🧙hua_gradio Random Seed

Hua_gradio_Seed (🧙hua_gradio Random Seed) looks like the most boring node in the pack - an int in, an int out - but the README singles it out with the only real warning in the whole document: don't build a packaged workflow without it. The author's claim, in plain terms: when ComfyUI receives two byte-identical API JSON payloads, it can decide the graph hasn't changed and won't actually run - the README describes it as refusing to infer and effectively hanging. The seed node exists to make every payload different so the run always fires.

Mechanically it's a pass-through. You give it a seed (INT, 0 to 2^64–1, default 0) and a name, and it returns that seed unchanged, plus a show_help string that's just a hardcoded link to the author's repo. The real work happens on the Gradio side: when you launch the pack's web UI, it inspects your saved workflow, finds this node, and shows seed controls - a mode dropdown (随机/random, 递增/increment, 递减/decrement, 固定/fixed) and a fixed-seed field. Each run, the frontend writes a fresh seed back into this node before POSTing the prompt to ComfyUI on port 8188. In "random" mode that's exactly the variability that stops the identical-payload problem from biting you.

Wire the seed INT output into your KSampler's seed input like any other seed node. The show_help output is cosmetic; you can ignore it, or use it as a route to the author's other work.

Install: ComfyUI Manager → "ComfyUI_to_webui", or

cd ComfyUI/custom_nodes
git clone https://github.com/kungful/ComfyUI_to_webui.git

restart, done. No model files. The pack's Gradio app then runs on http://127.0.0.1:7861.

Two gotchas worth knowing. First, if you run the workflow from inside ComfyUI's own UI rather than the Gradio front, the seed is just whatever you typed - the "random" behavior only kicks in via the frontend. Second, there's a duplicate class definition in the pack's source (two Hua_gradio_Seed classes, later one wins); it doesn't affect usage, but it tells you this pack is a personal project evolving quickly, not a polished commercial product. Keep it in your packaged graphs anyway - it's the difference between a webui that generates and one that stares at you silently.