Turn an image into a useful prompt

A reference image can say more than a paragraph of text, but a text-to-image generator only understands words. Turning one into the other is a translation job, and the quality of that translation decides whether you get a useful starting point or a generic result. This guide shows how to read an image like a prompt writer, and how to use the Image to prompt tool without expecting it to do something it can't.

What an image-to-prompt tool can and can't do

An image-to-prompt tool describes what is visible: the subject, the setting, the composition, the light and the overall style. It cannot recover the prompt that originally made the picture, and it cannot know the seed, the model, the artist or the camera. Any wording about lenses or "film look" is an interpretation of what the picture appears to show.

That distinction matters. If you treat the output as a recovered recipe you will be disappointed. If you treat it as a careful written description that you can edit, it becomes a fast way to start.

The five parts of a useful image prompt

1. Subject: who or what is the picture about? Be concrete: "a ceramic bowl of oranges", not "some fruit".
2. Setting: where is it, and what is around it? "On a rustic wooden table beside a window".
3. Composition: how is it framed? Close-up, wide shot, centred, off-centre, top-down.
4. Lighting and mood: soft window light, hard midday sun, dim candlelight, calm, dramatic.
5. Style and palette: editorial photograph, watercolour, flat illustration, muted warm colours.

The tool's output fields follow this structure on purpose, so you can keep the parts you like and rewrite the rest.

A worked example

Imagine a photo of oranges in a bowl on a table by a window. A weak prompt would be:

oranges in a bowl

A stronger prompt built from the five parts:

A ceramic bowl of oranges on a rustic wooden table beside a sunlit window, three-quarter view, soft diffused light from the left, shallow depth of field, warm palette of oranges and cream, editorial food photography.

(This is an example written for this guide, not live output from the tool.)

Notice what changed: nothing is invented that isn't visible, but every part of the scene is named.

Choosing the output style

• Detailed gives you a long, descriptive prompt. Good when you want to reproduce a scene closely.
• Concise gives you a single short line. Good when a generator works better with fewer words.
• Structured separates subject, setting, lighting and so on. Good when you want to remix individual parts.

Choosing a target wording

The tool offers general, Midjourney-style, FLUX-style and Stable Diffusion-style wording. These are best-effort presets: FLUX-style prefers full descriptive sentences, Stable Diffusion-style uses short comma-separated phrases and can come with a negative prompt. They are not exact syntax for any product, and generators change over time, so always check the generator's own documentation.

Editing the result

Read the output against your image and fix anything wrong. Then try one change at a time: swap the lighting, change the palette, or move the camera. The suggested variations are a quick way to explore. If your prompt is nearly right but rough, paste it into the Prompt enhancer to sharpen it without changing your constraints.

Respecting the image and the people in it

Only upload images you have the right to use. The tool is instructed not to identify real people or guess sensitive traits, but you remain responsible for how you use any generated prompt, especially for images that show people.

Quick checklist

• Did I keep concrete nouns for the subject?
• Are the light, composition and palette named?
• Did I remove anything that isn't actually in the image?
• Did I check the target generator's current guidance?