Environment Setup

Most ComfyUI "setup" nodes are documentation with a button on it. This one is narrower and more committed: it verifies - and optionally downloads - the exact three files the FLUX.2 Klein workflow in this pack expects, into the right ComfyUI folders, with the right names.

If you're picking up a shared Klein workflow and don't want to hunt down which of the four Klein checkpoints it wants and which text encoder pairs with it, this is a genuine shortcut. If you're running anything else, it does nothing for you.

What it provisions

Three files, hardcoded:

| Model | Goes to | |---|---| | FLUX.2 Klein Base 9B FP8 | models/diffusion_models | | Qwen text encoder | models/text_encoders | | FLUX.2 VAE | models/vae |

Worth being straight about what that is. Klein Base is the undistilled 9B - the one BFL points at training, not the four-step daily driver most people generate with. Klein shipped as four checkpoints (4B and 9B, each in distilled and base forms), and mixing them up is the most common confusion about the release. So this node is provisioning a training-capable base rig plus the encoder and VAE that go around it. If you already have the distilled 9B and a working text encoder, you probably don't need it.

Yes, the upstream text-encoder repo is really spelled vae-text-encorder-for-flux-klein-9b. That typo is BFL's/Comfy-Org's, not the pack's.

Inputs and outputs

• verify_installation (BOOLEAN, default true) - check what's on disk and report, without fetching anything.
• download_missing_models (BOOLEAN, default false) - when true, actually download what's missing and wait for it to finish.
• status (STRING) - per-model lines like FLUX.2 Klein Base 9B FP8 (flux-2-klein-base-9b-fp8.safetensors): already exists ✅, or missing ❌, or failed - ….
• ready (BOOLEAN) - false if anything is missing or errored.

It's an output node, so it runs when you queue the prompt rather than only when something downstream asks for it. That's deliberate: you drop it, queue, read the console output, and ignore it afterwards.

About the token

Downloads from Hugging Face need auth, and this is where the node is unusually well-behaved. There is no token field on the node. It reads HF_TOKEN, then HUGGINGFACE_TOKEN, then the shared server store written by this pack's Hugging Face Token node. Nothing gets serialised into your workflow JSON, and the returned status text has any token string scrubbed out of it before you see it.

The reasoning is sound and worth internalising generally: ComfyUI saves STRING widget values into the workflow file. A token widget is a token that ships to everyone you share the graph with. If a pack offers you a password box, that's usually a liability, not a feature.

Prefer env vars on a shared or hosted box:

export HF_TOKEN=hf_xxxxxxxxxxxxxxxx

Install

Manager → search Orel's Nodes → Install → restart → hard-refresh.

cd ComfyUI/custom_nodes
git clone https://github.com/DarkOrel/Orels-Nodes.git orels-nodes
python -m pip install -r orels-nodes/requirements.txt

That requirements file (huggingface_hub, filelock) is what makes the download path work. Skip it and the node still loads - it just reports Downloader unavailable: install huggingface_hub and filelock, which is a clear enough failure to act on.

Gotchas

Big download, no progress bar. The pack's own model-download node has a chunked progress UI; this setup node reports a line of text and, with downloads enabled, waits for the job to leave the "Downloading" state before returning. On a slow connection expect a long, quiet queue. Budget tens of GB.

Set both toggles deliberately. verify_installation on with download_missing_models off is the safe default and the one you want on someone else's workflow. Download off/verify off gives you No checks requested. and nothing else - a confusing thing to see if you flipped a switch by accident.

It won't fix your other missing nodes. It knows about three files. The rest of the workflow's dependencies are still the usual ComfyUI lottery.