The problem
A phone running an image model needs the right model files, enough memory, and a way to cancel generation and diagnose failures.
Applied AI / 2026
EXOIMAGE
Private Android preview that runs an image model locally and verifies the imported model pack before generation.
My role
Android interface, local inference integration, and model-pack validation.
What I built
- Private Android preview APK
- Model-pack verification and import
- Locally generated images with save and share controls
The result
A fixed local pipeline generated a 512×512 image on a physical device. The private Android product UI and model-pack import still need end-to-end device testing.
Statusprototype
Evidencereleased
Record reviewed
ReleaseVersion 0.1.2-gpu-preview / build 3
Implementation
How it works
Import a versioned model pack, verify its files, and run the text encoder, denoiser, and decoder locally. The preview includes cancellation, image history, save and share controls, and runtime diagnostics.
Technical context and methods
The APK and protected model pack have fixed versions and SHA-256 checks. Requested GPU execution is reported separately from verified accelerator placement.
- Kotlin
- Jetpack Compose
- LiteRT
- Model-pack validation
- Thermal guards
- On-device generation