AI food photography: prompts, light and styling that make food look delicious

Where AI food photography fits, and where it doesn't

Food images do two different jobs, and AI fits them very differently. The first job is attention: a launch teaser, a social post, a seasonal campaign, a mood board for a new menu. Here the image sells an idea or a feeling, and AI is excellent. You can try ten concepts and three lighting moods before a stylist would have finished shopping for props.

The second job is representation: the photo on a menu, a delivery app or a package. That image is a promise about what the customer will get. A generated dish that looks better, bigger or different from the real one is a misleading claim, however good it looks. For that job, AI is a retouching and set-building tool around a photo of the real plate, not a replacement for it.

Everything below applies to both jobs. The techniques that make a generated dish look delicious are the same ones you use to improve a real one.

Appetite appeal: the cues that make food look edible

People judge food in a photo by surface cues that signal temperature, freshness and texture: is it hot, is it crisp, is it juicy, was it just made? Food photographers spend most of their effort on exactly those cues. Models respond to them too, because they are concrete and visible, unlike "delicious" or "mouth-watering", which describe a reaction rather than a picture.

This five-second clip has almost no subject matter: honey and pancakes. What makes it work is that every word describes something you can see. "Glistening" asks for specular highlights on the honey. "Macro" brings the camera close enough to show the bubbly crumb on the pancake edges. "Pouring slowly" gives the honey time to pool on top and then run down the side of the stack.

Here are the cues worth naming, and words that reliably produce them:

The pizza at the top of this page is what those cues do for a realistic shot. Its prompt is close to a checklist from the table: texture, gloss, steam, a light direction and an angle.

Nano Banana 2 · 20 credits

Prompt anatomyCopy

Wood-fired margherita pizza with a blistered, charred crust and glossy melted mozzarella, fresh basil leaves, on a dark slate board, side light from a low window, wisps of steam, 45-degree angle, dark and moody food photography

• Subject what the picture is of
• Detail materials, wardrobe, texture
• Setting where, and what surrounds it
• Light source, direction, time of day
• Camera lens, angle, framing, focus
• Style medium, genre, reference

Every cue shows up. The crust has the dark leopard spotting of a wood-fired oven, uneven from one side of the rim to the other, the mozzarella has melted into glossy pools with orange oil at the edges, and faint steam rises against the dark wall. The low window on the left rakes across the crust and leaves the far side of the room in shadow. The model also styled the table without being asked: a terracotta bowl of basil, a stoneware jug, a printed linen napkin and an old knife. It reads as a real food photograph, which is the point, and also the warning: it is a pizza, not your pizza, so it belongs on a mood board or a social post, not on a menu.

How to light AI food photos: backlight and side light

Food is rarely lit from the camera position. Front light flattens texture and kills the shine on sauces. Food photographers mostly use two directions instead:

• Backlight (light behind the food, facing the camera): makes steam, syrups, drinks and anything translucent glow, and puts a bright rim on edges. The classic look for drinks and hot dishes.
• Side light (light from the left or right, often low): rakes across the surface and reveals crust, crumb, grain and seeds. The classic look for bread, pastry and meat.

A porcelain teacup containing a miniature stormy ocean with a tiny sailing ship riding a wave of tea, the teabag tag flying like a flag, dramatic lighting, macro

A croissant made of translucent glass, inside it a swirling purple galaxy with tiny stars, on a marble counter next to an espresso cup, macro food photography, dramatic rim light

The teacup asks only for "dramatic lighting", and the model chose a dark, low-key set with a window behind the cup, so every droplet of spray catches the light. The glass croissant shows hard side light: one bright direction, a long shadow across the marble and a bright edge on the glass. Note the steam rising from the coffee cup behind it, added without being asked; it is a strong "freshly made" signal.

Useful light phrases for food, from dark to bright:

• "Dark and moody, low-key, window light from behind": rustic dishes, coffee, chocolate, red wine.
• "Side light from a low window, long soft shadows": bread, pastry, breakfast.
• "Bright and airy, diffused daylight": salads, brunch, desserts, healthy brands.
• "Hard summer sunlight, crisp shadows": cold drinks, ice cream, picnic food.

Hero angle, 45 degrees or overhead flat lay?

The camera angle should follow the shape of the dish. Ask where the interesting part is: on the side (layers, height), on top (arrangement, toppings) or both.

An ice cream cone where each scoop is a miniature planet, Saturn with rings dripping caramel, Earth with whipped-cream clouds, pastel studio background, playful product photography

A sushi platter arranged as a tiny Japanese zen garden, raked rice sand, a bonsai made of wasabi, a miniature torii gate of chopsticks, overhead shot, soft daylight

The ice cream cone is all height, so a straight-on shot against a plain pastel backdrop shows every scoop. The sushi platter is all arrangement: from overhead, raked rice becomes a graphic pattern and the chopstick gate reads as a shape. The pizza at the top breaks the table on purpose: one pie at 45 degrees shows the height of the puffed crust, which an overhead shot would flatten. One caution from the sushi image: the model hung a little sign with the characters for "sushi" and a malformed character between them. Check any lettering a model adds.

Food styling, props and plating in the prompt

Styling is everything around the food: the plate, the surface, the cutlery, the napkin, the second cup of coffee. Models know common food-styling conventions well, and they will apply them unasked. That can be a gift or a problem.

Nano Banana 2 · 20 credits

Prompt anatomyCopy

A miniature city built on a stack of fluffy pancakes, maple-syrup rivers, butter-pat skyscrapers, blueberry cars, tilt-shift photography, morning light

• Subject what the picture is of
• Detail materials, wardrobe, texture
• Camera lens, angle, framing, focus
• Light source, direction, time of day

The prompt names the food and the concept, and nothing about the set. The model supplied a whole breakfast table anyway: a syrup jug, a steaming coffee cup, a knife and fork, a linen napkin, scattered blueberries and a bright window behind. That is textbook styling, and it is why the image feels warm. It also added lettering to the butter towers and reused one name, and the tilt-shift look barely came through: the background is blurred, but there is no miniature-style focus band.

