ComfyUI-LevelPixel-Advanced

ComfyUI-LevelPixel-Advanced
★ 21

LLM节点VLM/LLaVaGGUF支持自动卸载VRAM
提供 LevelPixel 的高级 ComfyUI 节点，便捷使用 GGUF 格式的 LLM/VLM（含 LLaVa、Qwen2.5/ Qwen2.5-VL）与 RAM 模型，支持自动卸载 VRAM。
💡 在 ComfyUI 中加载并高效管理 GGUF 格式的多模态模型。
🍴 1 Forks💻 Python🔄 2026-03-21
📦
网盘下载
复制链接后前往夸克网盘下载
https://pan.quark.cn/s/9d76119b2ef2
📦 requirements.txt
torch>=2.0.1
torchvision>=0.15.2
pillow>=9.4.0
numpy>=1.26.4
huggingface-hub>=0.19.0
transformers>=4.35.0
safetensors>=0.3.0
scikit-build-core>=0.10.7
transparent-background>=1.2.4
timm>=0.4.12
fairscale>=0.4.4
rembg>=2.0.66
matplotlib
diskcache>=5.6.1
jinja2>=2.11.3
typing-extensions>=4.5.0
ram
@
git+https://github.com/Level-Pixel/recognize-anything.git
banner_LevelPixel_with_logo
level-pixel-advanced-nodes_2
📄 README

Level Pixel nodes for ComfyUI – Advanced nodes

The purpose of this package is to collect the most necessary and atomic nodes for working with LLM and VLM models with GGUF format. Installation and maintenance of LLM and VLM nodes based on LLaVA is more complex, so this node package should now be installed separately from the main Level Pixel node package.

*Our dream is to see the possibilities for convenient creation of full automation in ComfyUI workflows. We will try to get closer to it.*

In this Level Pixel Advanced node pack you will find:

LLM nodes, LLaVa and other VLM nodes, Autotagger, RAM autotagger, WD Autotagger

Recommend that you install the main node package from Level Pixel:

https://github.com/LevelPixel/ComfyUI-LevelPixel

The official repository of the current node package is located at this link:

https://github.com/LevelPixel/ComfyUI-LevelPixel-Advanced

Like our nodes? Then we’d be happy to see your star on our GitHub repository!

Contacts, services and support

Our official Patreon page:

https://www.patreon.com/LevelPixel

On Patreon you can get services from us on issues related to ComfyUI, Forge, programming and AI tools.

You can also support our project and support the development of our node packages.

For cooperation, suggestions and ideas you can write to email:

Installation

Pre-Installation

Before running ComfyUI with this node package, you should make sure that you have the following programs and libraries installed so that ComfyUI can compile the necessary libraries and programs for llama-cpp-python (the main library that allows you to use any current GGUF models and neural network architectures):

  • Download CUDA driver – install the 12.9 version (versions 13.0 and newer are not supported at this time).

    • After installation, make sure the following lines are in the Path field of System Environment Variables:

    C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.9\libnvvp

    C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.9\bin

    And in System Environment Variables, add the CMAKE_ARGS variable and set it to the following:

    -DGGML_CUDA=on -DCMAKE_GENERATOR_TOOLSET='cuda=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.9'

  • Download Visual Studio 2022 Community IDE with libraries for compiling C++ programs, specifically with individual components (select them in Visual Studio Installer when installing/modifying Visual Studio 2022):

    • For Windows 10/11:

  • MSVC v143 – VS 2022 C++ x64/x86 build tools (Latest)
  • C++ Cmake tools for Windows

  • Download CMAKE official distribution – install the latest version.

    • After installation, make sure that the Environment Variables line in Path is:

    C:\Program Files\CMake\bin

    It is also recommended to update ComfyUI to the latest version (including all dependencies) by running the file ComfyUI\update\update_comfyui_and_python_dependencies.bat (don’t forget to press enter after the download of the latest libraries is complete)

    Installation package using ComfyUI Manager (recommended):

    Install ComfyUI Manager and do steps introduced there to install this repo ‘ComfyUI-LevelPixel-Advanced’.

    The nodes of the current package will be updated automatically when you click “Update ALL” in ComfyUI Manager.

    Alternative installation package

    Clone the repository:

    git clone https://github.com/LevelPixel/ComfyUI-LevelPixel-Advanced.git

    to your ComfyUI custom_nodes directory

    The script will then automatically install all custom scripts and nodes.

    It will attempt to use symlinks and junctions to prevent having to copy files and keep them up to date.

