Base Config (OpenRouter)

"Base Config (OpenRouter)" is the friendliest door into this whole pack, because it's the only node where "the setup is done" actually feels true. Two required fields, no server to run, no proxy to babysit. Paste an OpenRouter key, pick a model, and the rest of the OpenRouter nodes in the pack will happily talk to Gemini for you.

Like its LiteLLM sibling, this node is pure data - it makes no API calls and needs none. It exists so one configuration can be built once and branched into both Chat Params (OpenRouter) and Image Params (OpenRouter). That's the whole job, and it does it cleanly.

The fields that matter

• api_key - the only thing that's truly required. Grab it from openrouter.ai/keys. It isn't validated here; it's just carried along until a request actually happens.
• model - defaults to google/gemini-3-pro-image-preview. OpenRouter uses the provider/model naming, so if you want the faster flash variant you'd type google/gemini-2.5-flash-image-preview. You can also set any model OpenRouter serves - that's part of the appeal: one key, many providers.

Then three optional fields, worth a paragraph because they're easy to ignore:

• api_base - defaults to https://openrouter.ai/api/v1. You basically never touch this unless you're pointing at a compatible proxy.
• site_url and site_name - OpenRouter's app-attribution fields, sent as HTTP-Referer and X-Title headers on every request. Completely optional, but filling them in is the polite way to make your usage show up properly on your OpenRouter dashboard. If they're empty, the node just drops the headers.

Installing it

# ComfyUI Manager: search "Gemini LiteLLM", Install, restart

# or:
cd ComfyUI/custom_nodes
git clone https://github.com/ZUENS2020/ComfyUI-Gemini-LiteLLM

No pip install, no model downloads, no GPU footprint - the pack is stdlib urllib plus the numpy/torch/Pillow ComfyUI already ships. The one real requirement is an OpenRouter key and (sooner than you think) credit on the account, because Gemini image generation bills per request.

Gotchas

Two things. First, the output type is OR_BASE_CONFIG, and the OpenRouter params nodes expect exactly that - don't try to feed this into the LiteLLM side's Chat Params, the types won't match and ComfyUI will refuse the connection. Second, and this is the real one: your key sits in the workflow JSON. It's trivially easy to accidentally share a workflow with a live key embedded. If you're going to distribute a workflow, blank the key field first - and for that matter, don't expose your ComfyUI instance to the internet unauthenticated, since its API executes arbitrary workflows by design.