AI video for restaurants — dish hero shots and ambience B-roll

Food accounts die of irregularity, not of bad taste. One good month of posting, then service gets busy, then three weeks of nothing, then a photo of a chalkboard. The constraint is never ideas — it is that a proper dish shot means a quiet hour, a window seat, someone holding a bounce card, and a kitchen that has better things to do. This is how to build a repeatable weekly set on eroq, what it costs, and which shots you should still take with your own hands.

The signature dish goes into Elements

Before anything else, put your actual plates into Elements — the library of reusable places, objects and styles, up to 50 per account, each one @mentionable in a prompt. An Element holds a name, a description and reference photos.

The photos are the point. With references attached, the engine works from your dish instead of inventing a lookalike, and with a single photo the render becomes image-to-video and animates the real plate. Nobody reads a garnish in a five-second clip, but everyone notices when the burger in the ad is not the burger on the pass.

Build one Element per signature dish and one for the room itself. After that, every clip this quarter is an @mention plus a sentence.

Three shots carry a food account

Forget the thirty-second brand film. A restaurant needs a cadence, and a cadence is made of three repeating shapes:

1. The dish hero — the plate alone, slow push-in or a tight orbit, steam if there is steam, shallow focus at f/1.4.
2. The ambience — the room before service, the pass under heat lamps, rain on the window, the espresso machine. This is the filler that makes the account feel like a place rather than a menu.
3. The hands — pouring, tearing, dipping, the first cut into something. Handheld, daylight, no faces needed.

Set the director's rack once and leave it: Film type "Commercial", Tempo "Calm", Camera gear "35mm film", Palette "Warm film", Lighting "Window light". Then change only the lens and aperture between shapes — Macro at f/1.4 for the hero, clean and sharp at f/4 for the room.

The dish hero prompt

Written the way the engine likes it, one flowing paragraph naming subject, motion, camera, light and mood:

A wide ceramic bowl of ramen on a dark walnut counter, slow orbit to the right as steam curls up through a shaft of late afternoon light, soft-boiled egg and chili oil catching the highlights, macro lens at f/1.4, warm film palette, window light from the left with deep shadow behind, calm tempo, commercial look on 35mm film, close-up.

Two habits save credits here. First, use the prompt enhancer — the Enhance and Surprise me buttons rewrite a rough line into a director-grade prompt and cost nothing. Second, shoot takes rather than re-renders: each scene accepts 1 to 4 takes, so when a dish matters you ask for two at once and keep the better one instead of discovering the problem, editing, and paying for a whole second round.

A weekly cadence you can keep

One recording session a week, three clips out of it, scheduled across seven days:

• Monday — dish hero, vertical 9:16, the week's special.
• Wednesday — ambience, whatever the weather is doing.
• Friday — hands, and the still that goes on the menu board.

Shoot each clip in the ratio you publish. Video renders 16:9 wide, 9:16 vertical and 1:1 square, and images add 21:9, 4:3 and 3:4 for the site banner and the printed insert. Cropping a wide master into vertical throws away the headroom you framed for; the same pipeline is laid out in more detail on the social shorts use case.

What a month costs

One week, with the dish hero shot twice at 5 seconds on Kling 2.5 Turbo for 1080p, two ambience clips at 5 seconds on Seedance 1.0 Lite, three stills on Image One, and 600 characters of menu voice-over on Voice One at 3 credits per 100:

• Dish hero, 2 takes — 2 × 120 = 240
• Ambience — 2 × 60 = 120
• Stills — 3 × 10 = 30
• Voice-over — 600 ÷ 100 × 3 = 18

That is 408 credits a week, so 4 × 408 = 1,632 credits a month — about a quarter of Creator at $49 for 6,000 credits, leaving room for a launch week or a second location.

On Hobby at $15 for 1,800 credits, Kling is out of reach and the swap is Motion One at 100 credits for 5 seconds. The week becomes 200 + 120 + 30 + 12 (the same script on Voice Turbo at 2 credits per 100) = 362, so a month is 1,448 credits and Hobby covers it with 352 to spare. Unused credits roll over forever, and failed renders refund themselves automatically.

When the clip needs its own sound

Veo 3 Fast is the one engine that renders a soundtrack — ambience, effects and dialogue, in a fixed 8-second clip for 256 credits, on Creator and up. For food that is a genuine shortcut: the sizzle, the pour, the room tone arrive with the picture instead of being hunted down in a library. A two-shot sizzle spot is 2 × 256 = 512 credits.

Worth saying plainly, because it trips people up: eroq does not generate music. Veo gives you diegetic sound, not a track. Your bed comes from wherever you license music today, and the assembly — trims, transitions, grade — happens in your editor, since eroq renders clips and stops there.

FAQ

Do I need a photographer to start?

One clean photo per dish is enough to attach as a reference. More angles make the hero shot steadier. What generated video cannot do is invent your actual plate from a text description, so the reference is the whole trick.

Which engine for food?

Kling 2.5 Turbo for the hero when you want 1080p and fluid motion, Seedance 1.0 Lite at 60 credits for the filler, Motion One at 100 credits on any account including free. Veo 3 Fast when the clip needs its own sound.

How long should a food clip be?

Five seconds per shot is almost always right, and it is also the cheapest anchor. Engines go further — Seedance 2.5 runs from 4 to 30 seconds on Creator and up — but a longer take of a bowl of soup is rarely a better one.

Got one good photo of your best plate? Open the video studio, make the dish an Element first, and shoot Monday's hero.