Generate AI B-roll clips from videos with Veo 3

How it works

1. Analyze the video – Downloads a 2-minute video and uses Gemini 2.5 Flash to identify emotional peaks, hook moments, and timestamps

2. Select best moments – Uses GPT-4o-mini to pick the 3 best moments for B-roll insertion based on confidence scores and emotional impact

3. Generate B-roll prompts – For each selected moment, GPT-4o creates an 80–120 word Veo 3 prompt with camera movement, lighting, shot type, and music direction

4. Create B-roll clips – Generates four 8-second AI video clips with Veo 3.0 for each moment and uploads to GCS

5. Deliver – Aggregates all generated clip URLs and sends them to Telegram

---

Set up steps