What it does SDAI is a Kotlin/Android client for generating images with Stable Diffusion and friends. It speaks to eight different backends—your own AUTOMATIC1111 or SwarmUI instance, AI Horde’s crowdsourced cluster, Hugging Face, OpenAI DALL-E, Stability AI, or two experimental local runtimes (ONNX and MediaPipe). Generated images live in an in-app gallery with export to zip, and the app promises no ads, no telemetry, no spying.

The interesting bit The “local diffusion” options are the eyebrow-raisers: Microsoft ONNX Runtime and Google AI MediaPipe both run txt2img on-device, though the README is honest that these are beta and limited. The rest of the app is essentially a well-organized remote client, but the ambition to go fully offline on a phone is notable.

Key highlights

• Supports txt2img, img2img, and inpainting with full parameter control (CFG 1.0–30.0, steps 1–150, sizes 64–2048px)
• Batch generation up to 20 images for A1111 and Horde
• Picks LoRAs, textual inversions, hypernetworks, and SD models when connected to A1111
• Gallery with pinch-to-zoom, generation metadata, and zip export
• “Demo mode” returns mock images if you have no server handy
• F-Droid flavor exists; feature differences between Play Store/F-Droid/GitHub builds are documented

Caveats

• Local diffusion is beta, txt2img-only, and MediaPipe is restricted to Play Store/full flavors
• The README’s “no ads, no telemetry” claim is stated but not independently verifiable from the source
• Eight backends means eight different setup flows; the first-launch experience could overwhelm

Verdict Worth a look if you want one Android app that can talk to your home server, a free crowdsourced grid, or paid APIs without juggling separate clients. Skip it if you need reliable local generation today—the on-device options are still experimental.