---
title: "Run AI Toolkit on Google Colab to Train Flux, Stable Diffusion, Z-Image, and Qwen-Image LoRAs"
date: 2026-02-05T10:00:00.000Z
author: Z.SHINCHVEN
tags: [ai, ai-toolkit, google-colab, flux, stable-diffusion, z-image, qwen-image]
canonical: https://atlassc.net/2026/02/05/run-ai-toolkit-on-google-colab-to-train-loras
---
Want to train LoRAs for [Flux.1](https://huggingface.co/black-forest-labs/FLUX.1-dev), [Stable Diffusion](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0), [Z-Image](https://huggingface.co/Z-Image/Z-Image-v1.0), or [Qwen-Image](https://huggingface.co/Qwen/Qwen-Image) but lack a powerful local GPU? The best way is to run [AI Toolkit](https://github.com/ostris/ai-toolkit) on [Google Colab](https://colab.research.google.com/). This guide shows you how to get the toolkit up and running in minutes using a cloud-hosted UI.

## Quick Start

To get started, simply run the following setup block in a Google Colab notebook with a GPU runtime (T4, L4, or A100).

```python
#@title Setup ENV

!git clone https://github.com/ostris/ai-toolkit.git
%cd ai-toolkit
!pip install -r requirements.txt

from huggingface_hub import login
from google.colab import userdata

# Access the secret using its name
hfToken = userdata.get('HF_TOKEN')

# Use the API key in your code
print("API key loaded successfully.")

login(hfToken)

!wget https://github.com/cloudflare/cloudflared/releases/latest/download/cloudflared-linux-amd64.deb
!dpkg -i cloudflared-linux-amd64.deb
!pip install gradio --upgrade

%cd /content/ai-toolkit/ui
!npm install && npm run build
```

## Running the AI Toolkit UI

Once the environment is set up, you can run the web interface. We use `cloudflared` to create a secure tunnel, allowing you to access the UI from your local browser.

```python
#@title Run AI Tooklit UI
%cd /content/ai-toolkit/ui

import subprocess
import threading
import time
import socket
import urllib.request

def iframe_thread(port):
  while True:
      time.sleep(0.5)
      sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
      result = sock.connect_ex(('127.0.0.1', port))
      if result == 0:
        break
      sock.close()
  print("\nAI Toolkit UI finished loading, trying to run cloudflared\n")

  p = subprocess.Popen(["cloudflared", "tunnel", "--url", "http://127.0.0.1:{}".format(port)], stdout=subprocess.PIPE, stderr=subprocess.PIPE)
  for line in p.stderr:
    l = line.decode()
    if "trycloudflare.com " in l:
      print("This is the URL to access AI Tookit UI:", l[l.find("http"):], end='')

threading.Thread(target=iframe_thread, daemon=True, args=(8675,)).start()

# Set a secure password for the UI
!AI_TOOLKIT_AUTH=YOUR_SUPER_SECRET_PASSWORD_HERE npm run build_and_start
```

## Prerequisites

1.  **Hugging Face Token:** You need a Hugging Face write token to download base models and upload your results. Store this in your Colab Secrets as `HF_TOKEN`.
2.  **GPU Runtime:** Ensure your Colab notebook is using a GPU (Change runtime type -> T4 GPU or better).
3.  **Storage:** Depending on the model size, you may want to mount Google Drive to save your checkpoints permanently.

## Key Features

-   **Multi-Model Support:** Train LoRAs for Stable Diffusion XL, Flux.1, Z-Image, and Qwen-Image.
-   **Web UI:** A user-friendly interface for configuring training parameters without touching JSON files.
-   **Cloudflared Integration:** Bypass Colab's local networking restrictions to access the full UI features.

## How to Train LoRAs

Once you have the UI running, follow these steps to start your training:

1.  **Upload Dataset:** Prepare your training images and corresponding `.txt` caption files (ensure they have the same filename, e.g., `image1.jpg` and `image1.txt`). Upload them to the toolkit's dataset directory via the UI.
2.  **Create a New Job:** In the interface, add a "New Job." You will need to select your desired base model (e.g., Flux or SDXL), point to your uploaded dataset, and provide an example prompt for validation.
3.  **Start the Training Queue:** Note that jobs are not automatically started upon creation. You must navigate to the **Training Queue** and manually start the jobs you've created to begin the training process.

## What is AI Toolkit?

[AI Toolkit](https://github.com/ostris/ai-toolkit) is a powerful, all-in-one training suite designed by Ostris for fine-tuning the latest diffusion models. It supports high-performance models like Flux.1, SDXL, Z-Image, and Qwen-Image. Its standout feature is the built-in web interface, which simplifies the complex process of setting up training parameters, managing datasets, and monitoring progress without needing to manually edit JSON configuration files.

## What is Google Colab?

[Google Colab](https://colab.research.google.com/) is a cloud-based service that allows you to write and execute Python code in your browser. It is particularly popular in the AI community because it provides free or low-cost access to powerful GPUs like the NVIDIA T4, L4, and A100. By running AI Toolkit on Colab, you can perform heavy-duty LoRA training without needing to invest in expensive local hardware.