A spec reel for a fictional matcha brand: a granite stone mill grinding, brushing, sifting and a tin with a personalised label. It follows your brief: vertical, gloved hands only, no music and no voice.
Please watch it with sound. Four shots are AI motion clips, trimmed to their clean seconds. The sifting shot and the label shot are stills with a slow push-in. The shots are joined with short soft dissolves.
These are the start frames, all generated from the same hero frame. With your reference photos of the real mill and tins, this is the step where your mill goes in.
AI video models smear small text, so I never let them generate the label. The tin is filmed blank, and the label text is laid on top as a separate layer that follows the paper and the light. Names, batch numbers and Christmas lines come from a list, and each one renders in seconds.



labelData.ts
{ brand: "GRANIT & BLATT", product: "Matcha", line: "STEINGEMAHLEN",
personal: "für Lena", batch: "CHARGE 01 · 30 g" }
Your photos of the mill, the tins and the workspace become a fixed set of start frames per angle. Every clip starts from one of them, so the mill is always the same mill.
Each clip is image-to-video from a start frame, 6–10 s, with one simple action: the stone turns, the drawer opens, the brush moves. I keep only the 2–4 s where the motion is clean.
Is the stone rigid? Does it turn around its centre? Does the handle move with it? Are there five fingers and only one pair of gloved hands? Is the camera steady? If a take fails, it gets redone, not hidden in the cut.
Your reference footage sets the speed and direction of each motion: how fast the stone turns, how the drawer slides. I match the prompts and the cut to it.
Tin labels, printer output and scale displays are added as sharp layers on top of the footage and never generated, so text never warps.
Cut, dissolves, sound and export are one template, so the 58 reels share the same rhythm and loudness. A new batch is new clips plus one config file.
1080 × 1920, 30 fps, H.264. Every start frame and clip was checked at full size before it went into the cut.
Only one gloved hand per shot, five fingers and a stable grip. No face, no second pair of hands.
ASMR sound only, one layer per action, mixed to a steady level for Reels. No music, no voice.
Same granite, same handle, same oak tray, same window light and the same matte green across all six shots.
I'm in Thailand (UTC+7), six hours ahead of Germany, so feedback sent in your afternoon is usually worked in by your next morning.
The first batch, including the Christmas videos, can be planned backwards from the end of November. I'd start with your reference photos and footage, then lock the start frames with you before any clips are made.