Qwen3-ASR ComfyUI integration
Hi Qwen team 👋
Thank you for releasing Qwen3-ASR, it’s an excellent open-source ASR model with great quality and long-audio support.
We’ve built a ComfyUI custom node based on Qwen3-ASR to make speech-to-text and subtitle generation easier to use in visual, node-based workflows. It supports local model caching, long audio chunking, and optional timestamped subtitles.
GitHub repo:
https://github.com/1038lab/ComfyUI-QwenASR
Big thanks again for your work and for supporting the open-source community, it enables many downstream tools like this one
Chunking long audio while keeping the timestamps honest across boundaries is the part that usually breaks first, since a word split across two windows tends to get counted twice or dropped entirely. Do you overlap the windows and reconcile, or cut on detected silence?