Inspiration
During high school, I researched about the impact of generative AI on creative labor, digital identity, and public trust. These days, models are able to tell if something is AI generated. However, conflicts on a person's right of publicity and an entertainment's studio written policy introduce tension in many digital assets.
What it does
Framed is a multi-agent media governance tool. A user uploads an image or video, and specialized agents collect provenance evidence, check available C2PA Content Credentials, optionally inspect selected video frames, and compare the findings with review policies. The system then produces an evidence-backed report and flags uncertain or higher-risk assets for human review.
How we built it
We designed the workflow around separate provenance, visual-analysis, and policy-review responsibilities. Each agent contributes structured findings to a shared case so the final recommendation is traceable. Framed treats missing metadata as uncertainty—not proof that content is authentic or synthetic—and keeps a person in the loop for final decisions.
Challenges we ran into
The hardest challenge was handling incomplete or conflicting evidence without overstating what automated analysis can prove. Video also adds cost and complexity, so frame analysis is optional and targeted.
What we learned
Responsible media AI needs more than a detector. Provenance signals, clear policies, and human judgment work best together.
What's next
Next, we want to expand policy customization, strengthen audit trails, and connect Framed with digital asset-management workflows.
Built With
- ai
- c2pa
- computer
- vision
