Matt Wolfe · Video summary

I Built An AI Slop Detector

Matt builds an AI slop detector to identify AI-generated videos, navigating the challenges of current AI models and API limitations.

Matt Wolfe · September 9, 2026 · 20-min episode · 4-min read

Key takeaways

👁️AI-generated video content is becoming increasingly prevalent and can easily deceive viewers, highlighting the urgent need for reliable detection tools.00:05
🧠Google's Gemini models are uniquely capable of understanding video content, making them a valuable asset in developing AI video detection systems.01:58
🚧Initial attempts at building an AI detector using current LLMs may produce flawed results, as evidenced by the detector misidentifying an obvious AI video.04:53
🛠️AI models, even advanced ones, struggle with self-correction and often require extensive iterative testing and human intervention to improve accuracy.05:43
💰Relying on third-party APIs like Sightengine can significantly enhance AI detection capabilities, but introduces unforeseen costs that can hinder public deployment.07:00
📉The high cost of API usage for AI detection makes it difficult to offer free, publicly accessible tools, limiting widespread adoption and content verification efforts.15:55

Chapters

0:00⚠️ The Problem with AI Video Slop
1:55🤖 Initial Exploration with ChatGPT and Existing Tools
5:42❌ The First Attempt: A Promising Start, But Flawed Execution
8:45🔄 Iterative Development and Model Upgrades
11:37🔗 The Pivot to a Third-Party API: Sightengine
16:49✅ Testing the Sightengine Integration
26:30💸 The Unforeseen Challenge: API Usage Costs
27:30🤔 The Broader Implications: AI's Struggle with Self-Detection

The digest

In an era saturated with AI-generated content, distinguishing between authentic and synthetic videos has become increasingly challenging. This project chronicles the ambitious endeavor to create an "AI Slop Detector," a tool designed to help users identify AI-generated videos on popular platforms like YouTube, Instagram, and TikTok.

The Problem with AI Video Slop

AI-generated video content is ubiquitous, often fooling even discerning viewers. The ease with which AI can create convincing, yet fake, videos poses a significant challenge to content authenticity. The goal was to develop a simple web-based tool where users could paste a video URL, and the app would provide a verdict on whether the video was AI-generated. This would empower users to verify content before sharing, combating the spread of misinformation.