Generate AI Images Locally on iPhone, iPad, and Mac. Nothing Leaves Your Device.

You need iOS, iPadOS, or macOS 26 and Apple Silicon: a recent high-performance iPhone, an iPad with at least 8 GB of memory, or any Apple Silicon Mac. Every model needs at least 8 GB of memory.

Yes. You download a model once; after that, Private Diffusion is an offline AI image generator — everything runs on your device. No account, no sign-in, no server.

Between about 2.5 GB (Anima) and 12 GB (Krea 2 Turbo, Boogu Image Turbo) each. Downloads continue in the background even when the app is closed, and you can delete any model to reclaim space and download it again later.

Models are gated by memory. Anima and Flux.2 Klein 4B run on 8 GB devices; Mage Flow Turbo, Z-Image-Turbo, and Juggernaut Z Fast need a 12 GB iPhone (the Z-Image models also run on an 8 GB iPad); Krea 2 Turbo, ERNIE Image Turbo, Chroma1-Flash, and Boogu Image Turbo need an iPad or Mac with 16 GB or more.

Seven aspect ratios, three output sizes on most models (plus an extra-large tier on high-memory devices), seed control everywhere, a latent upscaler up to 4x, negative prompts on four models, and an on-device Rewrite Prompt enhancer on four models. Each model page lists exactly what it supports.

No. Private Diffusion generates images from text prompts.

Yes. On iPhone and iPad, generation continues in the background and finishes even if you switch apps. On Mac, you get a notification when your image is ready. A live preview shows the image forming step by step.

Yes. Every image keeps its prompt, model, seed, size, steps, and upscale factor, and Recreate in Studio restores all of it in one tap. The same seed and settings reproduce the image.

No. After a model downloads, Private Diffusion makes no app-code network calls. There is no analytics SDK, no crash reporter, and the privacy manifest declares zero collected data types.

On your device. If you turn on gallery sync, images sync through your own private iCloud database — Numen has no server and no access.

Three layers: Incognito Mode keeps new generations out of your gallery entirely, the Hidden tab locks images behind optional Face ID or Touch ID, and you can hide the Hidden tab itself. In Incognito Mode, a privacy cover also shields the screen in the app switcher.

Generation happens on your device; no cloud service reviews your prompts. If your device's system-level Sensitive Content Warning is on, potentially sensitive images are blurred until you tap Show — nothing is blocked or deleted, and the setting lives in iOS and macOS Settings, not in the app.

HEIC, at high quality. Save to Photos — the app can only add images, it never reads your library — or share, copy, or export to a file on Mac.

Exports carry a Private Diffusion wordmark in the lower-right corner. Images stay clean inside the app — the mark is added only when an image leaves it.

Yes. Exports embed AI-provenance metadata — the prompt, model, seed, and settings — following industry provenance standards, so an image can always be traced back to how it was made and recreated. Worth knowing before you share an image publicly.

The images you generate are yours. Most models permit commercial use; Krea 2 Turbo outputs may be used commercially only while your total annual revenue is under one million US dollars — above that, Krea requires an enterprise license. Mage Flow Turbo has no published license yet. Each model page summarizes the license on record.

Enter your Apple ID email in the signup form and confirm the email we send; your TestFlight invite follows. Free during the beta.

Pre-order is coming to the App Store. Until then, the TestFlight beta is the way in — and it keeps running until launch.

Private Diffusion is free during the TestFlight beta. Launch pricing has not been decided.

The app ships in English today. Prompts are free-form text — describe your image in whatever words feel natural.