This repository introduces ComfyUI nodes designed to integrate the latest vision-language and text-only checkpoints from the Qwen model family. It supports both Qwen3 VL and Qwen2.5 VL for multimodal reasoning, as well as text-only operations using Qwen2.5 models for prompt generation.
- Provides multimodal capabilities with Qwen3 VL and Qwen2.5 VL models, allowing for diverse input types.
- Supports text-only workflows through Qwen2.5 instruct models, ensuring flexibility in generation tasks.
- Offers customizable parameters, including temperature and token count, enhancing user control over the output.
Context
This tool serves as a wrapper for ComfyUI, enabling users to leverage advanced Qwen models for both multimodal and text-only tasks. Its primary aim is to enhance the functionality of ComfyUI by allowing seamless integration of sophisticated AI models that can process and generate content based on both visual and textual inputs.
Key Features & Benefits
The primary features include nodes for multimodal generation using Qwen3 VL and Qwen2.5 VL, which can handle images or videos along with text prompts. Additionally, the text-only Qwen2 node allows users to generate text-based outputs using the Qwen2.5 instruct models, providing a versatile toolkit for various creative and analytical applications.
Advanced Functionalities
The nodes offer advanced settings such as temperature control, maximum token count, and quantization options (none, 4-bit, or 8-bit) to optimize memory usage. Users can also enable model caching between runs by setting keep_model_loaded to True, which enhances performance by reducing loading times in repetitive tasks.
Practical Benefits
This tool significantly improves workflow efficiency within ComfyUI by providing robust options for generating complex outputs. It allows users to maintain greater control over the generation process and output quality, facilitating a smoother and more intuitive creative experience.
Credits/Acknowledgments
The repository is authored by Alex Cong and is openly available for use under the relevant licensing terms, contributing to the growing ecosystem of AI art tools and workflows.