Dish photo for delivery apps
Delivery platforms measure conversion by the photo. Tight framing, neutral background, warm light: the edit node changes only that, never the contents of the plate, otherwise the customer is disappointed at delivery.
AI food photography for restaurants
The phone snapshot taken in the dining room becomes a menu photograph, then a vertical story, then a shot of rising steam. The dish itself does not change. Visible here without an account.
Interactive demo
Real workflow, real results. Pan and zoom freely.
Use this workflowThe raw photo enters the editing node, which only changes the light and the background, then it is repurposed vertically and animated. Open the nodes to read the prompts.
At a glance
- What the workflow produces
- 2 images in 1:1, 9:16 · 1 video of 5s in 1080p, 1:1
- Canvas structure
- 7 nodes wired by 5 connections
- Models wired in
- Flux Kontext, Seedream 5 Pro, Kling 2.5 Turbo
- Cost of one full run
- about 37 credits, roughly €0.37
A menu, a website, a delivery page, an Instagram account: a restaurant needs dozens of visuals a year, and a food shoot costs more than it returns for a menu that changes every season. The phone snapshot taken in the dining room, on the other hand, already exists.
The whole workflow rests on one very firm editing instruction: change nothing on the plate. Left free, the model adds herbs, tidies the plating and returns a dish better looking than the one leaving the kitchen. The photo becomes a lie, the guest is disappointed at the table, and the owner can no longer use it. So only the light, the background and the framing change.
Flux Kontext is the editing model that respects the original subject best, Seedream 5 Pro handles the other formats, and Kling 2.5 Turbo raises the steam for five seconds of video. The guide writing a prompt that works explains how to frame an edit without letting it drift.
How it works
- 1
Take the photo in the dining room
A sharp photo of the dish as it is served, even with yellow light and a cluttered table. It is the background and the light that get redone, not the dish.
- 2
Write the instruction
It always starts with what must not move: same quantity, same plating, same crockery. Then what changes: the background, the light, the frame.
- 3
Repurpose the formats
The menu shot runs into a second image node that turns it vertical, with empty space for a caption. Two formats, one single shot.
- 4
Add the steam
Five seconds are enough, still camera. Steam gives a static shot the life that stops the thumb, without turning the photo into an advert.
Variations and use cases
The same canvas, tuned for different needs: each variation is two or three lines of prompt away.
Menu visuals for a restaurant
A menu that changes every season does not justify a shoot. Thirty dishes photographed on a phone in the dining room run through the same canvas, with the same light, for under two euros of credits.
Pastry and bakery
Fine textures are what Flux Kontext preserves best in editing. Ask for raking light: that is what brings out the lamination and the grain of the sugar.
Cocktails and drinks
Glass is the hardest subject, because the model recomposes reflections. Shoot against a dark background and check the liquid level on the render, that is the detail that betrays an edit gone too free.
Vertical story and a steam shot
The second Image node reframes the photo to 9:16 with room for a headline, and the Video node lifts steam for five seconds. Two formats and a moving shot from a single phone snapshot.
Common mistakes
Letting the model improve the plating
Without a firm instruction it adds herbs and straightens the plate: the photo lies about what leaves the kitchen. The instruction always starts with what must not move.
A source photo shot from too far away
Editing does not create detail that is not there. Move the phone closer, even if it crops the plate, rather than photographing the whole table.
Changing background and framing at once
Two changes in one instruction give an unpredictable result. Redo the background first, reframe afterwards in a second node.
Recommended models
Flux Kontext
Seedream 5 Pro
Kling 2.5 TurboFrequently asked questions
Does the dish really stay identical?
That is the role of the editing instruction, which forbids any added garnish. Check every render anyway: an editing model is still a model, and a retouched dish has to match what leaves the kitchen.
Is a business account needed?
No, the canvas is the same for everyone. The visuals you generate are yours and can be used on a menu, online or in advertising with no particular credit.
How many dishes can be processed?
As many as you have photos, each dish costing a few cents of credits for two visuals. A menu of thirty dishes comes to less than two euros, detailed on the calculator.
What if the model adds ingredients?
Run it again with a harder opening line: same plate, same quantity, no added garnish. Editing models follow bans placed at the top of the instruction, much less those slipped in at the end.
Can the visuals be used in advertising?
Yes, the images are yours and work on a menu, online or in paid ads. The constraint is commercial rather than legal: the photo must stay true to the dish.
Do I need a camera?
No, a recent phone is enough, even under the yellow light of a dining room. That light is exactly what the workflow redoes, starting from a sharp photo of the dish as served.
What does a full menu cost?
About thirty-seven credits for one dish turned into two visuals and a video, and six credits for a simple edited menu photo. A thirty-dish menu stays under two euros.
AI food photography for restaurants
Use this workflowFree account, no card required. The workflow opens pre-filled in your canvas.