AI content moderation with structured metadata (video tutorial)

Overview

Learn how to use MediaFlows with AI Vision and structured metadata to automate content moderation workflows. This tutorial demonstrates how to build an intelligent system that automatically analyzes user-generated images, detects specific criteria (like visible damage), and routes assets based on AI-powered decisions.

In this example, you'll see how a used car marketplace can automatically review vehicle images, detect dents and damage, add detailed descriptions to metadata fields, and route images to appropriate folders. This automation dramatically reduces review time while maintaining quality control.

Video tutorial

Tutorial contents

This tutorial presents the following topics. Click a timestamp to jump to that part of the video.

Introduction to content moderation

0:00	Discover how to use MediaFlows to automate content moderation for user-generated images. This demo shows a used car marketplace that needs to review vehicle photos for visible damage, demonstrating how AI Vision can replace manual review processes while maintaining quality control.

Understand the business process

0:08	Learn about the business requirement: manually reviewing each user-uploaded car image to check for visible damage. This time-consuming process requires someone to label whether each image shows dents or damage, creating a bottleneck in the workflow.

Automate the workflow

0:18	See how to automate this workflow using Cloudinary AI Vision and MediaFlows. The automation reduces the time spent on manual labeling by having AI analyze each image and automatically determine whether damage is present, then route the image accordingly.

Set up the flow trigger

0:35	Configure the flow to begin with an upload trigger. The automation activates when images are uploaded to Cloudinary in specific folders, ensuring that only relevant uploads trigger the AI analysis workflow.

AI Vision analysis

0:41

Set up the AI Vision Analyze By Prompt block to analyze uploaded assets. The AI evaluates each image with a simple prompt: "yes: the car in the image contains a visual damage or dent. no: the car in the image doesn't have a visual damage or dent return only yes or no" The AI responds with "yes" or "no," providing the foundation for automated routing decisions. See the AI Vision documentation for more capabilities.

Implement condition logic

1:13	Use the condition block to create different workflow paths based on AI responses. If damage is detected (yes), the flow continues with additional analysis. If no damage is found (no), the flow routes the image to a "good-to-go" folder. This branching logic enables sophisticated automation.

Update structured metadata

1:36	Configure structured metadata fields to store AI analysis results. The automation populates metadata fields with "yes" or "no" values indicating whether damage was detected, creating a permanent, searchable record of the AI's assessment.

Advanced AI prompts

1:59	For images with detected damage, the flow asks a second, more detailed question to get a comprehensive description of the dent location and severity. It only processes images that need it, optimizing performance. The description of the damage is stored in another metadata field.

View processing results

2:52	Review the automated results after uploading car images. Watch as the AI processes each image, categorizes them based on damage detection, populates metadata fields with detailed damage descriptions, and automatically routes images to the appropriate folders.

Understand the complete workflow

3:36	Examine the logs to understand the full workflow. See how images flow through the upload trigger, AI Vision analysis, condition checks, metadata updates, and folder routing. This comprehensive automation demonstrates how to use AI Vision to tag assets, add metadata, and organize content to fit your business needs.

Keep learning

If you like this, you might also like...

 

 

Check out the Cloudinary Academy for free self-paced Cloudinary courses on a variety of developer or DAM topics, or register for formal instructor-led courses, either virtual or on-site.