  • For uninstallation:
  • Delete the cloned repo in custom_nodes
  • Ensure web/extensions/levelpixeladvanced has also been removed
  • For manual update:
  • Navigate to the cloned repo e.g. custom_nodes/ComfyUI-LevelPixel-Advanced
  • git pull

    • Troubleshooting

    If you have problems running ComfyUI with this node package, check and do the following:

  • The path to Cmake (official distribution, for example, path C:\Program Files\CMake\bin) must be at the very top of the Path list in the System environment variables. Make sure that the path is at the top of the Path in the System variables, and not only in the user variables. If Cmake is not at the top of the list, then some libraries may not compile due to the lack of the current version of the Cmake compiler.
  • The path to the current version of your CUDA must be at the very top of the Path list in the System environment variables (right after Cmake). Make sure that the path is at the top of the Path in the System variables, and not only in the user variables.

    • Here are the paths that should be in Path for CUDA (this is an example, substitute your CUDA version for “12.9”):

    C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.9\libnvvp

    C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.9\bin

  • In System environment variables, add the CMAKE_ARGS variable and set it to the following:\

    • -DGGML_CUDA=on -DCMAKE_GENERATOR_TOOLSET='cuda=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.9'

    Save the changes made to System environment variables.

    Reboot the system.

    If the node package still does not install at startup, try to forcefully remove the remains of the llama-cpp-python package for your python.

    For example, you can run the following command in the python_embeded folder (if you are using the portable version of ComfyUI):

    python.exe -m pip uninstall llama-cpp-python

    After that, run ComfyUI again.

    If you still get errors, restart your PC, this may help (sometimes during installation the cache gets damaged and remains in the computer’s RAM).

    Regarding onnxruntime – usually, various node packages install the onnxruntime library instead of the onnxruntime-gpu package if your computer has a GPU. Some other packages may install onnxruntime-gpu by default. However, due to the strange implementation of the library by its authors, we have a contradiction between onnxruntime-gpu and onnxruntime, which leads to errors when running ComfyUI.

    To fix the error with onnxruntime that you may have, you can use the script at the path (only for Windows!):\

    .\ComfyUI\custom_nodes\ComfyUI-LevelPixel-Advanced\scripts\remove_onnxruntime.bat

    After which you need to run ComfyUI again, our node package should automatically install the correct version of onnxruntime for your system.

    If these tips don’t help – study the logs and the cause of the error, read docs about building llama.cpp https://github.com/ggml-org/llama.cpp/blob/master/docs/build.md, and then talk to some powerful neural network about this error – it will probably help you solve your problem.

    If examining the logs doesn’t help, remove all the components listed below from the system and repeat the steps above.

    What needs to be removed:

  • Uninstall MSVC (uncheck the box in the VS Installer’s Modify mode for Visual Studio 2022 and confirm the changes), restart the PC.
  • Uninstall all CUDA drivers and the NVIDIA toolkit using the standard Windows uninstaller, then restart the PC.
  • Remove all obsolete or unnecessary paths from the Path in System Environment Variables.
  • Download the latest new version of ComfyUI and update it.

    • After that, follow the instructions above again.

    Features

    All nodes Level Pixel Advanced:

    Multimodal Generator node

    Multimodal Generator Advanced – New node on new technology of multimodal neural models based on GGUF.

    Supports Qwen2.5-VL of GGUF format.

    How to use Multimodal Generator node:

  • Download the Qwen 2.5 VL gguf file:\

    • https://huggingface.co/Mungert/Qwen2.5-VL-7B-Instruct-GGUF/tree/main\

    Choose a gguf file without the mmproj in the name

    Example gguf file:\

    https://huggingface.co/Mungert/Qwen2.5-VL-7B-Instruct-GGUF/resolve/main/Qwen2.5-VL-7B-Instruct-q8_0.gguf\

    Copy this file to ComfyUI/models/LLavacheckpoints.

  • Download the Qwen 2.5 VL mmproj file (this is clip model):\

    • https://huggingface.co/Mungert/Qwen2.5-VL-7B-Instruct-GGUF/tree/main\

    Choose a file with mmproj in the name.\

    Example mmproj file:\

    https://huggingface.co/Mungert/Qwen2.5-VL-7B-Instruct-GGUF/resolve/main/Qwen2.5-VL-7B-Instruct-mmproj-bf16.gguf\

    Copy this file to ComfyUI/models/LLavacheckpoints.

  • Run ComfyUI and add to workflow node Multimodal Generator Advanced [LP]. Choose ckpt model and clip, pin image and write prompt.

    • LLaVa nodes

    A node that generates text using the LLM model and CLIP by image and prompt with subsequent unloading of the model from memory.

    Our LLava nodes support the latest LLM and VLM models, and should support future ones (please let us know if any models stop working).

