Novaflow

Inspiration With the explosion of user-generated content, manual video moderation is impossible. We wanted to build a "Digital Lawyer" that doesn't just flag issues but actively fixes them using the cutting-edge reasoning of Amazon Nova. What it does Novaflow is an automated brand-safety engine. It audits videos for unauthorized logos or safety hazards, extracts the exact frame coordinates, and uses generative AI to "heal" the video by intelligently removing violations while preserving the original content. How we built it We used FFmpeg to split videos into frame-level data, stored in MongoDB GridFS. The "brain" uses Amazon Nova 2 Pro for multimodal auditing and Nova Canvas for surgical inpainting. The frontend is built with Next.js 16 for a real-time "Mission Control" audit experience. Challenges we ran into Synchronizing high-frequency frame extraction with AI inference speeds was tough. We solved this by implementing an event-driven AWS Step Functions architecture to handle parallel processing of frame batches without crashing. Accomplishments that we're proud of We successfully achieved "surgical remediation"—the ability to identify a violation at a specific $(x, y)$ coordinate and remove it so seamlessly that the viewer cannot tell the video was ever edited. What we learned We mastered Multimodal Prompting, learning how to force AI to output precise JSON data. We also learned how to scale heavy video processing using a serverless cloud infrastructure. What's next for Novaflow We plan to implement Audio Auditing (detecting copyright music) and Live-Stream Integration, allowing Novaflow to protect brands in real-time during broadcasts.

Built With

• brand-safe
• built-with-next.js-16-and-aws-step-functions
• final
• for
• reconstructed
• video
• we-use-ffmpeg-to-deconstruct-videos-into-mongodb-gridfs.-amazon-nova-2-pro-identifies-violations-via-json-coordinates
• while-nova-canvas-inpaints-and-"heals"-the-frames.-the-result-is-a-surgically-audited