How to run uncensored offline AI image generation on your phone

Imagine having a powerful, uncensored AI model running directly on your phone—no internet, no filters, and no subscriptions. That’s exactly what this open-source app delivers. It runs locally on Android and iOS, can generate images, understand photos and PDFs, and even act as a local AI server for other devices on your Wi‑Fi.

What this offline AI app can do

This app turns your phone into a mini AI workstation. Once you’ve installed it and loaded a model, everything runs locally on your device. Here’s what it can handle:

• Uncensored chat: The offline models don’t apply the usual safety filters you see in cloud AIs, so they won’t refuse prompts based on content policies.
• Image understanding: With the right model, you can upload a photo and the AI will describe what’s in it or answer questions about it.
• PDF and document Q&A: Attach a PDF or document and ask questions about any part of the file, like having a personal research assistant in your pocket.
• Image generation: With supported models, you can generate images directly on your phone, completely offline.

If you’re interested in how far image generation can go in the cloud, it’s worth comparing this setup with tools like advanced online image models such as ChatGPT Image 2.

Offline and online AI in one app

While the main focus is offline AI, the app also supports cloud providers if you want extra power or specific models. You can plug in API keys for services like:

• DeepSeek
• OpenAI
• Google Gemini
• Other providers via platforms like OpenRouter or Nvidia NIM

The app will validate your API key and automatically fetch the list of models linked to your account. You can then switch between offline models and online models in the same interface, depending on what you need.

Supported model formats: Lite RT vs GGUF

The app supports two main types of models, each with different strengths:

1. GGUF models
• Usually text-only (chat)
• Often marked with tags like GGUF and uncensored
• No image generation or file understanding
• Good for pure offline chatting and text tasks

2. Lite RT models
• Optimized for speed and smooth performance on mobile
• Often come with a vision tag for image/PDF understanding
• Some include an image tag for image generation
• Faster and more capable for multimodal tasks than GGUF on phones

In short: use GGUF for text-only chat, and Lite RT (especially with vision or image tags) for anything involving images or documents.

How to download and import models the right way

When you open the app for the first time, it won’t do anything until you load a model. You’ll see two ways to get one:

Option 1: In-app “Get” button
• Taps directly into a model download from inside the app
• Simple, but not ideal because the file is stored only inside the app’s data

Option 2: Download to your phone’s Downloads folder (recommended)
1. Tap the download icon instead of the “Get” button.
2. This saves the model file into your normal Downloads folder.
3. The file stays on your phone even if you clear app data, until you manually delete it.
4. Once downloaded, tap the import icon in the app and select the model file.
5. The app will import it and the model will appear in your models list.

This approach makes it easier to reuse models, back them up, or move them between apps and devices.

Choosing the right model for your phone

Not every model will run well on every device. The key rule is simple: don’t load a model that’s bigger than your phone’s RAM. If you do, the app may crash or become extremely slow.

When browsing models, pay attention to tags:

• GGUF + uncensored: Text-only chat, no image or file support.
• Lite RT + vision: Can understand images and PDFs, and runs faster.
• image: Supports image generation (text-to-image).

Once you load a vision-enabled Lite RT model, you’ll see a file attachment button appear in the chat. That’s your signal that the model can handle images, PDFs, and other files. If you switch to a plain GGUF chat model, that attachment button disappears.

How to generate images offline

To generate images directly on your phone, you need a model that explicitly supports image generation. In the app, look for models with the image tag. One example mentioned is a model called “Cyber Realistic,” which is tuned for realistic images.

After you’ve downloaded and loaded an image model, there are two main settings you should understand:

1. Generation steps
• More steps usually mean better image quality.
• More steps also mean longer generation time.
• For most phones and models, 8–12 steps is a good balance between speed and quality.

2. Image size
• Larger images look more detailed and sharp.
• Bigger resolutions use more RAM and take longer to generate.
• If your phone is mid-range or older, start with smaller sizes and work up.

For example, you might prompt something like “a girl dancing in a dark room” and wait a few minutes for the result. Generation time can vary a lot depending on your phone’s CPU/GPU and the model size—anything from a couple of minutes to longer for high-res images.

If you’re curious about other “no limits” image and video tools, you might also like this guide on free uncensored AI image and video generators.

Using online models inside the same app

If offline performance isn’t enough or you want access to frontier models, you can switch to the app’s online section. There you’ll find integrations for providers like:

• DeepSeek (for models like DeepSeek V4 Pro)
• OpenAI (GPT models)
• Gemini
• Others via OpenRouter or Nvidia NIM

To use them, you simply:

1. Choose a provider in the online models section.
2. Paste in your API key.
3. The app validates the key and fetches your available models in real time.
4. Select a model and start chatting or generating content, just like with offline models.

This setup gives you the best of both worlds: fully private offline AI when you want it, and powerful cloud models when you need extra capability.

Turn your phone into a local AI server

One of the most advanced features in this app is the built-in AI server mode. With Lite RT models, you can start a local server on your phone that exposes an API over your Wi‑Fi network.

Key points:

• The server uses OpenAI-compatible endpoints.
• Other devices on the same Wi‑Fi can send requests to your phone as if they were calling the OpenAI API.
• This means you can plug your offline phone model into tools, scripts, or apps that already support the OpenAI API format.

In practice, your phone becomes a small, private AI backend for your laptop, tablet, or even other phones—without sending data to any external cloud.

Installation, iOS notes, and troubleshooting

The app is fully open-source and built with Flutter. You can:

• Download prebuilt releases and install them directly on your Android device.
• For iOS (iPhone or iPad), follow the specific installation steps described in the project’s README, as iOS has extra signing and installation requirements.
• If you’re a developer, you can clone the GitHub repo, customize the code, and build your own version.

If you run into issues like models failing to load or the app freezing:

• Try reloading the model from the models list.
• If that doesn’t help, restart the app and load the model again.
• Make sure the model size is appropriate for your device’s RAM.

Because everything is local and open-source, you’re in full control: no ads, no subscriptions, and no hidden paywalls—just your phone and your models.