Adaptive Post-Production Product Placement
Traditional product placement in movies is static, generalized, and often irrelevant to global audiences. We built a post-production system that dynamically replaces branded objects in videos based on viewer context—without reshooting or altering the storyline.
Instead of changing the movie, we change the ads inside the movie.
How It Works
1. Video Input & User Control
Users can submit full videos for automatic detection or use manual on-screen selection to target specific regions. This creates a lightweight video-editing workflow combining user intent with automated understanding.
2. Video Understanding with Gemini
Gemini watches the video end-to-end, identifies frames containing relevant branded objects, and filters semantically meaningful scenes based on user requests and manual selections.
3. Object Segmentation with SAM 3
SAM 3 (Segment Anything Model v3) performs precise pixel-accurate object segmentation, handling occlusion, lighting consistency, and perspective to generate masks for selected objects.
4. Brand Replacement
Using generated masks, we apply post-production inpainting and embedding with ComfyUI to replace brands with context-aware alternatives based on location, availability, and viewer relevance.
Tech Stack
AI Models
Gemini & SAM 3
Backend
Python & FastAPI
Frontend
Next.js & TypeScript
Deep Learning
PyTorch & ComfyUI
Infrastructure
Vast.ai GPU Compute
Styling
Tailwind CSS
Example Use Cases
iPhones in U.S. releases → Huawei for China
Starbucks cups → Milo in regions without Starbucks
Ford F-150 → VW Golf for viewers in Germany
Coke cans → Pepsi in local markets
Why It Matters
Instead of advertising at everyone, we advertise to the right audience. This system transforms static product placement into a dynamic, personalized advertising layer without breaking immersion—no reshoots required, post-production only, and scalable to long-form content.