How to narrate a food video to what the hands are doing

Last updated 2026-08-25 · 7 min read

What is the short version?

Food narration fails when the words and the hands are out of step. A line about lemon zest that plays while someone is still arranging fish reads as generic marketing, and the fix is to find the beats in the footage first and write to them.

1. Go to Vizard Agent and upload your cooking footage.
2. Name the dish and every ingredient exactly.
3. Describe the voice you want — accent, warmth, pace.

What do you need before you start?

The footage and the correct names. Vizard Agent finds the beats in the process itself, so a shot list is unnecessary, but ingredient and brand names have to be given precisely because they are spoken aloud and a wrong one is worse than no name at all.

• The footage. One continuous take works.
• The dish. What it is, in the order it is built.
• The ingredients. Named exactly as you want them said.
• The voice. Accent, tone, how knowledgeable it should sound.
• The length. Thirty to forty seconds for a Reel.

What do you type into Vizard Agent?

Describe the dish and the voice, not the edit. Vizard Agent works out the beats from the footage, so what it needs from you is the vocabulary — the ingredient names — and a clear picture of who is meant to be speaking over it.

Prompt

Variants worth knowing:

• Voice tests first. Several candidates generated so you can hear them.
• A named finishing beat. The zest, the drizzle, the last thing that happens.
• A hero frame. The shot the whole reel builds toward.

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real food reel cut from one take. The step that shows what the method actually is comes in the middle: it zoomed into the frame to confirm the zest sprinkle was genuinely visible before writing a line about it.

1. Inspects and downloads the footage, then builds a contact sheet and reviews it.
2. Locates the key visual beats precisely across the recording.
3. Measures the source audio loudness and checks the sections in question.
4. Extracts a frame grid for each beat — the tin opening, the toast, the first placement.
5. Reviews every beat's grid by eye, including the onion and the zest.
6. Zooms into the sprinkle moment and confirms it rather than assuming it happened.
7. Checks the hero frame at full size and verifies the finishing touches.
8. Searches the voice library, generates voice tests, and listens to them before choosing.
9. Generates the narration, pinpoints the placement beats again, and assembles the track against them.

Step 6 is the discipline. It is very easy to write "a final flourish of lemon zest" over footage where the zest never actually appears in frame, and the only way to avoid it is to look at the pixels.

What does the result look like?

From the run this page is written from, probed on the delivered file: 1080x1920, H.264, 30fps, 39.83 seconds, AAC audio. Vertical, just under forty seconds, cut from a single take of a dish being assembled, with narration in a specific requested accent landing on each stage of the build.

Forty seconds is roughly six or seven beats. Each line has three to five seconds, which is long enough to say something about an ingredient and short enough that the hands are never waiting for the words to catch up.

When does this not work well?

Food reels depend on what the camera caught during the build, and Vizard Agent can only narrate what is genuinely on screen. These are the situations where the method runs into something it cannot fix from the edit alone.

• A beat that was not filmed cannot be narrated. No zest shot, no zest line.
• One fixed wide angle limits everything. There are no close-ups to cut to.
• Speed matters. A rushed build gives each line under two seconds.
• Claims about food are yours. Provenance, sourcing and health claims are your responsibility.
• Accents are approximations. A generated voice suggests a region, it does not come from one.

How do you fix a result that came back wrong?

Name the line or the beat. Vizard Agent keeps the beat map, every frame grid, the voice tests and the narration word timings, so a line can be rewritten or re-timed without recutting the whole reel around it.

• "The zest line is too early." Re-timed against the located beat.
• "Try the other voice." Recut with a different test from the audition.
• "Say the brand properly." The narration is regenerated with the correct name.

How does Vizard Agent compare to doing it yourself?

By hand this means scrubbing the take to find each stage, writing a line for each, recording it, then discovering the words run long and the hands have moved on. The re-timing pass is where most food reels lose their rhythm.

Common questions

Do I need to shoot close-ups? They help a great deal. Vizard Agent can crop in, but a fixed wide angle limits how much detail it can show.

How does Vizard Agent find the beats? It analyses the footage for the key visual moments, then pulls a frame grid for each one.

Will it invent a step that is not on screen? No. Vizard Agent verifies a moment is visible before narrating it, zooming in when unsure.

Can I hear the voice before committing? Yes. Vizard Agent generates voice tests and listens through them before generating the narration.

Can I specify an accent? Yes. It matched a requested Southern English accent on this run, calm and conversational.

Why zoom in to check a single sprinkle? Because narration that describes something the viewer cannot see is the fastest way to make a food reel feel written by someone who never watched the footage. Confirming the pixels costs one step and prevents that.

How long should a food reel be? Thirty to forty seconds. That is six or seven beats with room to say something about each.

Can Vizard Agent use my own voice? Yes, cloned. Vizard Agent uses it the same way it uses a library voice, timed to the same beats.

Does it work for a whole recipe? Yes, but a full recipe needs longer. Forty seconds lets Vizard Agent cover an assembly, not a bake.

Can I get a wide version for a website? Yes. Vizard Agent re-frames from the same beat map and the same narration, so the lines still land on the moments they describe rather than needing to be re-timed.

Does Vizard Agent check the finished reel? Yes. It verifies the narration lands on the beats before delivery.