WaveSpeed Client

The honest headline

This node does almost nothing, and you will almost never need to place it. It exists so a workflow can hand a WaveSpeed API key to downstream nodes. Every WaveSpeed node in the ERPK Collection (Seedream, Qwen Image, the video nodes) can run completely standalone if your key lives in ComfyUI's settings - the client only matters when you want to pass a key explicitly, per workflow.

Think of it as the login screen you walk past because you're already signed in.

How it works

WaveSpeedAIAPIClient builds the WAVESPEED_AI_API_CLIENT object that other ERPK WaveSpeed nodes accept on their client input. Give it a key, it hands back a client. That's the whole job.

Where the key actually comes from matters more, because the node checks three places in priority order:

1. ComfyUI Settings - right-click the canvas → ERPK Settings, or Settings → ERPK → API Keys. This is the way to do it. Keys are stored per-user, never saved into workflow JSON, so you can share a workflow without leaking anything.
2. The node's own api_key widget.
3. config.ini - the pack writes a stub config file on first run; you can drop a key in there as a last resort.

If all three are empty, the node raises a ValueError that helpfully lists those three options. That "No API key found" error is the classic first-run failure with this pack, and the fix is always option 1.

Its only input is api_key (optional, empty by default) and its only output is client, which wires into the client input of any WaveSpeed node.

Installing the pack

ERPK Collection is a single monorepo that ships all the WaveSpeed, Gemini, Claude, OpenAI and Grok nodes at once. The easy route:

1. Open ComfyUI Manager → Install Custom Nodes.
2. Search erpk, find ERPK Custom Nodes, install, restart.

Manual, if you prefer:

cd /path/to/ComfyUI/custom_nodes/
git clone https://github.com/eRepublik-Labs/comfyui-nodes-erpk.git erpk
cd erpk
pip install -r requirements.txt

Then restart ComfyUI. There are no model downloads and no VRAM requirement - everything these nodes do happens on WaveSpeed's servers, not your GPU. The dependencies (requests, Pillow, pydantic, cryptography, plus the OpenAI/Anthropic/Gemini/xAI SDKs) are thin. Old workflows referencing the pre-V3 node ID "WaveSpeed Custom Client" get migrated to this node automatically on load.

Gotchas worth knowing

• The key setup is the whole game. Skip the client node entirely, set the key once in ERPK Settings, and every WaveSpeed node just works.
• WaveSpeed is a pay-per-image API. You'll need a key from wavespeed.ai, and there's a minimum top-up (around $10) - the community grumbles about that, and it's worth knowing before you sign up for a couple of test renders.
• It's not the official WaveSpeed pack. The real WaveSpeed maintains its own ComfyUI-WaveSpeed repo. This is ERPK's independent integration, so don't file bugs about it with WaveSpeed - and the API behavior can lag theirs slightly.

Keep this node in your back pocket for the workflows you hand to someone else or run on a shared machine. For your own daily use, set the key once and never think about the client again.