Safe Image is an AI-powered image filtering library designed to promote safe and culturally appropriate content.
The library uses advanced deep learning models to automatically detect and blur inappropriate or sensitive content in images, with full control over the blur strength, ensuring that visuals remain respectful and aligned with Islamic values.

Make sure Maven Central is added to your project-level build.gradle (or settings.gradle for Gradle 7+):

repositories {
    google()
    mavenCentral()
}

Add Safe Image in module-level build.gradle:

dependencies {
    implementation("io.github.devvura:safe-image-compose:1.0.0")
}

If you only want to display an image with automatic blurring, you can use SafelImage with just the model parameter:

SafeImage(
    model = "https://example.com/sample.jpg",
    contentDescription = "Sample Image"
)

You can fully customize SafeImage by providing parameters such as loadingContent, errorContent, onBlurContent, blurRadius, and more:

SafeImage(
    modifier = Modifier.fillMaxWidth(),
    model = "https://example.com/sample.jpg",
    contentDescription = "Sample Image",
    loadingContent = { Text(text = "Loading...") },
    errorContent = { Text(text = "Error loading image") },
    onBlurContent = { Text(text = "Sensitive content")},
    blurRadius = 20,
    contentScale = ContentScale.FillWidth
)

Add Safe Image in module-level build.gradle:

dependencies {
    implementation("io.github.devvura:safe-image:1.0.0")
}

If you only want to display an image with automatic blurring, you can use SafeImage with just the model parameter:

val imageView = findViewById<ImageView>(R.id.imageView1)
imageView.loadSafeImage(
    model = "https://example.com/sample.jpg",
    placeholderRes = R.drawable.loading_bar,
    errorDrawableRes = R.drawable.error_svgrepo_com
)

You can fully customize SafeImage by providing parameters such as onLoading, onError, blurRadius, and more:

val imageView = findViewById<ImageView>(R.id.imageView1)
imageView.loadSafeImage(
    model = "https://example.com/sample.jpg",
    blurRadius = 70,
    onLoading = { 
        Toast.makeText(context, "Loading image...", Toast.LENGTH_SHORT).show() 
    },
    onError = { 
        Toast.makeText(context, "Failed to load image", Toast.LENGTH_SHORT).show() 
    },
    onSuccess = { 
        Toast.makeText(context, "Image loaded successfully!", Toast.LENGTH_SHORT).show() 
    },
    crossFadeEnabled = true,
    placeholderRes = R.drawable.loading_bar,
    errorDrawableRes = R.drawable.error_svgrepo_com
)

Safe Image integrates seamlessly with Coil’s image loading pipeline, working specifically on the Transformation step to detect and blur unsafe images before they are displayed.
This ensures maximum performance and that sensitive content never flashes on the screen unblurred.

When you load an image with Safe Image, it is first passed through Coil’s request pipeline, which includes a custom transformation. This transformation uses an AI-powered detector to analyze the image before it is displayed. If the detector identifies unsafe content, the image is blurred with a configurable blur radius; otherwise, it is shown normally.

Under the hood, the detection is powered by a TensorFlow Lite model trained to distinguish between safe and unsafe content. Each image is resized, normalized, and classified, and if the probability of sensitive content exceeds a threshold, the image is flagged as NSFW.

To ensure smooth performance, Coil’s memory and disk caching are enabled so images are not repeatedly processed.

• The AI model was trained on a custom dataset consisting of 11,304 images, categorized as either safe or unsafe.
• The unsafe category includes 5,652 images containing sensitive or explicit content.
• The safe category also contains 5,652 images, featuring appropriate content such as individuals in everyday attire, including hijabi women, men, and women in normal contexts.
• To improve generalization and accuracy, the dataset was carefully collected from individuals of diverse nationalities.
• The trained model achieved an accuracy of 96% on the training set and 94% on the testing set, effectively covering the majority of sensitive content detection cases, with only a few exceptions.
• The deployed TFLite model size is ~10 MB.
• Supported image formats: PNG, JPG, JPEG2000, and WEBP.

Safe Image is built as a Kotlin Multiplatform (KMP) project.
Below are the currently supported targets:

We are continuously improving Safe Image. Here are some of the planned features and enhancements:

• Extended Platform Support – bringing Safe Image to iOS, Desktop, and Web through Kotlin Multiplatform.
• Smarter AI Model – improved accuracy with larger and more diverse training datasets.

🙌 Contributions and suggestions are always welcome!

Safe Image is proudly developed and maintained by:

• Abdulrahman Khattab
• Fares Mohamed
• Malak Raef
• Muhammed Edrees

Copyright 2025 Team Devura

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

   http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License