Base Config

"Base Config" is the node every workflow in this pack starts from - and it does absolutely nothing by itself. It never calls an API, never makes a request, never generates a pixel. It's a data node: three text fields in, one base_config object out, and that object gets threaded through Chat Params or Image Params into the execution node that actually does the work. If you're debugging a pack like this and it feels like the config node is "dead," that's correct - it's supposed to be.

The three fields that matter

• api_base - defaults to https://your-litellm-server.com/v1, which is a placeholder, not a working endpoint. This is the URL of your LiteLLM proxy. If you haven't got one running, this is where the pack stops working. This is the field that separates the LiteLLM path from the OpenRouter path in this same pack: OpenRouter's Base Config points at OpenRouter instead of your own server.
• api_key - your LiteLLM server's credential. Note that this node doesn't validate it; the key just gets carried along and used at request time.
• model - the model name your LiteLLM server recognizes, conventionally prefixed with the provider, e.g. gemini/gemini-3-pro-image-preview. Get this string wrong and the request will fail at the server with a model-not-found error.

Why a separate config node?

Because it's reusable. Put one Base Config in the graph, then branch it into both a Chat Params (for chat) and an Image Params (for image generation). Change the model or key once, and every downstream node picks it up. It's the same "carry config on one wire" pattern the ecosystem's bigger packs use, just scoped small.

Installing it

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

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

There's no requirements.txt, no pip install, and no model files to download - this is a thin API-client pack using only Python's standard urllib plus what ComfyUI already ships. The one thing you genuinely have to provide is a running LiteLLM server and a key for it; if that sounds like more than you signed up for, the Base Config (OpenRouter) variant in the same pack gets you the same graph shape with nothing to host.

Gotchas

The output is LLM_BASE_CONFIG, and the input types matter: Chat Params expects exactly this type, and Image Params expects it too. Keep the key handling in mind as well - whatever you type into the api_key field is stored in the workflow JSON, so a shared workflow file carries your credential with it. For a one-line config node, that's the one thing actually worth being careful about.