Set DeepSeek LLM Service Connector 🐑

This node is the one-step way to hook DeepSeek up to the rest of the MieNodes pack. It's a preconfigured connector: DeepSeek's endpoint is already baked in, so you drop the node, paste your key (or leave it in a config file), pick a model, and you've got a live LLM connection ready to feed the prompt-generator nodes. No base-URL fiddling like you'd do with the general connector.

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

Why you'd use it

The MieNodes prompt tools - the Kontext generator, the Flux2 generator, plain CallLLMService - all need an LLM to actually do the thinking. This node produces that connection, specifically pointed at DeepSeek.

DeepSeek is a reasonable default for prompt work: cheap, fast, strong at following instructions, and it doesn't need to be multimodal for text-in/text-out prompt rewriting. If you're using these nodes to expand a rough idea into a full prompt, translate, or restructure - the exact workflow the KB's prompt-engineering doc describes as "having an LLM write the instruction your LLM-encoded model reads" - DeepSeek is a fine engine to hang off it. The dedicated node just saves you from looking up the endpoint URL every time.

The inputs that matter

Required:

• api_token - your DeepSeek API key (sk-...). Can be left empty if you store it in the config file.
• model_select - a dropdown: deepseek-v4-pro, deepseek-v4-flash (the default), or Custom. Flash is the cheap-and-quick one; Pro when you want stronger reasoning. Pick Custom and fill custom_model if DeepSeek ships something newer that isn't in the list yet.

Optional plumbing, shared with every connector in the pack:

• custom_model - the exact model id, used only when model_select is Custom.
• config_file (default mie_llm_keys.json), config_key (default deepseek), and prefer_local_config (default true). With local config preferred, the JSON file's deepseek entry wins and the api_token field is just a fallback.

Output is llm_service_connector - feed it straight into any node that has an llm_service_connector input.

Installing it

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

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

then restart. Nothing to download beyond the pack itself; the model lives on DeepSeek's servers. Nodes appear under the 🐑 MieNodes menu.

Common issues

Auth failures. Check whether your key is in the node or in mie_llm_keys.json, because by default the file wins. If you typed a key into api_token but there's a stale one under the deepseek entry in the JSON, the JSON one gets used and you'll be confused. Either keep the key in one place, or flip prefer_local_config to false to force the node field.

Model not found. If deepseek-v4-flash/-pro ever stop resolving because DeepSeek renamed things, switch model_select to Custom and paste the current id into custom_model. The dropdown is a convenience, not a hard constraint.

Keys leak through saved workflows. Typing your key into the node means it can end up in an exported workflow JSON. The config-file route (copy mie_llm_keys.json.example, fill the deepseek entry, leave the node field blank) keeps it out. The file is deliberately not committed by the pack.