Get started

The fastest verified way to try KMP-Vision is the Vision Lab tutorial. It builds directly against the modules in this repository and runs on Android and iOS.

Run from source

git clone https://github.com/lamplis/kmp-vision.git
cd kmp-vision
./gradlew :tutorial:android:installDebug

For iOS, open tutorial/ios/VisionLab.xcodeproj in Xcode and run the VisionLab scheme on an arm64 simulator. See the tutorial for requirements and a walkthrough.

Add a checkout to an existing project

Clone KMP-Vision alongside your application. In your application's settings.gradle.kts, include it as a composite build and map the artifact names to the source modules:

includeBuild("../kmp-vision") {
    dependencySubstitution {
        substitute(module("io.github.lamplis:vision-camera"))
            .using(project(":vision-camera"))
        substitute(module("io.github.lamplis:vision-gallery"))
            .using(project(":vision-gallery"))
        substitute(module("io.github.lamplis:vision-picker"))
            .using(project(":vision-picker"))
        substitute(module("io.github.lamplis:vision-overlay"))
            .using(project(":vision-overlay"))
        substitute(module("io.github.lamplis:vision-ocr"))
            .using(project(":vision-ocr"))
        substitute(module("io.github.lamplis:vision-infer"))
            .using(project(":vision-infer"))
    }
}

Then choose your modules in the application's shared source set:

commonMain.dependencies {
    implementation("io.github.lamplis:vision-ocr:1.0.0")
    implementation("io.github.lamplis:vision-overlay:1.0.0")
    // Optional:
    // implementation("io.github.lamplis:vision-camera:1.0.0")
    // implementation("io.github.lamplis:vision-gallery:1.0.0")
    // implementation("io.github.lamplis:vision-picker:1.0.0")
    // implementation("io.github.lamplis:vision-infer:1.0.0")
}

Here the composite build resolves these coordinates to your local checkout. 1.0.0 is the source version in this repository; this example does not rely on Maven Central. Pin the checkout to the commit you have tested when adopting it in your application.

1.0.0 is also published on Maven Central for every module in the table below.

Choose only what you need

Artifact	Source module	Purpose
vision-camera	:vision-camera	Compose camera UI and analysis frames
vision-gallery	:vision-gallery	Device photo gallery
vision-picker	:vision-picker	System image selection
vision-overlay	:vision-overlay	Rectangles and labels
vision-ocr	:vision-ocr	Native text recognition
vision-infer	:vision-infer	Compatible on-device models

vision-camera exposes vision-overlay on its compile classpath. Add vision-ocr separately when recognizing text; the camera dependency alone does not include OCR.

Platform setup

• Android: minimum API 24. The tutorial uses compile SDK 37 for the current Compose dependencies. Camera capture requires the camera permission; the tutorial declares it and handles the library's permission UI.
• iOS: the tutorial targets iOS 15+, with arm64 device and simulator targets. Add NSCameraUsageDescription before using camera capture. Physical camera capture needs a device.
• Models: package compatible .tflite and .mlmodelc assets and supply your own preprocessing. See model integration.

The camera and picker guide covers app configuration and API examples.