Set MiniMax Token Plan LLM Service Connector 🐑

This node connects MiniMax to the MieNodes pack - specifically the Token Plan (a.k.a. Coding Plan) tier, the fixed-fee subscription whose keys start with sk-cp-.... If you're on MiniMax's standard pay-as-you-go Open Platform instead (JWT-style eyJ... keys), you want the plain SetMiniMaxLLMServiceConnector node, not this one. The distinction is real: the two tiers hit different endpoints.

It's part of ComfyUI-MieNodes (ComfyUI_MieNodes), MieMieeeee's utility pack, under LLM Service Config.

Why you'd use it

Like every connector in this pack, its job is to produce an llm_service_connector that the prompt-generator nodes can call. This one aims it at MiniMax's Token Plan endpoint so your flat-rate subscription is what pays for the requests.

MiniMax's M-series models are solid at instruction-following, which is what prompt rewriting actually needs - the KB's prompt-engineering doc is blunt that on 2026-era LLM-encoded diffusion models your prompt is read as a message, so an LLM that reliably restructures messages is exactly the right tool to sit in front of them. If you're already paying for the Token Plan, routing your ComfyUI prompt work through it is close to free at the margin.

There's a small compatibility detail the README flags: the MiniMax connectors strip detail: "auto" from any image_url parts before sending, because MiniMax rejects that value with an HTTP 400. That's baked into the node, so multimodal prompt work (feeding a reference image to a vision model) just works instead of erroring - one of those quiet reasons a vendor-specific connector beats the generic one.

The inputs that matter

Required:

• api_token - your Token Plan key (sk-cp-...). Optional if it lives in the config file.
• model_select - dropdown led by MiniMax-M3 (the default), then MiniMax-M2.7, -M2.7-highspeed, MiniMax-M2.5, -M2.5-highspeed, MiniMax-M2.1, -M2.1-highspeed, MiniMax-M2, and Custom. The -highspeed variants trade a little quality for latency; handy when you're batching a lot of prompt rewrites.

Optional plumbing (same across the pack's connectors):

• custom_model - used only when model_select is Custom.
• config_file (default mie_llm_keys.json), config_key (default minimax), prefer_local_config (default true; the JSON entry wins over the node field).

Output: llm_service_connector → wire into CallLLMService or any generator node.

Installing it

ComfyUI Manager → search ComfyUI-MieNodes → install → restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/MieMieeeee/ComfyUI-MieNodes

then restart. No local model download; MiniMax runs remotely. Nodes appear under the 🐑 MieNodes menu.

Common issues

Wrong-tier key. A sk-cp-... key belongs here; an eyJ... JWT belongs on the standard connector. Cross them and auth fails. This is the single most common trip-up with the MiniMax nodes.

Config file overrides your typed key. The minimax entry in mie_llm_keys.json beats the api_token field by default. Keep the key in one place or set prefer_local_config to false.

A renamed model. If a listed MiniMax-Mx id stops resolving, switch to Custom and paste the current id into custom_model.