ComfyUI-QwenVL

Why we included this project

ComfyUI users who want to actually read the frames in their pipelines rather than just produce them can drop the Qwen-VL vision-language models straight into the node graph. The nodes handle image understanding, video analysis, and text generation, and the GGUF backends keep quantized checkpoints runnable on GPUs with limited VRAM. Recent releases added a video auto-scaling step that computes a per-frame token budget, so context-overflow and out-of-memory failures become far less common, plus a config file for registering your own Hugging Face or GGUF models. There is also a Python API and CLI that expose the same engine outside the ComfyUI canvas, which helps when your review-and-edit steps live in other tools. For teams running multi-stage generation and verification pipelines, that means adding a vision-language check step without leaving the environment they already use.