AI menu generator: make the artwork, type the prices

An AI menu generator is most useful for the menu's artwork, not its words: generate the background, borders, and dish photos with an image model, then set the dish names, descriptions, and prices yourself in a layout tool, where you can check and change every character. Generate at a portrait shape for a printed menu and at 16:9 for a TV menu board.

Below is how to make the artwork with Sume's Image API, read on 2026-09-29. The same art-then-words pattern for flyers is in AI flyer generator.

What should the AI make, and what should I type?

Split the menu by what must be exact. Text inside a generated image is drawn, not typed, so a price can come out wrong and every price change would mean a new image. Sume's terms also say generated outputs may be inaccurate, and that you review them before publication.

• AI: the background texture, borders and dividers, a header illustration, and empty panels where text will go.
• AI, from your own photos: dish images that look like the plates you serve.
• You: restaurant name, section titles, dish names, descriptions, prices, allergen notes, and opening hours.
• Ask the prompt for “no text, no letters” and for plain areas of calm color where the menu text will sit.

What size should an AI menu be?

Pick the aspect ratio from where the menu will be seen. A model only accepts the ratios its catalog descriptors list, so read supported_parameters from GET /v1/images/models first. For print, generate at a high tier and upscale; AI image for print has the pixel math for 300 DPI.

How do I generate menu artwork with AI?

Send a prompt to POST /v1/images with the ratio and tier. For an exact pixel size, ChatGPT Image 2.5 takes a custom image_size: both edges multiples of 16, a maximum edge of 3840, a ratio of at most 3:1, and 655,360 to 8,294,400 pixels in total.

curl -X POST "https://api.sume.com/v1/images" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "bytedance-seed/seedream-4.5",
    "prompt": "Rustic Italian trattoria menu background: warm cream paper texture, thin olive-branch border, small hand-drawn tomato and basil illustration at the top, large empty plain areas for text, no text, no letters",
    "resolution": "2K",
    "aspect_ratio": "3:4",
    "n": 3
  }'

Can I use photos of my own dishes?

Yes. Send your photo of the plate as a reference in input_references on a model that takes references, with a prompt for the backdrop and light, and aspect_ratio: "auto" to keep the photo's shape. Reference URLs must be public HTTPS. AI food photography covers the prompts. A dish generated from text alone shows food you didn't cook, so keep menu photos to real plates.

How much does an AI menu cost to make?

A menu is a handful of images: a background or two and a few dishes. Each completed image is billed at the model's endpoint price, the pricing line of GET /v1/images/models/{model_id}/endpoints, and cost_usd × n is what you pay. Failed or cancelled generations are not billed. Result URLs are signed, so download the files you keep. For a looping video on a menu screen instead of a still, see restaurant video advertising with AI.