ComfyUI LLM

Custom ComfyUI Nodes for interacting with any OpenAI-compatible API server (/v1/... endpoints).

Integrate the power of LLMs into ComfyUI workflows easily — text generation, vision tasks, multi-turn conversations, and structured output.

Originally created by Stav Sapir as comfyui-ollama. This fork migrates to OpenAI-compatible API endpoints with broader server support.

Supported Servers

Any server that exposes an OpenAI-compatible /v1/... API:

Server	Base URL Example	Notes
Ollama (built-in /v1 proxy)	http://127.0.0.1:11434	No API key needed by default
LM Studio	http://127.0.0.1:1234	Local inference
vLLM	http://127.0.0.1:8000	High-throughput serving
OpenAI	https://api.openai.com	Requires API key
LiteLLM proxy	http://127.0.0.1:4000	Unified proxy for many providers

Installation

Install the custom nodes

Use ComfyUI's built-in extension manager to install the nodes. Search for comfyui-llama by Clusters.

Or

If you prefer ComfyUI-Manager, search for comfyui-llama and select the one by clusters

Or

1. git clone into the custom_nodes folder inside your ComfyUI installation or download as zip and unzip the contents to custom_nodes/comfyui-llama.
2. pip install -r requirements.txt
3. Start/restart ComfyUI

Nodes

LLM Generate

A node for text generation via OpenAI-compatible /v1/chat/completions. Supports system prompts, vision tasks (images), multi-turn context, and JSON structured output.

Inputs:

• LLM Connectivity (optional)
• LLM Options (optional)
• images (optional) — for vision-capable models
• context (optional) — context from another LLM Generate node
• meta (optional) — passes connectivity and options from a previous LLM Generate node

Notes:

• For this node to be operational, LLM Connectivity or meta must be provided.
• If images are used in a meta chain, all images need to be passed as well to subsequent nodes.

LLM Chat

A node for conversational interaction using /v1/chat/completions. It manages full conversation history natively and supports chained sequences of chat nodes.

• Functionality: Designed specifically for multi-turn conversations.
• Key Features:
  • Conversation history is handled natively within the node instance.
  • Dedicated history outputs for chaining multiple chat nodes.
  • Option to reset the current conversation history on demand.

Inputs:

• LLM Connectivity (optional)
• LLM Options (optional)
• images (optional) — for vision-capable models
• meta (optional) — passes connectivity and options from another LLM Chat node
• history (optional) — passes history ID from another LLM Chat node

Outputs:

• result: The generated text
• thinking: Extracted thinking/reasoning text (if model outputs <think> tags)
• meta: Metadata to pass to the next LLM Chat node
• history: History ID to pass to the next LLM Chat node

LLM Connectivity

Configures the connection to your API server.

Field	Description
url	Base URL of the OpenAI-compatible API server (e.g., http://127.0.0.1:11434)
api_key	Bearer token for authentication. Leave empty if no auth is required.
model	Model name/ID available on the server
keep_alive	How long to keep the model loaded in memory (-1 = forever, 0 = unload immediately)

LLM Options

Advanced inference parameters mapped to OpenAI-compatible API names:

Node Option	Mapped To	Description
temperature	temperature	Creativity of output
top_p	top_p	Nucleus sampling threshold
top_k	top_k	Top-k sampling
seed	seed	Reproducibility seed
stop	stop	Stop sequences
repeat_penalty	presence_penalty	Repetition penalty
num_predict	max_tokens	Max tokens to generate

Note: Each option has an enable/disable toggle. Only enabled options are sent in the API request. The debug flag enables CLI debug output (not part of the API).

Deprecated nodes

Old V1 nodes are still available, please replace them with the above ones. Here's the documentation of V1 nodes.

Usage Example

Please see the example_workflows folder or use ComfyUI's template browser.

The custom Text Nodes in the examples can be found here: https://github.com/pythongosssss/ComfyUI-Custom-Scripts