🤖大炮API模型配置@炮老师的小课堂

This node does almost nothing by itself, and that's the point. DapaoAPILLMConfigNode is the configuration brick for the multi-turn chat workbench in ComfyUI-dapaoAPI - the node that holds the API key, picks the LLM, and sets the timeout and context budget for every chat call the workbench makes. You never wire its output into a sampler. You wire it into the chat node, and it makes the difference between "chat works" and "chat works with a model I actually picked."

The whole thing is a pipe: this config node feeds DapaoAPIMultiTurnChatNode, which optionally also takes a settings node and a Skill loader. Think of the config as the power supply - you need exactly one, and everything downstream draws from it.

What it actually sets

Four fields, none of them glamorous:

• 🔑 API密钥 - your dapaoAI key. The tooltip is worth quoting: it's only ever sent to https://api.dapaoai.com, and it never gets written into chat history or Skills. That's a real privacy-consciousness win for this pack, though remember the key still lives in your saved workflow JSON.
• 🤖 LLM模型 - the mapping model. The menu is a who's-who of 2026 frontier models: gpt-5.5, the gpt-5.6 luna/terra/sol trio, claude-fable-5 and the claude-opus/claude-sonnet 4.x–5.x line, plus gemini-3.5-flash, gemini-3.5-flash-lite and the default gemini-3.7-flash. You're not running any of these locally - dapaoAI relays them.
• 📚 上下文上限 - context budget in tokens. Default 0 means "use the node's built-in conservative value." It only drives local trimming and the little token-gauge ring in the chat UI; the model's real limit is whatever the API enforces.
• ⌛ 请求超时 - seconds, default 300. Long Skill runs with video reference analysis can need more; the chat node's settings govern most of the tuning.

How it works

When the chat node runs, it reads this config, builds an OpenAI-compatible chat request against api.dapaoai.com/v1/chat/completions, and sends your message plus whatever history and Skill context survive trimming. The node itself just validates the model name and returns a tiny DAPAO_API_LLM_CONFIG object on its single 🤖API模型 output. If the model string isn't in the option list it raises a clear error - the author guards against typos, which is more than most API packs do.

One thing worth knowing before you pick a model: not every entry in that list is multimodal. Wire an image into a text-only model and the chat node stops you with "该模型不支持图片输入，请切换多模态模型" (switch to a multimodal model). The gemini-3.x and gpt-5.x entries are your safe bets for image chat.

Install and first run

The pack ships as ComfyUI-dapaoAPI. Easiest path: ComfyUI Manager → search "dapaoAPI" → install → restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/paolaoshi/ComfyUI-dapaoAPI.git
cd ComfyUI-dapaoAPI
pip install -r requirements.txt

Then you need a dapaoAI key: register at api.dapaoai.com, buy a redeem code in the wallet, apply for an API key in the default group, and paste it here. The README's registration links carry the author's referral code, so that's the business model, but the API itself is straight-up pay-per-call.

Troubleshooting

• 401 / key rejected - the key is only valid for the api.dapaoai.com endpoint and the "default" group. If you used a key from another group or another provider's relay, it won't work here.
• Model not in the list - you can't free-type one; the enum is the contract. If dapaoAI adds a model before this pack updates, you're stuck with the current menu.
• Seed does nothing - there's no seed field here, but if you're coming from the chat node: its seed only controls ComfyUI caching, never the LLM. Don't hunt for a reproducibility knob where the API doesn't provide one.

Chinese-only menus, like the rest of this pack - if you don't read the labels, your tab key and the defaults will get you most of the way, because the defaults are genuinely sensible.