ComfyUI-PDiD-LLM-Nodes

<div align="center"> <img src="https://img.shields.io/badge/Requires-Python 3.10%2B-blue" alt="Python"> <img src="https://img.shields.io/badge/Version-0.1.0-green" alt="Version"> </div>

Intro

Some nodes created to do job with LLMs and VLMs (Such as caption, draw bbox, simple chat).

There are alternatives, why created this node?

Some nodes rely on numerous libraries. Due to the project's non-modular structure, missing just one dependency can make all nodes inoperable.

Additionally, certain libraries required by these nodes—even ones I don't use—conflict with my local virtual environment. For example, the project depends on aisuite, causing dependency mismatches during ComfyUI deployment and preventing tools like uv from managing these dependencies properly.

Nodes

API LLM Loader

This node load LLMs or VLMs for use. Supports many clients.

• Openai-compatible chat client
• Openai-compatible responses client (Which OpenAI strongly recommend)
• Ollama client
• Mistralai client.

I cannot make sure that Anthropic client works well because I cannot reach any Anthropic services in my country. But it works on modelscope on Anthropic client.

Extra Parameters

This node controls some of the extra parameters of a model. Only works on Openai-compatible client (chat and resposes).

Thinking enabled or disabled only works on some models on volcengine. (Seed 1.6 series.) Thinking effort works on some newest thinking models so far. Such as GPT-5, GPT-5.1, Seed-1.6 and so on.

API Chat

A chat node for LLMs and VLMs.

unload_model_after_chat works only on Ollama. llama.cpp will be supported when they finished their jobs.

top_k works only on Ollama.

Generate BBOXes

A node to generate bboxes from an image.

The bbox is competitable with some bbox nodes on ComfyUI_LayerStyle. You can use it with segment anything.

Captioner

A node to generate caption from a image. (Maxium 10 sentences.)

You can specify the language to give a corresponding caption.

About num_max_sentences:

• Use 2 for CLIP-based model. (Such as SD-1.5 SDXL or CLIP part of the ClipTextEncodeFlux)
• Use 4 for T5-based model or Qwen-VL model. (Such as t5xxl part of the ClipTextEncodeFlux or Qwen-Image)
• Use 4 and above for Qwen-Image.

If you want to use it to make caption for training or finetuning. These are also accepted.

Install

Make sure you are using ComfyUI Version 0.3.51 and above.

Just install it by searching ComfyUI-PDiD-LLM-Nodes in ComfyUI-Manager

Contributing

Any contributions on this repostory are welcome. (Especially about Anthropic nodes.)

TODO

Clients

• [x] OpenAI client support.
• [x] Anthropic client support.
• [x] Ollama client support.
• [x] Mistralai client support.
• [ ] ComfyUI Clip client support.

Functions

• [x] Chat
• [x] Generate bbox
• [x] Generate Caption
• [ ] Custom context.

Licence

MIT