How the NARA matcha ad was built: one AI creator, a character-design sheet, a shot-by-shot UGC storyboard, and a Seedance 2.0 video — for a brand that does not exist. The full production system.
One Nano Banana 2 hero shot: describe the person, the outfit, the set and the product (short label text on packaging renders fine). This single image becomes the reference for EVERYTHING else.
02
02Step 2
Build the character sheet
Generate 6 headshots (front, laugh, wave, profiles, 3/4) through the edit endpoint with the hero as reference, repeating the identity description every time. Lay them out as a character-design sheet — that sheet IS shareable content.
03
03Step 3
Storyboard the ad in 6 shots
Hook wave → product to lens → ritual → texture → payoff sip → CTA. Generate each as a still with the same reference. Write one script line per beat. This board is your ad BEFORE you spend a cent on video.
04
04Step 4
One video call brings it alive
Seedance 2.0 reference-to-video with the hero image + ambient audio ON. ⚠️ Two moderation walls we hit: spoken dialogue gets flagged (keep her silent, ambience only) and photoreal faces are blocked as image-to-video input — use reference-to-video, and if the face drifts, regenerate the STILLS from a video frame instead of re-rolling the video. Stills cost cents, videos cost dollars.
The identity lock (paste into every prompt)
Same woman as the reference image (late 20s, warm auburn wavy hair, light freckles, cream knit cardigan), same cozy bright kitchen, same soft morning light, same photorealistic phone-camera UGC look.
+ pass the hero image as reference in EVERY still and video call
Want this running for you?
Build it step by step inside AI Creative OS, with me in the thread.
