This unofficial privacy-focused version of Gallery source code removes Firebase analytics and GMS Play services dependencies (but read the warning below!). It includes other changes different from the official release.
Warning
Unfortunately, some app dependencies, such as Google's closed-source ML Kit library, may themselves pull in unwanted code and send telemetry-- use ./gradlew app:dependencies from the Android/src/ directory to audit this for yourself. Even worse, ML Kit expressly declares in its Privacy statement that "When you use ML Kit APIs, processing of the input data (e.g. images, video, text) fully happens on-device, and ML Kit does not send that data and the resultant outputs to Google servers.
The ML Kit APIs may contact Google servers from time to time in order to receive things like bug fixes, updated models and hardware accelerator compatibility information. The ML Kit APIs also send metrics about the performance and utilization of the APIs in your app to Google."
To address potentially unwanted data leaks, the AndroidManifest.xml file includes aggressive instructions intended to disable specific analytics and data transport services. These countermeasures may or may not be successful.
Other changes to this code include:
- Strip out Firebase Analytics, Play Services Measurement, and Firebase Cloud Messaging from app code (dependencies remain, see Warning above)
- Strip out Android Tools developer analytics
- Request mlkit not to send analytics and add some manifest/proguard build filters remove AD_ID permissions, disable auto app-backups, etc. in
AndroidManifest.xmlto help with privacy (additional ideas for this are invited!)
- Switch from GMS play-services-tflite to standalone TFLite
- Add missing libraries (e.g., documentfile)
- Update most libraries to their latest versions
- Fix build issues with new protobuf 0.9.6 and newest AGP
- Migrate from kapt to ksp
- Upgrade to Gradle 9.5.1
- Target SDK 37 (Android 17)
- Turn off cloud app backup
- Enable minification and shrinkResources for release builds
- Add ProGuard rules to allow installRelease build to complete
- Auto-copy extra featured skills into the APK
- Update Kotlin to 2.4.0
- Update Java from 11 to 21
- Remove onboarding Play ToC agreement (Play-related class was removed, but the agreement is still available for review in the menu)
- Update to Material 3 DayNight theme
- Add build timestamp to Settings screen
Note
The model list is pulled from Google's GitHub by default. To see extra models, change the default repo in the settings to another user, such as "fat-tire".
- Selectable GitHub-based model lists (if you want to extend beyond the official list)
- Added the 12B Gemma-4 model (for systems with 12GB+ memory)
- Added abliterated model option to menu. See related licenses and terms of use before using, and you agree to them use entirely at your own risk.
- Larger context windows (may break functionality if you exceed context though)
- Increased constants for images, audio, and more
- Speculative Decoding (MTP) enabled by default on supported models
Note
As code is committed upstream, changes may be rebased over the newer code, so updates to this branch may require --force.
Explore, Experience, and Evaluate the Future of On-Device Generative AI with Google AI Edge.
AI Edge Gallery is the premier destination for running the world's most powerful open-source Large Language Models (LLMs) on your mobile device. Experience high-performance Generative AI directly on your hardware—fully offline, private, and lightning-fast.
Now Featuring: Gemma 4
The latest version brings official support for the newly released Gemma 4 family. As the centerpiece of this release, Gemma 4 allows you to test the cutting edge of on-device AI. Experience advanced reasoning, logic, and creative capabilities without ever sending your data to a server.
| Install the app today from Google Play | Install the app today from App Store | Download for macOS |
|---|---|---|
 |
|
 |
For users without Google Play access, install the apk from the latest release
-
Agent Skills: Transform your LLM from a conversationalist into a proactive assistant. Use the Agent Skills tile to augment model capabilities with tools like Wikipedia for fact-grounding, interactive maps, and rich visual summary cards. You can even load modular skills from a URL or browse community contributions on GitHub Discussions.
-
AI Chat with Thinking Mode: Engage in fluid, multi-turn conversations and toggle the new Thinking Mode to peek "under the hood." This feature allows you to see the model’s step-by-step reasoning process, which is perfect for understanding complex problem-solving. Note: Thinking Mode currently works with supported models, starting with the Gemma 4 family.
-
Ask Image: Use multimodal power to identify objects, solve visual puzzles, or get detailed descriptions using your device’s camera or photo gallery.
-
Audio Scribe: Transcribe and translate voice recordings into text in real-time using high-efficiency on-device language models.
-
Prompt Lab: A dedicated workspace to test different prompts and single-turn use cases with granular control over model parameters like temperature and top-k.
-
Mobile Actions: Unlock offline device controls and automated tasks powered entirely by a finetune of FunctionGemma 270m.
-
Tiny Garden: A fun, experimental mini-game that uses natural language to plant and harvest a virtual garden using a finetune of FunctionGemma 270m.
-
Model Management & Benchmark: Gallery is a flexible sandbox for a wide variety of open-source models. Easily download models from the list or load your own custom models. Manage your model library effortlessly and run benchmark tests to understand exactly how each model performs on your specific hardware.
-
100% On-Device Privacy: All model inferences happen directly on your device hardware. No internet is required, ensuring total privacy for your prompts, images, and sensitive data.
- Check OS Requirement: Android 12 and up, and iOS 17 and up.
- Download the App:
- Install the app from Google Play or App Store.
- For users without Google Play access: install the apk from the latest release
- Install & Explore: For detailed installation instructions (including for corporate devices) and a full user guide, head over to our Project Wiki!
- Google AI Edge: Core APIs and tools for on-device ML.
- LiteRT: Lightweight runtime for optimized model execution.
- Hugging Face Integration: For model discovery and download.
Check out the development notes for instructions about how to build the app locally.
This is an experimental Beta release, and your input is crucial!
- 🐞 Found a bug? Report it here!
- 💡 Have an idea? Suggest a feature!
Licensed under the Apache License, Version 2.0. See the LICENSE file for details.