Real Time Detection Nudity And Violence Images & videos

Real-time AI image & videos analysis for detecting nudity and violence, ensuring safer digital content and platforms using the Gemini 3 Flash Preview API.

Comment

Inspiration

The idea for this project came from my daily experience surfing the internet, where I constantly encounter violent and explicit images and videos **. This type of content can be especially harmful for **teenagers, who may be exposed without any warning or protection.

I wanted to build something useful that could help make online platforms safer by detecting sensitive content automatically and in real time.

What it does

  • Analyzes images in real time to detect nudity and violent content.
  • Supports real-time video detection, analyzing the entire video by gemini in the background.
  • Accepts images from URLs or external sources.
  • Uses the Gemini 3 Flash API for fast and reliable image understanding.
  • Automatically flags sensitive content to protect users, especially teenagers.
  • Provides a clear safety decision: safe or inappropriate.
  • Designed for easy integration into web platforms, applications, and moderation systems.
  • Integration of ** test.html ** file in order to test the extension .

How I built it

I started by integrating the Gemini API for image analysis, focusing on detecting nudity and violence.

The workflow is simple:

  1. Collect an image or a video
  2. Fetch and preprocess the data
  3. Send it to the *Gemini *
  4. Analyze the response and flag sensitive content

In theory:

$$ \text{Safety Decision} = f(\text{Image / Frame}) \rightarrow {\text{Safe}, \text{Sensitive}} $$

Where ( f ) is powered by the Gemini vision model.

Challenges I ran into

  • Gemini 1.5 did not perform well for this use case
  • Gemini 2.5 Flash required tuning to achieve stable results
  • Handling images from URLs caused parsing and format issues
  • Ensuring real-time performance without latency was challenging
  • Extending detection to processes the entire video holistically in the background before deciding if it is harmful in real time

After multiple iterations, Gemini 3 Flash provided the best balance of speed, stability, and accuracy.

Accomplishments that I’m proud of

  • Built a real-time detection system for both images and video streams
  • Successfully integrated the Gemini 3 Flash API
  • Solved the challenge of processing images directly from URLs
  • Implemented real-time video frame analysis
  • Designed a clear and scalable moderation workflow
  • Created a project focused on online safety, especially for teenagers
  • Gained hands-on experience in API debugging, AI model selection, and performance optimization

What I learned

  • How to work with AI vision APIs, especially Gemini models
  • How to handle image and video frame inputs from external sources
  • How to debug API limitations and version issues
  • How to design a real-time moderation system
  • That choosing the right AI model version is critical for success
Share this project:

Updates