The rule: if you care about the set, specify it. A few styling principles worth writing in:

• One hero, few supporting props. Props should come from the dish's story (its ingredients, its cuisine, how it is eaten), not from a generic kitchen.
• Odd numbers and negative space. Three berries look arranged by a person; four look like a grid. Leave clear space where an ad's headline will go.
• Color contrast. Blueberries against golden pancakes, green herbs on a red sauce. A complementary garnish makes the main color look richer.
• Quiet tableware. Matte, neutral ceramics and plain linen keep attention on the food.
• Garnish only with what is in the dish. A sprig of an herb that isn't in the recipe is a small lie on a menu.

Motion for social: pours, drips and steam

Short, slow, satisfying motion is a staple of food content on social: a pour, a drizzle, a cheese pull, steam curling off a bowl. These clips work because they show texture and temperature better than any still, and they loop well.

The honey clip earlier in this guide was generated directly from text on PixVerse V6, as a five-second 9:16 vertical, for 57 credits. The camera is effectively locked off, which is right for this kind of shot: the motion belongs to the food.

For your own dish, start from a photo of it and use image-to-video, so the food stays exactly as served and you only direct the motion. Here we animated the pizza still on the same model. The prompt describes only what moves, because the image already fixes everything else.

Start frame

→

steam, gloss and a push-in

Image promptNano Banana 2 · 20 creditsCopy

Wood-fired margherita pizza with a blistered, charred crust and glossy melted mozzarella, fresh basil leaves, on a dark slate board, side light from a low window, wisps of steam, 45-degree angle, dark and moody food photography

Motion promptPixVerse V6 · 5 s clip · 57 creditsCopy

Steam rises slowly from the pizza, the mozzarella glistens, slow push-in, then settles

The pizza itself stays the same pizza: same crust spots, same basil, same pools of cheese. Steam drifts up on the left and the cheese catches the light. The push-in is stronger than "slow" suggests: by the last second the frame is a tight close-up on the center of the pie, the props have left the frame, and the motion only eases off at the very end. That works for a vertical crop on social; if you need the whole table in the final frame, ask for a locked-off camera or a gentler move and check the last frame, not just the first. It cost 57 credits.

Our image-to-video prompting lesson covers how to write motion prompts, and the guide to AI video ads for social covers hooks and testing.

Concept food: surreal shots that stop the scroll

Surreal food is where AI is unmatched: a burger made of clouds, a bowl of ramen as a swimming pool, a coffee bar run by an octopus. These are attention images for social and campaigns, not menu photos, and they live or die on one rule: keep the food recognizable.

A gourmet burger where every layer is a different kind of cloud: cumulus bun, stormy-grey patty with tiny lightning, sunset-coloured sauce, floating above a plate, surreal food photography

A tiny person in a swim ring floating in a giant bowl of tonkotsu ramen, noodles as waves, a chashu slice as a lounger, bright pop-art colours, playful

An octopus barista behind a copper espresso bar pouring four latte-art hearts at once with four arms while two more arms grind beans and one wipes the counter, warm café light, photorealistic

• The cloud burger works because the stack order is still bun, patty, sauce, bun. You read "burger" before you read "clouds".
• The glass croissant in the lighting section is the counterexample: the galaxy and the light are beautiful, but the shape reads as a glass horn rather than a croissant. When you transform a food, name its defining features ("crescent shape, laminated layers") so the silhouette survives.
• The ramen pool came back as a flat pop-art illustration, and the model painted a grey transparency checkerboard into the background, a known artifact. Check the edges before you post.
• The octopus barista is a convincing café, but the grinder and espresso machine carry real equipment brand names that nobody asked for. In a commercial post, that is a trademark problem.

AI food photos for menus and delivery apps: keep them honest

Advertising rules treat a food photo as a claim about the product. In the UK, the Advertising Standards Authority looks at the overall impression an ad gives, images included, and its guidance says pictures that exaggerate the size or quality of a product, or show extras that don't come with it, are likely to mislead. In one well-known ruling, the ASA cooked the advertised product itself and found the ad showed more pasta and mushrooms than the real thing. In the US, the FTC Act prohibits deceptive advertising in general. Delivery platforms also publish their own photo guidelines, so check the one you list on.

A generated dish on a menu fails this test almost by definition: it shows a dish, not your dish. The honest workflow keeps the food real and uses AI for everything around it.

1. 01

   Photograph the real plate

   Plate it exactly as served, same portion and garnish. A phone in daylight is enough. Use the angle from the table above that suits the dish.

2. 02

   Edit around it

   Use an image-editing model with the photo as the reference to change the surface, background and light. Say explicitly to keep the food unchanged.

3. 03

   Compare against the plate

   Put the edit next to the original. Same portion, same components, same garnish, no added steam on a cold dish? If the food changed, redo the edit.

4. 04

   Keep the originals

   Store the source photo with the final, so you can show how any menu image was made.

What AI food photography costs

Every food image and clip in this guide, with what it cost to render. The spread is wide: the cheapest images cost 3 credits and the most expensive 20. The two clips are in the last rows.

A sensible workflow is to explore concepts, angles and light on the cheapest models, then render the chosen direction on two or three stronger models and keep the best. In PromptLab, the Benchmark runs one food prompt on up to four models side by side, and our four-model comparison shows how differently models read the same brief. Current prices are in the model catalog.

Questions people ask