Comfy Colab Kit
Wake-on-demand ComfyUI

Sleeping → Waking → Ready

ComfyUI on Colab GPUs, awake only when you need it

Break free from baby-sitting Colab browser tabs. Comfy Colab Kit lets you wake ComfyUI on demand on Google Colab L4 GPUs from your phone, laptop, or autonomous AI agent. Protected by your personal Cloudflare Access zero-trust gate with automated idle shutdown.

Quick Start GitHub

L4 · ~1.54 CU/h · 28.4 s per 1024² image

Pain Points

The Problem

Dedicated cloud GPUs drain your budget 24/7, while raw Colab notebooks force you to sit at a PC and baby-sit fragile browser tabs.

💸 Idle Waste

Cloud GPUs bill continuously

A rented cloud GPU bills around the clock, even when hours pass without generating a single image.

🔌 Tethered to PC

Colab notebooks are fragile

Notebooks time out, public URLs change randomly, laptop sleep kills the session, and manual dependency setups waste time.

👥 Collaboration & Agents

Sharing & automation are messy

Teammates accidentally spin up duplicate VMs, and headless AI agents have no clean way to trigger or orchestrate generation.

Comfy Colab Kit gives Colab a single, stable front door that sleeps when idle and wakes in one click from anywhere.

Architecture

How It Works

A lean, zero-trust bridge connecting your personal devices and AI agents with on-demand Colab hardware.

Browser / Agent mobile, web or API Cloudflare Access zero-trust auth VPS Gate Caddy + comfy-gate Google Colab L4 ComfyUI runtime

Browser / Agent

mobile, web or API

Cloudflare Access

zero-trust auth

VPS Gate

Caddy + comfy-gate

Google Colab L4

ComfyUI runtime
  1. Open your URL or call the API. Cloudflare Access verifies identity before anything touches your infrastructure.
  2. If the GPU VM is asleep, the gate presents a sleek wake card (or handles agent API requests). Click "Wake" to begin.
  3. The gate automatically allocates a Colab L4, sets up ComfyUI, and syncs models. Everyone sees identical live progress.
  4. Once ready, the same URL seamlessly proxies ComfyUI. Idle timeouts or budget caps return it safely to sleep.
Waking… downloading models62%
The Paradigm Shift

Why Comfy Colab Kit

Compare the daily friction of raw Colab notebooks with our zero-trust, wake-on-demand architecture.

The Old Way

Traditional Colab Notebooks

  • Tethered to PC & Browser

    Must sit in front of a computer, manually run notebook cells, and baby-sit the browser tab.

  • Ephemeral & Fragile Sessions

    Laptop sleep or network blips kill the session. URLs change unpredictably on every restart.

  • Zero Access Control

    Either unauthenticated public tunnels or sharing your private Google account credentials.

  • Inaccessible to AI Agents

    Headless AI coding agents cannot trigger or orchestrate workflows without human manual setup.

  • Overnight Compute Waste

    Forgetting to terminate notebooks burns through dozens of costly Compute Units (CU).

Comfy Colab Kit

Remote & Autonomous Gate

  • Remote Wake Anywhere

    Wake from your phone, tablet, or laptop with one tap. No open Colab tabs or PC required.

  • Single Persistent Custom Domain

    Stable custom URL fronted by Caddy reverse proxy; background gateway persists independently.

  • Cloudflare Access Zero-Trust

    Protected by enterprise SSO and Email OTP. Granular owner roles and service tokens.

  • Autonomous AI Agent Native

    Claude, OpenClaw, or LangGraph agents can wake, generate, and spin down via REST/MCP.

  • Smart Auto-Shutdown

    Monitors idle activity, max session duration, and CU hard caps to automatically release the GPU.

Capabilities

Feature Highlights

Engineered for friction-free generation with enterprise-grade session governance.

⚡ Instant Launch

One-Click Remote Wake

Wake from your phone, laptop, or API. The gate provisions Colab, configures ComfyUI, and proxies it on the same URL.

🛡️ Zero-Trust Auth

Cloudflare Access Security

Protected by zero-trust identity verification (SSO / Email OTP / Service Tokens). No exposed ports, no shared Google passwords.

🤝 Idempotent

Single Shared Session

Wake is strictly idempotent. Everyone sees identical live progress, and a second visitor joins the active VM without duplicating cost.

⏱️ Budget Guard

Smart Auto-Shutdown

Sessions automatically go back to sleep on idle timeout, max duration limit, or when the allocated CU budget is reached.

🎨 App First

App Mode by Default

Boots straight into ComfyUI App mode: type a prompt, get an image. Switch back to the full node graph anytime with one click.

🤖 AI Agent Ready

AI Agent & MCP Native

Headless AI coding agents can wake the gate, trigger image generation, and release GPUs via REST API and MCP protocol.

🧭 Roadmap

MCP server integration based on joenorton/comfyui-mcp-server, mounted at /mcp so AI agents can generate images through your gate.

Configurations

Model Profiles

A profile is a single profile.yaml: public model weights, community workflows, and ComfyUI args. Pick one before waking.

⭐ Recommended

MOSS-Image

Primary profile: OpenMOSS MOSS-Image on a Colab L4, defaulting to MATLOWAI DeBand int8 workflow with 4-step turbo and official quality graphs.

profiles/examples/moss-image
⚡ Fast Turbo

SD 3.5 Large

Stable Diffusion 3.5 on a Colab L4, defaulting to the official SD 3.5 Large fp8 graph with TurboX and Medium turbo fast variants.

profiles/examples/sd-latest
💾 Int8 Quant

Qwen-Image 2.1

Qwen-Image 2.1 official templates on a Colab L4, defaulting to 2.1 Turbo text-to-image with int8 weights that effortlessly fit 24 GB.

profiles/examples/qwen-image
Real-World

Measured Benchmarks

Transparent metrics measured on Google Colab L4 GPU infrastructure.

~1.54 CU/h Colab L4 compute cost per hour
28.4 s One 1024² image, 25 steps
461–557 s Cold start from sleep to ready

Measured on Colab L4 with the MOSS-Image profile. Actual figures will vary depending on network and GPU quota.

Deployment

Quick Start

Deploy the gate on a lightweight VPS behind Cloudflare Access, then wake Colab from your browser or AI agent. Full production walkthrough in docs/AGENT.md.

Read the Deploy Guide Star on GitHub
Terminal 
# Clone the repository
git clone https://github.com/FireTable/comfy-colab-kit.git
cd comfy-colab-kit

# Configure your environment secrets
cp .env.example .env

# Follow docs/AGENT.md to set up Caddy, Cloudflare Access and the gate
# Then open your personal domain and press Wake
Help & Answers

FAQ

Common questions about compute, quotas, workflows, and self-hosting.

Is it free?
The kit is open source under the MIT license. Colab compute comes from your own Google account: free tier or paid compute units.
Can I share it with friends?
It is designed for personal or small trusted team use. Colab compute belongs to your personal Google account. Keep access to yourself or invite friends via Cloudflare Access emails rather than making it public.
What happens when nobody uses it?
It shuts down automatically on idle timeout, max session duration, or when it reaches your CU budget limit, and the gate safely goes back to sleep.
Do I lose the node editor?
Not at all. It opens in clean App mode by default for speedy generation, and a single toggle in the top bar switches to the full ComfyUI node graph editor.
How do I add a workflow?
Add it to a profile.yaml and run git pull on the gate VPS. No container rebuilds or restarts required.
Is it affiliated with Google or Comfy Org?
No. It is an independent open-source project created by the community.