From Veo to Motion Transfer: Building Demo Videos for a BJJ Training App

Описание и применение

Создание реалистичных демонстрационных видео для тренировок по бразильскому джиу‑джитсу с помощью AI оказалось невозможным для базовых движений, поэтому автор записал собственные движения, перенёс их на виртуального атлета с помощью Wan 2.2 и применил пост‑обработку, чтобы получить качественные циклические клипы.

Задача

Generating realistic demo videos for BJJ drills using AI failed for basic movements like hip escape because the model had never seen them.

Результат

The app now delivers realistic, loopable demo videos for all BJJ drills, offline voice guidance, and a streamlined HIIT experience.

Рабочий процесс

• Generate starting pose with Gemini‑3‑pro‑image
• Feed pose to Veo 3.1 to create 8‑second loop
• Convert to 960×540 H.264 with ffmpeg
• Film author performing movements
• Transfer motion with Wan 2.2
• Stabilize footage with OpenCV
• Grey background with MediaPipe
• Slow down, interpolate, speed up with ffmpeg
• Repair freeze‑frames with script
• Remove loop seam with crossfade

Доказательства

It worked for push‑ups. It completely failed for the hip escape — the most basic movement in BJJ — because the model had simply never seen it.

Источник

I switched to motion transfer with Wan 2.2 (Wan2.2‑Animate on fal.ai): it reads the skeleton and motion from a source video and redraws it onto the reference athlete.

Источник

Raw transfer wasn’t shippable either. The post‑processing pipeline that made it work: Stabilization (OpenCV), Background greying (MediaPipe), Slow it down, then speed it back up (ffmpeg), Freeze‑frame repair, Loop seam removal.

Источник