DeepMist AI

The demo reel  /  Kibo breakdown

Production Breakdown

How we made the KIBO ad sound-first

This is a full breakdown of one 10-second ad: the brief, the opening line, the stack, and the finished video. KIBO is a fictional brand we invented; this is a self-initiated spec ad, not client work. Food and drink is judged on texture and sound in the first second, so we led with both.

Kibo is a brand we invented. Self-initiated demo, not client work.

The finished ad. 10 seconds, 9:16, sound on. Every frame AI-generated. The “portfolio demo” watermark is ours — client deliveries ship clean.

The brief we wrote ourselves

Product
KIBO, a fictional matcha starter kit.
Audience
People ready to quit an expensive latte habit.
Constraint
Ten seconds, one quiet kitchen shot, no filmed footage.
Format
Close kitchen prep · voiceover
Niche
Food & Beverage

The opening line

“POV: you quit your $7 latte habit.”

The opening line is 'POV: you quit your $7 latte habit.' It never mentions matcha. Instead it hands the viewer an identity they already want, someone who escaped an overpriced routine, and lets the drink on screen be the proof. The product arrives as the answer, not the pitch.

Why this format. Food sells through appetite, which lives in texture and sound before any claim registers. So we kept the camera close on the prep and let a voiceover carry the line, with no face competing with the matcha for attention.

How it was built

  1. 01

    Hero still

    One image, generated with Nano Banana Pro at 9:16. Framed like a phone photo rather than a product shot: the item in hand, the brand name legible on the label, light coming from somewhere plausible.

  2. 02

    Motion and voice

    That still becomes frame one. Seedance 2.0 takes it to ten seconds of 720p video and generates the audio in the same pass. Spoken lines are written into the prompt, so the delivery is lip-synced from the start instead of dubbed on afterwards.

  3. 03

    Check before keep

    Nothing survives because it looked fine as a thumbnail. We read the stream data with ffprobe, build a contact sheet of frames with ffmpeg, and compare the spoken audio against the script that was written. Takes that drift from the brief get regenerated, not patched.

  4. 04

    Finish

    The approved take is cut and packaged with ffmpeg for wherever it will run. Vertical for Reels and TikTok here; the same take can be reframed for feed or landscape if the brief asks.

Format
9:16 vertical, 1280×720 source
Runtime
10.0 seconds, audio included
Stills
Nano Banana Pro
Motion & voice
Seedance 2.0, image-to-video
Assembly & QA
ffmpeg / ffprobe
On location
Nobody. No crew, no samples shipped

What to watch for

The prep sound

Audio is generated in the same pass as the picture, so the whisking and pouring carry the appetite cue instead of music laid underneath.

The close framing

The shot stays tight on the powder, the water, and the whisk, so the texture of the matcha reads before any word is spoken.

The kit on frame

The matcha box and the whisk stay in frame with the KIBO name visible, so the kit reads as a real object, not a prop.

What this page is not

Every frame here is AI-generated, and KIBO is a brand we invented for this demo. We claim no view or sales numbers, because this ad was never run as paid media. With a real product we'd shoot around the actual kit and a real pour.

We have no published reviews yet, so we are not going to borrow credibility we haven't earned. No client names, no engagement numbers, no invented case studies. What we can show you is the work and the method, in full, and let you judge whether it clears your bar.

We'll write three hooks for your food brand, free.

Get 3 custom hooks — free Or skip ahead and brief us on Fiverr to have the whole ad built — script, presenter and edit.