Portable Comfyui

ComfyUI Image Generator - Hermes Skill

You have access to a portable ComfyUI installation for AI image generation with Stable Diffusion and other models.

When to Use

Use this skill when the user wants to install or manage the portable ComfyUI extension, work with its models or nodes, or generate images through a local ComfyUI instance.

Server Info

• Installer API: http://127.0.0.1:5000 (management)
• ComfyUI Web UI: http://127.0.0.1:8188 (when instance is running)
• Location: extensions/comfyui/

Quick Start

If not running, start the management server:

cd extensions/comfyui && launcher.bat

If not installed yet:

cd extensions/comfyui && install.bat

Architecture

ComfyUI has two layers:

1. Installer/Manager API (port 5000) — manages instances, models, nodes
2. ComfyUI instances (port 8188+) — the actual image generation servers

Management API Endpoints (use via terminal with python httpx)

System Status

import httpx
r = httpx.get("http://127.0.0.1:5000/api/status")   # Overall status
r = httpx.get("http://127.0.0.1:5000/api/gpus")      # GPU info
r = httpx.get("http://127.0.0.1:5000/api/settings")  # Current settings

Install ComfyUI

r = httpx.post("http://127.0.0.1:5000/api/install")        # Full install
r = httpx.post("http://127.0.0.1:5000/api/update")         # Update ComfyUI
r = httpx.post("http://127.0.0.1:5000/api/install/sage-attention")  # Install SageAttention

Instance Management (multi-GPU)

# List all instances
r = httpx.get("http://127.0.0.1:5000/api/instances")

# Create new instance
r = httpx.post("http://127.0.0.1:5000/api/instances", json={
    "gpu": 0,       # GPU index
    "port": 8189,   # Port for this instance
    "vram": "normal" # normal, low, cpu
})

# Start/Stop instances
httpx.post("http://127.0.0.1:5000/api/instances/{id}/start")
httpx.post("http://127.0.0.1:5000/api/instances/{id}/stop")
httpx.post("http://127.0.0.1:5000/api/instances/start-all")
httpx.post("http://127.0.0.1:5000/api/instances/stop-all")

Model Management

# Browse model registry (101+ pre-defined models)
r = httpx.get("http://127.0.0.1:5000/api/models/registry")
r = httpx.get("http://127.0.0.1:5000/api/models/categories")
r = httpx.get("http://127.0.0.1:5000/api/models/search?q=sdxl")

# See locally installed models
r = httpx.get("http://127.0.0.1:5000/api/models/local")

# Download a model
r = httpx.post("http://127.0.0.1:5000/api/models/download", json={
    "id": "sd15-base"  # Model ID from registry
})

Custom Nodes

# Browse node registry
r = httpx.get("http://127.0.0.1:5000/api/nodes/registry")

# See installed nodes
r = httpx.get("http://127.0.0.1:5000/api/nodes/installed")

# Install custom node
r = httpx.post("http://127.0.0.1:5000/api/nodes/install", json={
    "url": "https://github.com/author/custom-node.git"
})

# Update nodes
httpx.post("http://127.0.0.1:5000/api/nodes/update-all")

ComfyUI Target Directory

r = httpx.get("http://127.0.0.1:5000/api/comfyui/target")           # Current target
httpx.put("http://127.0.0.1:5000/api/comfyui/target", json={"path": "..."})  # Change target
httpx.post("http://127.0.0.1:5000/api/comfyui/target/reset")        # Reset to default

Jobs & Logs

r = httpx.get("http://127.0.0.1:5000/api/jobs")          # List background jobs
r = httpx.get("http://127.0.0.1:5000/api/jobs/{id}")      # Job status
r = httpx.get("http://127.0.0.1:5000/api/logs")           # Server logs

Cleanup

httpx.post("http://127.0.0.1:5000/api/purge")      # Remove ComfyUI (keep models)
httpx.post("http://127.0.0.1:5000/api/purge-all")   # Remove everything

Generating Images (via ComfyUI Instance API)

Once a ComfyUI instance is running (e.g., on port 8189):

# Queue a prompt/workflow
r = httpx.post("http://127.0.0.1:8189/prompt", json={
    "prompt": {workflow_json},
    "client_id": "hermes"
})

# Get generated images
r = httpx.get("http://127.0.0.1:8189/view", params={"filename": "output.png"})

Available Tools (use these instead of raw API calls)

Tips

• Install ComfyUI first via the installer API, then download models, then start an instance
• Use api/models/registry to browse all 101+ pre-defined models
• Each GPU can run its own instance on a separate port
• The web UI at the instance port provides a visual workflow editor
• Models are stored in extensions/comfyui/comfyui/models/
• All portable — nothing installed system-wide