Auto Install All Custom Nodes Requirements_txt

Every custom node pack ships a requirements.txt, and if you've ever set up a fresh cloud ComfyUI instance you know the drill: clone ten packs, then run ten separate pip installs, hoping you caught all the dependencies before some node redlines on import. Auto Install All Custom Nodes Requirements_txt tries to automate exactly that. Give it your custom_nodes folder, run it, and it walks the whole tree, finds every requirements.txt, and pip-installs them one after another.

It's a genuinely useful idea with one huge caveat attached: it's the "scorched earth" version of dependency management. ComfyUI Manager already does this per-node in a controlled way (its "Install Missing Custom Nodes" is the sane version of the same job). This node does it in one loop, against whatever Python environment ComfyUI is running in, with no dry run and no undo.

How it works

The input is a single string, custom_nodes_path - the absolute path to your custom_nodes directory. The node then:

1. Recursively walks that directory for every file named requirements.txt.
2. Skips anything under a folder called ComfyUI-Manager (a deliberate, sensible exception - it doesn't want to reinstall the manager's own deps).
3. For each remaining file, runs pip install -r <file> using the same Python interpreter that's running ComfyUI (sys.executable).
4. Returns a status string concatenating every command's stdout, or the error text when one fails, so you can see which installs blew up.

That's it. No dependency resolution across packs, no "skip if already installed" check beyond pip's own behavior, no rollback. Each pack's requirements are installed into the shared environment, in arbitrary walk order.

Where this shines, and where it burns

The good case: a brand-new cloud box where you just cloned a big workflow's worth of custom nodes and want everything installed before you load the graph. One node run, all requirements in. For a provisioning script that's genuinely handy.

The bad case: your daily-driver local install, where every pack's requirements are already resolved and working. Running this can upgrade a transitive dependency that three other nodes quietly depend on, and now your working setup is broken in ways that are miserable to untangle. It also installs every pack's requirements, including packs you don't use - more surface area, more version drift. This is the node I'd only run inside a throwaway environment or a container.

Install

Same pack-wide route: ComfyUI Manager → search ComfyUI_AutoDownloadModels → install, or

cd ComfyUI/custom_nodes
git clone https://github.com/AIExplorer25/ComfyUI_AutoDownloadModels
cd ComfyUI_AutoDownloadModels
pip install -r requirements.txt

then restart. The pack's own requirements (huggingface_hub, hf_transfer, rapidfuzz) are tiny and install fast.

Troubleshooting notes

• Path matters. Point it at the directory that contains your custom node folders (e.g. ComfyUI/custom_nodes), not an individual node folder, or you'll only hit that one pack.
• It skips ComfyUI-Manager by design. If you're missing something that lives in the manager's own requirements, this won't fix it - install those separately.
• Slow runs are normal. Walking a folder tree and pip-installing dozens of files takes minutes; the node blocks the queue until every install finishes. Don't panic if ComfyUI looks frozen.
• The output is a wall of text. The status string includes every install's full output. If a pack failed, the error appears in there - but expect to scroll.

Bottom line: this node is a convenience for disposable setups and a hazard elsewhere. Use ComfyUI Manager for your real install; reach for this only when you're provisioning something fresh and want it over with.