😀 LLMs_Chat

Straight answer first: 😀 LLMs_Chat doesn't exist in the current ComfyUI-LLMs pack. This page exists because the node used to, and enough old workflows still reference it that you'll hit a "missing node" error and go looking. What you actually have is a workflow built against a pre-1.0.0 version of the pack, and the fix takes about a minute.

What happened

The pack's changelog is blunt about it: version 1.0.0 (January 2024) "restructured the entire project" and replaced the old separately-registered nodes with two cleaned-up ones - LLMs Chat (shown as "🤖 LLMs Chat | 智能对话") and LLMs Vision Unified. The old 😀 LLMs_Chat was the chat node of its era, and the current LLMs Chat is functionally the same thing: same config, same OpenAI-compatible API calls, same prompt-generator default system prompt. The class literally still exists in the code, just registered under a new name.

So if your workflow complains it can't find 😀 LLMs_Chat, you don't need a different node or a different model. Install the current pack, delete the dead node, drop in LLMs Chat, and re-wire the two prompt strings. It's a rename, not a redesign.

What the old node looked like

Its inputs are basically identical to today's, which is why the migration is painless:

• api and model - dropdowns built from your settings.yaml config. The old model list shows the pack's original sample lineup: glm-4, chatglm_pro, ERNIE-Bot-4, qwen-turbo-internet, plus the GPT-3.5/4 family.
• system_prompt - defaults to the "act as prompt generator…" instruction, still the pack's signature move: turn a rough idea into a detailed image prompt.
• user_prompt, temperature (default 0.99), and optional top_p.

Output was gpt_response, a STRING - same as today. The whole node was a thin wrapper around an OpenAI-compatible chat completion, and it still is.

Installing the replacement

Same pack, current version:

cd ComfyUI/custom_nodes
git clone https://github.com/leoleelxh/ComfyUI-LLMs
cd ComfyUI-LLMs
pip install -r requirements.txt
cp settings.yaml.sample settings.yaml

Note the README's own clone URL has a typo (leoleexh instead of leoleelxh) - use the one above or ComfyUI Manager, which searches "ComfyUI-LLMs". Then set your API key and endpoint in settings.yaml, restart ComfyUI, and the node will be waiting under the "🌟 LLMs" category as LLMs Chat.

One habit worth keeping from this episode: old workflows rot faster than the nodes they reference. When a downloaded workflow shows a missing node with an emoji name you can't find, check whether the pack it came from renamed things before you hunt for a mysterious fork.