AI image prompt examples: a template and 8 prompts to copy

Good AI image prompts follow one pattern: a sentence or two that names the subject, the setting, the style or medium, the light, and the framing. Even a short prompt can follow it: “a red panda astronaut floating in space, studio lighting” names a subject, a place, and a light. Below are a template, eight examples by use, and the choices that belong in request settings rather than in the prompt.

The red panda prompt is a request example in Sume's Image API docs. The docs' other example prompts and the request fields come from those docs and the request schema in the Sume API reference, read on 2026-09-28. The template and the longer examples are illustrations, not wording any model is documented to follow.

How do I write a prompt for image generation?

Build it in five parts. Each line adds one part to the docs example:

• Subject: who or what. “A red panda astronaut.”
• Setting: where and when. “Floating in space above the Earth.”
• Style or medium: a photo, a 3D render, a watercolor, a flat illustration. “A 3D render.”
• Light: the source and the mood. “Studio lighting.”
• Framing: shot size and position. “Full body, centered, with empty space around it.”
• All five together: “A 3D render of a red panda astronaut floating in space above the Earth, studio lighting, full body, centered, with empty space around it.”

What are some AI image prompt examples?

Eight prompts built on the template. Four grow out of short prompts in Sume's docs: “a product photo”, “a dramatic portrait”, “a landscape photo”, and “a cute cat”.

• Product: “A product photo of a glass bottle of olive oil on a white marble counter, soft morning window light from the left, 45-degree angle, shallow depth of field.”
• Portrait: “A dramatic portrait of an older fisherman in a yellow raincoat, hard side light against a dark background, close-up at eye level.”
• Landscape: “A landscape photo of a mountain lake at sunrise, mist over the water, pine trees in the foreground, wide shot, golden light.”
• Pet: “A cute cat asleep on a sunny windowsill, fur lit from behind, close-up, soft background.”
• Food: “Overhead photo of a bowl of ramen with a soft-boiled egg and scallions on a dark slate table, soft window light.”
• Interior: “A cozy reading nook with a green armchair, a floor lamp, and a full bookshelf, warm evening lamp light, wide shot at eye level.”
• Illustration: “A watercolor illustration of a fox reading a book under an oak tree, soft pastel colors, the whole scene in view.”
• Poster art: “A flat illustration of a bicycle in a city park in autumn, a limited palette of teal and orange, empty space at the top for a title.” Add the title yourself; AI image with text generator covers words in images.

What goes in the settings instead of the prompt?

Anything the request has a field for. A model only accepts the values its catalog lists, and a parameter the selected model does not list is rejected with 400 unsupported_parameter rather than silently dropped. Check each model's lists with GET /v1/images/models. The example below returns 200 with the images within 30 seconds, or 202 with a job to poll when the generation takes longer.

curl -X POST "https://api.sume.com/v1/images" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-image-2.5",
    "prompt": "A 3D render of a red panda astronaut floating in space above the Earth, studio lighting, full body, centered, with empty space around it.",
    "aspect_ratio": "16:9",
    "n": 4
  }'

How do I test and improve a prompt?

Change one part of the template at a time and keep every setting the same between runs, so you can see what the change did. Set n to compare several results of one prompt in a single call; on most models the ceiling is 4 today. No image model serves seed yet, so each run is a new take. Download the results you like, because result URLs are Sume-hosted and signed. Each completed image is billed, and a failed or cancelled one is not.

How do I write a prompt that edits a photo?

Send the photo in input_references to a model that edits from reference images, and write what changes and what stays. The docs' edit example is one line: “make this scene look like a watercolor painting”. AI photo editing prompts has examples for backgrounds, light, color, and style.