AI food photography: restyle your real dish or generate one

AI food photography works two ways. For a menu, a delivery listing, or a product page, photograph the real dish and let an AI image model restyle the scene around it (backdrop, light, props) with a prompt that says the food stays exactly as it is. Where no specific dish is promised, such as a blog header or a mood image, you can generate the food from a prompt alone. An edit redraws the whole picture, food included, so compare every result with the plate you serve.

On Sume, both are POST /v1/images requests. The facts below come from the Image API docs and the Sume API reference, read on 2026-09-28.

How do I restyle a photo of my real dish with AI?

• Shoot the plated dish from the angle you want to publish, in even light.
• Put the photo at a public HTTPS URL and send it in input_references to a model that edits from reference images, with aspect_ratio: "auto" to keep its shape.
• Say what stays first (the food, the portion, the garnish, the plate), then what changes around it. How to change the background of a photo with AI has more backdrop prompts.
• Ask for n: 4 and keep the version whose food matches the plate you serve.

curl -X POST "https://api.sume.com/v1/images" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "google/nano-banana-pro",
    "prompt": "Keep the dish identical: the same food, portion, garnish, and plate. Replace the background with a dark wood table, add a linen napkin and a fork beside the plate, and light it with soft window light from the left.",
    "input_references": [
      { "type": "image_url", "image_url": { "url": "https://example.com/dish.jpg" } }
    ],
    "aspect_ratio": "auto",
    "n": 4
  }'

What should a food photography prompt include?

Five parts, and only the last one is about the food:

• Angle: overhead (a flat lay), 45 degrees, or eye level.
• Light: soft window light from one side, bright daylight, or warm evening light.
• Surface and backdrop: marble, dark wood, linen, or colored paper.
• Props: cutlery, a napkin, raw ingredients, a drink.
• The food: for a real dish, what stays (the food, the portion, the garnish, the plate). Don't ask for a garnish or sauce the dish isn't served with.

What are some prompts for AI-generated food pictures?

Where no menu item is promised, the whole picture can come from the prompt, with no reference photo. These are illustrations to adapt, not wording any model is documented to follow:

• “Overhead photo of a bowl of ramen with a soft-boiled egg, scallions, and nori on a dark slate table, chopsticks resting on the bowl, soft window light from the left.”
• “45-degree photo of a stack of blueberry pancakes with maple syrup running down the side, on a white plate on a light wood table, bright morning daylight.”
• “Eye-level photo of an iced latte in a tall glass on a café counter, warm afternoon light, background softly out of focus.”
• “Flat lay of fresh pasta ingredients on a floured wooden board: eggs, flour, a rolling pin, basil, and tomatoes, even daylight.”

Should I edit my dish photo or generate one?

Use a generated picture only where it doesn't stand for a specific dish. On a menu, a delivery app, or a product page, the photo tells customers what they will get, so start from a photo of the real dish and check the result against the plate. If you list on a delivery app or marketplace, read its photo rules before you post an edited photo.

Can AI enhance or enlarge a food photo?

Yes, in two different ways. An edit such as “brighten the photo and warm the light; keep the food and the plate exactly as they are” redraws the picture, so check the food again afterwards. To make a photo bigger, for example for print, send its public HTTPS image_url to Image Upscale 1.0 at POST /v1/image-upscale-1.0/upscale with an upscale_factor from 1 to 4 (default 2). API pricing lists it at $0.20 per image, plus a 5.5% agent fee by default, and describes its output as generative, so check the enlarged photo too. AI image upscaler API covers the job.

What does it cost, and what are the limits?

• Each completed image is billed, so n: 4 is four images, and a failed or cancelled generation is not billed. Each model's price is in the pricing line of GET /v1/images/models/{model_id}/endpoints, and usage.cost in the response is the USD amount billed.
• The call waits up to 30 seconds and returns 200 with the images; a slower one returns 202 with a job to poll.
• Result URLs are Sume-hosted and signed, so download the photos you keep.
• For packaged products in lifestyle scenes rather than plated food, see AI lifestyle product photography.