    The core functionality is taken from ComfyUI_VLM_nodes and belongs to its authors.

    Mainly supports Qwen2.5-VL (with clip for images), Mistral (with clip for images) and Llama (with clip for images) models.

    At the moment the nodes are obsolete (but still in support status), instead of them it is supposed to develop “Multimodal Generator nodes” based on llama-mtmd for using Qwen2.5-VL, Bagel and other multimodal neural networks.

    LLM nodes

    A node that generates text using the LLM model with subsequent unloading of the model from memory. Useful in those workflows where there is constant switching between different models and technologies under conditions of insufficient RAM of the video processor.

    Our LLM nodes support the latest LLM and CLIP models, and should support future ones (please let us know if any models stop working).

    The core functionality is taken from ComfyUI_VLM_nodes and belongs to its authors.

    Autotagger nodes based on WD tagger models

    An image autotagger that creates highly relevant tags using fast and ultra-accurate, highly specialized models. More diverse models are planned to be added to the list of models in the future.

    This node allows it to be used in cycles and conditions (in places where it is not necessary to execute this node according to the specified conditions), since it is not a node with mandatory execution.

    The core functionality is taken from ComfyUI-WD14-Tagger and belongs to its authors.

    Image Remove Background

    A more improved version of nodes for removing background for ComfyUI with an extended list of models.

    There are three separate nodes that implement different functionality for different neural models:

  • Image Remove Background (RMBG) – RECOMMENDED! The most powerful node to use, which uses the most powerful model RMBG-2.0 for background removal.
  • Image Remove Background (BiRefNet) – Recommended for super-fast background removal with high quality. Uses the latest generation BiRefNet models that perfectly remove any background in a fraction of a second on the GPU.
  • Image Remove Background (rembg) – Not recommended for normal use and requires additional settings (read below). It differs in that it allows you to include other special types of neural networks to remove the background in certain situations, but the models will not always be the latest generation for this node.

    • To use on GPU, at least CUDA 12.4 (Pytorch cu124) is required, so I recommend upgrading to newer versions of ComfyUI and Pytorch.

    To use Image Remove Background (rembg) effectively on your GPU, you should make sure that you do not have onnxruntime installed together with onnxruntime-gpu. When you run ComfyUI, my package will tell you in the console that you have a conflict between onnxruntime and onnxruntime-gpu.

    Solution:

    Remove onnxruntime, leaving only pure onnxruntime-gpu.

    To do this, do the following:

    Close ComfyUI and run the script at .\ComfyUI\custom_nodes\ComfyUI-LevelPixel-Advanced\scripts\remove_onnxruntime.bat

    The core functionality is taken from RemBG nodes for ComfyUI and from ComfyUI-RMBG and belongs to its authors.

    Recognize Anything Model (RAM++)

    The counterpart to Segment Anything Model (SAM)

    This is an image recognition node for ComfyUI based on the RAM++ model from xinyu1205.

  • https://huggingface.co/xinyu1205/recognize-anything-plus-model
  • https://github.com/xinyu1205/recognize-anything
  • This node outputs a string of tags with all the recognized objects and elements in the image
  • 3 different models.
  • RAM and RAM++ outputs tags in English and Chinese language

    • Furthermore you need to download the RAM, RAM++ and tag2text models and place it in the /ComfyUI/models/rams/ folder or use the ComfyUI-Manager model downloader.

    You can also configure the location in ‘extra\_model\_paths.yaml’ in the ComfyUI folder.

    Update History

    27-07-2025 – Multimodal Generator Advanced stabilized by natively using the latest version of the llama-cpp-python library

    30-05-2025 – Added new node Multimodal Generation Advanced for neural models of multimodal type (for example, for Qwen2.5-VL)

    The license for this package has been changed from Apache 2.0 to GNU GPLv3

    Credits

    ComfyUI/ComfyUI – A powerful and modular stable diffusion GUI.

    VLM nodes for ComfyUI/ComfyUI_VLM_nodes – Best VLM nodes for ComfyUI.

    WD autotagger node for ComfyUI/ComfyUI-WD14-Tagger – Source ComfyUI nodes for WD autotagger

    RAM node for ComfyUI/ComfyUI-Hangover-Recognize_Anything – Source ComfyUI nodes for RAM (source repository is archived, but we will continue to support RAM nodes)

    RemBG software package/rembg – Software to remove background for any object in the picture.

    RMBG nodes for ComfyUI/ComfyUI-RMBG – Thanks for the awesome code and implementation of using BiRefNet and RMBG-2.0 models in very convenient and customizable nodes.

    License

    Copyright (c) 2024-present Level Pixel

    Licensed under GNU GPLv3