models for Qwen 3.5 is supported. It can be utilized in text processing and related tasks.
- Qwen 3.x Series: Inference performance optimization (Part 2) targeting the Qwen 3.x large language models has been introduced, aiming to speed up generation processing utilizing these models. Think-block template bugs in the TextGenerate node have also been resolved.
- PDD Baked-in Models: Official support is provided for models in formats where PDD (Position-Dependent Diffusion, etc.) is baked directly into the model.
- FLUX 3 Image: The name “FLUX 3 Image" is now used in node display labels.
- Quiver (Arrow 1.x) Discontinuation: In the partner node Quiver, Arrow 1.x models have been removed due to retirement. These models can no longer be used.
Additionally, documentation improvements regarding the custom quantization format “comfy quant" have been implemented.
Hardware and Platforms
- ARM Linux: “ARM Linux" has been newly added as a target environment for execution and unit tests. Improvements in stability and ongoing verification in ARM architecture Linux environments are progressing.
- Quantization Restrictions on CPU Execution: On CPU devices, low-bit quantization formats such as fp8, nvfp4, mxfp8, and int8 are no longer supported by specification. When executing these quantized models, it is necessary to prepare a compatible GPU environment or select a supported format.
How to Get It
Here are the environmental requirements and installation steps to update ComfyUI to the latest version and use it stably.
Checking Prerequisites
The recommended execution environment for ComfyUI is guided as follows. Please check your environment before updating.
- Python Version: Python 3.13 is very well supported. If issues arise with custom node dependencies, using Python 3.12 is recommended. Python 3.14 also works, but issues may occur with some custom nodes (free-threaded builds also work, but full support is not guaranteed as the GIL is enabled by some dependencies).
- PyTorch and CUDA Version: torch 2.7 or higher is minimally supported, but using newer versions is strongly recommended. When using Nvidia 20-series or higher GPUs, PyTorch version cu130 or higher is required. If more than 6 months have passed since installing PyTorch, updating to the latest major version is recommended.
Setup Procedure
After updating the existing environment’s code to the latest version (v0.42.0), update various dependency packages as needed.
To set up the dependencies for ComfyUI Manager, run the following command:
pip install -r manager_requirements.txt
To start with the manager functionality enabled, append the --enable-manager flag to the startup command:
python main.py --enable-manager
Settings When Using TLS/SSL
If you want to encrypt communications and securely access the Web UI, create a self-signed certificate and specify the startup options.
An example command to generate the certificate and private key is as follows:
openssl req -x509 -newkey rsa:4096 -keyout key.pem -out cert.pem -sha256 -days 3650 -nodes -subj "/C=XX/ST=StateName/L=CityName/O=CompanyName/OU=CompanySectionName/CN=CommonNameOrHostname"
After generating the certificate, starting with the following options enabled will activate TLS/SSL, allowing you to access via the https://… format:
python main.py --tls-keyfile key.pem --tls-certfile cert.pem
If you are using a container environment, you can place and load the key and certificate in the current directory by using relative paths with volume mounts (-v).
For detailed release contents and diff information, please check the official release page.
Related Articles
- ComfyUI v0.43.0 Released: MOSS-Image and Fibo 2.0 Support
- ComfyUI v0.40.0 Released: Native LoRA Stack and More
- ComfyUI v0.45.0 Released with Qwen 3.x Optimizations
- ComfyUI v0.39.0 Released with New CLI Flags and Partner Nodes
What to Read Next
- Follow this tool → ComfyUI overview and release history (163 releases tracked)
- Other image, video and audio → Diffusers / InvokeAI / whisper.cpp
Sources
- https://github.com/Comfy-Org/ComfyUI/releases/tag/v0.42.0
- https://github.com/Comfy-Org/ComfyUI/pull/17067
- https://github.com/Comfy-Org/ComfyUI/pull/17075
- https://github.com/Comfy-Org/ComfyUI/pull/17078
- https://github.com/Comfy-Org/ComfyUI/pull/17095
- https://github.com/Comfy-Org/ComfyUI/pull/17099
- https://github.com/Comfy-Org/ComfyUI/pull/17102
- https://github.com/Comfy-Org/ComfyUI/pull/17007
- https://github.com/Comfy-Org/ComfyUI/pull/17121
- https://github.com/Comfy-Org/ComfyUI/pull/17139
- https://github.com/Comfy-Org/ComfyUI/pull/17140
- https://github.com/Comfy-Org/ComfyUI/pull/17144
- https://github.com/Comfy-Org/ComfyUI/pull/17155
- https://github.com/Comfy-Org/ComfyUI/pull/17157
- https://github.com/Comfy-Org/ComfyUI/pull/17159
- https://github.com/Comfy-Org/ComfyUI/pull/17169
- https://github.com/Comfy-Org/ComfyUI/pull/17170

