Chat Params

"Chat Params" is the node that turns your Base Config into a complete chat configuration. It takes the base_config from Base Config, bolts two knobs onto it, and hands the result to the Chat node as an LLM_CHAT_CONFIG. Like every config node in this pack, it makes no API calls - it's just the middle of the pipeline: Base Config → Chat Params → Chat → text.

The two inputs that matter

• temperature - a float from 0 to 2, default 0.7. This is your creativity dial, and this pack's chat side is notably more permissive than its image side (chat allows up to 2, image caps at 1). Near 0, Gemini gives you the same conservative answer every time - great for extracting a fixed format. Near 2, it gets sloppy and interesting. For prompt rewriting, 0.7 is a sane starting point; for anything where you need consistency, drop toward 0.
• max_tokens - 1 to 128,000, default 2000. This caps how long the reply can be, not how smart the model is. For a caption or a rewritten prompt, 2000 is generous. If you're asking for long structured output - say, a full JSON dump - bump it and you'll stop getting truncated responses.

That's genuinely the whole node. It merges those two values with whatever your Base Config carried (api_base, key, model) and passes the bundle along. The only output is config of type LLM_CHAT_CONFIG, which wires straight into the Chat node's config input.

Installing it

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

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

Same pack, same zero-dependency story: no requirements.txt, no pip install, no model files. It's stdlib urllib plus the numpy/torch/Pillow that ComfyUI already ships.

Gotchas

The type system is the only real trap. A config coming out of this node is an LLM_CHAT_CONFIG, and it's meant for the Chat node. If you feed it into the LiteLLM Image node instead, that node checks for a flag only Image Params sets - you'll get a Gemini config required error and a confused afternoon. Use Chat Params for chat and Image Params for images, and the pipeline stays boring in the good way.