nerwander/comfy-chatbot repository overview

⁠ComfyUI Chat

A self-hosted web chat interface for generating images with ComfyUI⁠. Type a prompt, watch progress stream in real time, and see the generated image appear inline in the conversation.

⁠Screenshot

⁠Features

• Chat UI — prompt history (↑/↓), LoRA chip shortcuts, lightbox image viewer
• LoRA tags — include <lora:name:strength> anywhere in a prompt
• Live progress — Server-Sent Events stream status updates while ComfyUI runs
• Slash commands — /help, /server, /server-add, /t2i-workflow
• Multi-server — switch between ComfyUI instances at runtime; supports both Unix and Windows path conventions
• Persistent output — generated images saved to a configurable host directory

⁠Quick Start

The easiest way to run ComfyUI Chat is with the published image on Docker Hub (nerwander/comfy-chatbot⁠).

1. Download docker-compose.yml⁠ from this repo
2. Edit it — set COMFY_SERVER to your ComfyUI address and adjust the volume mounts
3. Run it

docker-compose up -d

1. Open http://localhost:5000⁠ — default login: user / password

To build from source instead, replace image: nerwander/comfy-chatbot:latest with build: . (or keep both — Docker Compose uses build when present and falls back to image for docker pull).

⁠Configuration

All settings are environment variables in docker-compose.yml:

⁠loras.json

The LoRA catalogue controls which chips appear in the chat UI and provides default strengths. Each entry can be a plain filename string or an object with an explicit strength:

{
  "loras": [
    { "name": "styles/cinematic.safetensors",   "strength": 0.8 },
    { "name": "styles/anime-flat.safetensors",  "strength": 0.7 },
    { "name": "characters/hero.safetensors",    "strength": 1.0 },
    "detail-enhancer.safetensors"
  ]
}

Plain strings default to strength 1.0. The name value is the path sent to the ComfyUI server — on Windows servers use backslashes (styles\\cinematic.safetensors) or set COMFY_SERVER_OS=windows and let the app convert them automatically.

⁠Volume mounts (docker-compose.yml)

volumes:
  - ~/dot-files/comfyui:/app/workflows:ro   # workflow templates + loras.json
  - ~/Pictures/ComfyUI:/app/output           # generated image output

⁠Slash Commands

⁠Adding Workflows

⁠1. Export from ComfyUI as API format

Open your workflow in ComfyUI, then in the menu enable "Dev Mode Options" (Settings → Enable Dev Mode Options). A new "Save (API Format)" button will appear. Use this — not the regular Save — to export the file. The regular format includes UI layout data that the chatbot cannot execute directly.

⁠2. Add placeholder tokens

The exported JSON is a static snapshot; you need to replace the values you want the chatbot to substitute at generation time with placeholder tokens. Open the file in a text editor and replace:

For example, a KSampler node's prompt input might look like this before and after:

// Before
"text": "a photo of a cat"

// After
"text": "<PROMPT>"

And a LoRA loader node:

// Before
"lora_name": "my-lora.safetensors",
"strength_model": 0.8,

// After
"lora_name": "<LORA_1_NAME>",
"strength_model": <LORA_1_STRENGTH>,

LoRA placeholders are optional — any slots not filled by the user's prompt are automatically removed from the workflow graph before submission.