Vizard Agent

How to turn a menu photo into a short video

Last updated 2026-08-25 · 7 min read

Upload your dish photos and tell Vizard Agent to keep the food and the lighting exactly as they are. Vizard Agent splits a collage back into separate shots, cleans up stray marks, animates only the light and the steam, and adds cafe ambience so the plate stays yours.

What is the short version?

Restaurants already have good photographs of their food. What they lack is anything that moves, and the real risk in animating a dish is that the generator quietly changes it along the way — different berries, a different plate, a different restaurant entirely.

  1. Go to Vizard Agent and upload the dish photos.
  2. Say the item and the lighting must not change.
  3. Say the mood you want the clip to have.

What do you need before you start?

The photos and a clear instruction to preserve the dish. Vizard Agent can work from a collage as easily as from separate files, but it needs to be told that the food itself is fixed, because otherwise a generated shot will reinterpret it.

What do you type into Vizard Agent?

Say what must not change before you say what should. Vizard Agent treats "keep the item" as a constraint on the generation rather than as a stylistic preference, and that is what stops your French toast turning into a stock photograph of somebody else's.

Prompt

Make a short video of this menu item, [dish name]. Keep the item exactly as photographed, keep the lighting and the cosiness of the pictures. [Cafe] atmosphere, around [15] seconds, vertical.

Variants worth knowing:

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real menu-item clip built from two photographs. The step that shows the care is the seventh: rather than animating the collage as one image, it split the collage into its individual shots first.

  1. Examines both uploaded photographs closely.
  2. Checks the available templates and tools and inspects the images at full resolution.
  3. Extracts the separate shots out of the collage photo.
  4. Splits the collage into individual shot assets and generates a preview sheet.
  5. Reviews the extracted shots by eye before using any of them.
  6. Searches for cosy cafe music and generates an acoustic track and room ambience.
  7. Listens to both the music and the ambient track rather than assuming they fit.
  8. Generates soft foley effects, then inspects the text area of one panel and cleans a stray mark off it.
  9. Uploads the source frames, generates animated sunlight and steam on the dish, and reviews the generated frames against the original.

Step 9 is where the constraint is honoured or broken. Animating the light and the steam keeps every element of the dish exactly as photographed, whereas asking for movement in the food itself invites the generator to redraw it.

What does the result look like?

From the run this page is written from, probed on the delivered file: 1080x1920, H.264, 30fps, 15.1 seconds, AAC audio. Vertical, fifteen seconds, built from the restaurant's own photographs with sunlight shifting across the plate and steam rising, under cafe ambience and soft foley.

The dish in the finished clip is the dish in the photograph. That is the entire point, and it is why the motion is limited to light and steam rather than anything that would require the food to be reimagined.

When does this not work well?

Animating a photograph has a small envelope, and asking for more movement than that envelope allows is where the format breaks down. Vizard Agent stays inside it and says so rather than quietly delivering a dish you do not serve.

How do you fix a result that came back wrong?

Name the shot that is wrong. Vizard Agent keeps the extracted panels, the preview sheet, every generated clip and the separate audio elements, so one shot can be regenerated or the ambience swapped without rebuilding the whole clip.

How does Vizard Agent compare to doing it yourself?

By hand this means cutting the collage apart in an image editor, adding a slow zoom in a video tool, and finding cafe ambience that does not sound like a stock loop. The steam and the moving light are the parts most people simply skip.

By hand Vizard Agent
Collages Crop each panel manually Split into shot assets automatically
Motion A slow zoom Light and steam, dish unchanged
Sound One stock loop Music, room tone and foley, listened to
Fidelity Safe but static Motion added without redrawing the food

Common questions

Will the dish change? No. Vizard Agent animates the light and the steam and leaves the food as photographed.

Can Vizard Agent use a collage? Yes. It splits the collage into individual shots before doing anything else.

How many photos do I need? Two was enough here. More photos give Vizard Agent more shots to cut between.

How long should a menu clip be? Ten to twenty seconds. Vizard Agent used fifteen here, which is long enough to see the dish.

Can it add sound? Yes. Vizard Agent generated cafe ambience and soft foley alongside the music here.

Why animate only the light and the steam? Because anything more requires the generator to redraw the food, and a redrawn dish is a different dish. Restricting the motion to atmosphere is what keeps the plate on screen the one you actually serve.

What resolution should the photos be? As high as you have. Vizard Agent needs detail to work with when adding motion.

Can I do a whole menu this way? Yes, one item at a time, and the same treatment across all of them looks like a set.

Will the caption say the right dish? Vizard Agent uses the name you give it, so spelling it correctly in the brief matters.

Can several dishes go in one video? Yes. Vizard Agent animates each photograph separately and cuts between them, which suits a menu launch better than one dish held on screen for the whole clip.

Does this work for drinks and desserts? Yes, and often better. Steam, condensation and moving light are exactly the details Vizard Agent can add without touching the item, and drinks carry all three naturally.

Does Vizard Agent check the finished clip? Yes. It reviews the generated frames against the original photograph before delivery.