How to run a portable, uncensored AI image generator on any PC

Imagine carrying a powerful AI image generator in your pocket that works on almost any computer you plug it into. No installs, no accounts, no filters, and no internet connection required. That’s exactly what this portable setup gives you: a fully local, cross-platform AI image generator that runs from a USB drive or SSD.

What this portable AI image setup can do

This setup turns any compatible PC into a local AI image studio. Once it’s running, you can generate and edit images entirely offline, with no cloud services involved.

Here’s what it supports:

• Text-to-image: Type a prompt and generate any kind of image you want, without online content filters.
• Image-to-image: Upload an existing image and ask the AI to modify it (change age, clothes, style, etc.).
• Hardware acceleration: Automatically uses the best option available on your machine—CPU, GPU, or NPU (neural processing unit).
• Cross-platform: Works on Windows, macOS, and virtually any Linux distro with the exact same files.

Because everything runs locally, you keep full control over your data and what you generate, similar to other local-first tools like uncensored local AI video generators.

Why run AI image generation from a USB drive?

The standout feature of this setup is portability. Instead of installing a different stack on every machine, you prepare one drive and use it everywhere.

With a single USB drive or external SSD you get:

• One shared model library: Download models once and reuse them across Windows, Linux, and macOS.
• Consistent environment: Same interface, same settings, same workflows on every machine.
• No reinstallation: Just plug in, run one script, and you’re ready to generate images.
• Offline by design: Ideal for air-gapped systems, privacy-focused workflows, or unreliable internet.

You can also run it directly from an internal SSD if you don’t care about portability and just want a local, uncensored image generator on a single PC.

Step 1: Prepare your drive (for portable use)

If you want the setup to work across Windows, Linux, and macOS using the same drive, you’ll need a file system that all three can read and write.

• Format the drive as exFAT: exFAT is supported by all major desktop operating systems and is ideal for large AI model files.
• You can use a USB stick, external SSD, or even an external HDD. For speed, an SSD is strongly recommended.

If you only plan to use this on one operating system and don’t need portability, you can skip the formatting step and just use any folder on your internal drive.

Step 2: Download the project from GitHub

The entire setup is packaged as an open-source project hosted on GitHub.

Here’s the basic process:

1. Open the GitHub link (provided with the project).
2. Click the Code button.
3. Choose Download ZIP to grab the full project as a compressed file.
4. Once downloaded, extract the ZIP to your chosen location: your exFAT USB drive, external SSD, or a folder on your internal disk.

After extraction, you’ll see several files and folders. You can ignore most of them—only three launchers really matter.

Step 3: Use the correct launcher for your OS

Inside the extracted folder you’ll find three key files:

• windows.bat – for Windows
• linux.sh – for Linux
• mac.sh – for macOS

To start the app:

• On Windows: double-click windows.bat.
• On Linux: run chmod +x linux.sh if needed, then execute linux.sh.
• On macOS: run mac.sh from Terminal (you may need to grant execute permissions first).

The first time you run the launcher on a given operating system, it will:

• Download all required dependencies automatically.
• Detect your hardware and configure the best backend (CPU, GPU, or NPU).
• Set up the local web interface.

This initial setup usually takes around 8–10 minutes. After it finishes, press Enter and the app will open in your default browser.

Step 4: Download and load an image model

Once the web interface is running, the next step is to install at least one AI model for image generation.

• Open the Model Manager inside the app.
• Choose a model that fits your hardware. For lower-end systems, a model like CyberRealistic is a good starting point.
• Download the model, or if you already have it on your drive, import it from storage.

After the model appears in your list:

• Click the Load button to load it into memory.
• The app will automatically use the best available hardware (for example, a Vulkan GPU backend on supported systems).

Once loaded, you’re ready to start generating images.

Generating your first images

With a model loaded, you can try basic text-to-image generation:

1. Enter a descriptive prompt (for example, “portrait of a woman in natural light, ultra detailed”).
2. Keep the default settings for your first test, then tweak later.
3. Hit the generate button and wait for the result.

On a system without a dedicated GPU, generation might take around a minute or more per image, but the quality can still be surprisingly good. On stronger hardware, it will be much faster.

Editing existing images with image-to-image

This setup isn’t limited to generating images from scratch. It can also edit existing photos or renders using image-to-image mode.

Typical workflow:

1. Upload an image you want to modify.
2. Write a prompt describing the changes, such as “make her look older and change her clothes to pink.”
3. Adjust strength or other settings if available (to control how much the original is preserved).
4. Generate the new image.

The AI will keep the core structure of the original image while applying the requested changes, often with very natural-looking results.

Unlocking NPU acceleration for extreme speed

If your machine has an NPU (neural processing unit), this setup can take advantage of it for huge speed boosts.

Here’s what changes with NPU support:

• The app will automatically show NPU-compatible models in the model list.
• When you load one of these models on an NPU-capable system, generation times can drop from tens of seconds to just a few seconds.

Examples from testing:

• A prompt that took around 77 seconds on a CPU-only system can drop to under 4 seconds with NPU acceleration.
• Even with maximum step counts for higher quality, images can still generate in around 5–6 seconds.

On Apple silicon Macs, you can switch to the NPU backend in the settings and then load an NPU-compatible model. The first load may take 4–5 minutes while it prepares everything, but after that, generations are dramatically faster.

Using the same setup on Windows, Linux, and macOS

One of the biggest advantages of this project is that your entire environment lives on the drive itself.

That means:

• You can plug the same USB or SSD into a Windows PC, run windows.bat, and generate images.
• Unplug it, move to a Linux machine, run linux.sh, and pick up exactly where you left off.
• Do the same on macOS with mac.sh, using Apple’s Metal or NPU acceleration.

All of the following are stored on the drive:

• Downloaded models
• Generated images
• Metadata and settings

Because of this, you maintain a single, unified library of models and outputs, no matter which operating system you’re using. This is similar in spirit to other cross-platform local setups, like running local code models via tools such as Ollama-based environments on any PC.

Why this kind of local setup matters

Running AI image generation locally and portably gives you several important benefits:

• Privacy: Your prompts, images, and models never leave your own storage.
• No filters: You’re not limited by platform content rules or blocked prompts (though you’re still responsible for how you use the tech).
• Reliability: Works even without an internet connection or when cloud services are down.
• Ownership: You control the environment, versions, and updates.

If you’re serious about local AI workflows, this kind of portable image generator is a powerful companion to other on-device tools for voice, video, and code.