A teleprompter for recipe videos — the script waits while you knead

A recipe video is instruction delivered while your hands are busy doing the thing you are describing. That is the whole format — and it is exactly what a teleprompter is worst at, because every other prompter expects a free hand to scroll with. Yours are in the dough.

The order is the content. A recipe said out of sequence is a different recipe: the oven to 220°C before the resting, the salt after the bloom, the pan already hot. Quantities too — “some flour” makes a video nobody can cook from, and 320 g is not a number you trust to memory across a forty-minute shoot.

There is a pacing problem peculiar to food, too: the dish sets the tempo, not you. Dough needs its two minutes of kneading whether the video wants them or not; caramel will not hold while you find your place; a hot pan punishes a lost line within seconds. A prompter that scrolls at a fixed speed is wrong twice a minute in a kitchen — running ahead through the long jobs, trailing behind when you talk your way through a quick one — and you cannot reach out to correct it, because correcting it means touching a screen with dough on your hands.

And half the sentences want the food in frame, not you. “Fold it like this” means nothing said to camera; it means everything said over the bowl. The switch from face to pan comes mid-sentence — exactly where stopping a recording ruins it.

Under all of it, the kitchen is loud. The extractor fan, the sizzle, the mixer — a phone microphone at the far end of the worktop hears the room, not you.

Most food creators cook first and talk later — a voiceover written to fit the footage. It can work; it also flattens the delivery, and the timing never quite fits the pictures, so the edit becomes a session of nudging clips to match sentences.

The voiceover route has a quieter cost as well: the recipe stops being checkable. When the words are written after the cooking, nobody re-tests that the numbers in the narration match what actually went into the bowl, and the comment section finds the mismatch within a day of posting. And a fixed-speed teleprompter app fails in this kitchen for the reason above — the scroll does not know the dough. It carries on through the knead, and by the time your hands are free the script is forty lines ahead and the take is dead.

The other route is a step per clip: read the step off a card, film the step, stop, next card. A single bake becomes thirty files, and the assembly takes longer than the cooking did.

Some tape the printed recipe to a cupboard door behind the camera. The eyes drift off the lens with every glance, and by the end of the shoot the printout carries floury fingerprints and the middle steps are memorised wrong anyway.

Talk2Camera scrolls the recipe as you speak it. Voice follow listens on the device: while you are talking, the script moves with you; the moment you stop to knead, it stops too. It waits through the whole fold — thirty seconds or two minutes — and picks up on your next word. Skip a line because the step is obvious on camera, and it catches up. Your hands never touch the screen.

When a sentence needs the food, you flip to the rear camera mid-sentence and keep reading — over the bowl, into the pan. The recording never stops through the switch, so there is one continuous file rather than thirty. Flip back when the point is made.

For the noise, pick your microphone in the app: a clip-on transmitter or any external input, chosen before the take rather than regretted after it. Switching the input re-routes the live capture without interrupting a take.

Afterwards, trim on a timeline you zoom with a pinch. At full magnification one pixel is milliseconds, so the cut between “into the oven” and “forty minutes later” lands on the exact frame the door shuts — no dead air served to the viewer, no step clipped short.

And because the script is the recipe, the numbers on screen are the numbers you say. You wrote 320 g once, when you tested the bake; on camera you read it, over the bowl, at the moment the flour goes in. The video can be cooked from — which is the difference between food content that gets watched and a recipe that gets made, saved and sent to a friend.

A focaccia video, start to finish. The script is the recipe itself: gram quantities, water temperature, three fold cycles, the oven at 220°C. Intro to camera; flip to the rear camera on “this is the flour we are using” and keep reading over the bowl.

First fold: hands in the dough, so the talking stops — and the script waits. Lift, turn, speak the next line, and it is moving again. The extractor fan goes on for the bake; the clip-on mic was selected before the first word, so the take does not care.

One file at the end. The proving gaps are trimmed out with the timeline zoomed right in, each cut landing where the cloth lifts. What is left is a recipe someone can actually cook from — every number said exactly, every step shown the moment it is said.