Loading...
Personal Projects
Signal Seal
Personal Project
As AI audio and video tools get better, distinguishing real from fake gets harder. Signal Seal was an attempt to solve that — protect content from AI training, prove authenticity, and detect AI-generated media. Worked on this with a developer colleague. He trained the models, I handled product and design. The technology actually worked. The problem was who we were competing against.
Role
Designer + Dev Collaboration
Year
2024
Status
Postponed
What We Built
Three core features:
- •
Content Protection
Watermark audio and video files. The watermark is practically inaudible — you can barely tell the difference. But it renders the content useless for AI training.
- •
Authentication
Prove who created something. Detect if content was edited after signing.
- •
AI Detection
Point at any audio or video file. Determine if it's AI-generated. If it was Signal Seal signed, identify the original author.
What Worked
The results were genuinely impressive. Audio protection broke clone attempts. Video protection made content invisible to AI. Detection worked against most tools.
Desktop
Protect
Mobile
AI Detector
Why It Stopped
The technology worked against most cloning tools — until we tested against Eleven Labs. Their models cut through our protection. That was the moment: we weren't competing with other startups. We were competing with billion-dollar research labs who update their models weekly. Every protection we built would need rebuilding every few months. As a side project, that's not sustainable. The lesson: pick fights you can win. This one needed venture backing and a full-time ML team, not weekends and determination.