how i steal viral videos with ai part 2 - branded dropshipping

Full walkthrough on how to recreate any viral video using AI — from extracting frames and motion data to generating each scene with a consistent character. I break down the exact process: downloading the reference video, pulling key frames from every scene, extracting detailed motion data with Gemini, and rebuilding each scene inside Freepik with AI generation. By the end, you'll know how to take any complex viral video with a real character doing real actions and turn it into an ad or organic content piece. Tools Used: Analysis & Swipe File: Floh (waitlist: https://app.floh.ai/waitlist) Movement Extraction: Gemini (video → JSON) Image Generation: Freepik Timestamps: 0:00 — Intro 0:26 — Downloading the viral video & extracting frames 2:17 — Extracting motion data with Gemini 3:04 — Setting up the workspace in Freepik 4:44 — Aligning frames with motion descriptions 7:30 — Regenerating the character with AI 9:33 — Swapping in your own product (Gruns example) 10:56 — Picking the best generations 13:00 — Turning images into video with Kling 3.0 15:39 — Reviewing & selecting clips 18:25 — Product detail refinement 19:10 — Final Results Key Points: • Frame extraction + motion extraction are two separate steps — frames give you the visual reference, motion data (via Gemini) gives you the movement instructions for AI generation • You need a separate JSON motion file for each scene — dumping all motion into one file doesn't work; each scene gets its own text prompt paired with its own reference frame • Gemini can analyze a full video and segment motion scene-by-scene automatically — prompt it to "extract in detail all motion, segment scene by scene, in a detailed JSON" • Consistent characters across AI-generated scenes require using the original frames as reference images — not just text descriptions • The motion descriptions often need manual refinement — AI extracts the broad strokes but you'll want to rewrite specific actions ("slamming the fridge door closed, turning