Marketing

AI Image Prompt Generator

Image models follow structure far better than they follow paragraphs. Describe roughly what you want and get a field-by-field prompt covering subject, composition, lighting, palette, lens, texture, and what to exclude, as XML or JSON, ready to paste into your generator. This tool writes the prompt, it does not render the image itself.

  • Free to use
  • No sign-up
  • Instant results
0
Image prompt
Your result will appear here Add your input, then write image prompt

How it works

  1. 01 Describe the image you have in mind, including the style if you have one.
  2. 02 Pick a format: XML, JSON, a single dense line, or maximum detail.
  3. 03 Generate the structured prompt.
  4. 04 Paste it into Midjourney, DALL-E, Stable Diffusion, Flux, or similar.

What makes an image prompt work

The difference between a flat render and a striking one is almost never prompt length. It is whether each field carries real visual information.

Do

  • Name the light's direction and quality: "low sun raking from frame left, hard-edged shadows"
  • Use actual hues — ochre, oxblood, cold slate — instead of "vibrant colours"
  • Give the camera real values; 35mm at f/8 and 85mm at f/1.4 are different pictures
  • Keep one subject and one focal point
  • Anchor scale against something known when the subject is very large or very small

Don't

  • Don't stack four style references — they average into mush
  • Don't contradict yourself; a wide establishing shot can't also resolve an iris
  • Don't rely on adjectives like "beautiful", "amazing", or "masterpiece" — they carry no visual information
  • Don't leave the negative field empty on people, hands and text are where models fail most

Frequently asked questions

Structure separates the fields. In a paragraph, "golden hour" and "blue palette" compete inside one run-on sentence and the model averages them. As tagged fields, lighting and palette land in their own slots, so you get the picture you asked for and can edit one field without disturbing the rest.

XML is the safest default and reads well when you are hand-editing. JSON suits anything programmatic, like a batch script or an API call. One-line is for Midjourney's prompt box and other places a single string is all you get. Max detail adds layered foreground/background description and three variations to try.

No. It writes the prompt. Paste the output into an image generator such as Midjourney, DALL-E, Stable Diffusion, or Flux to render the picture.

They don't parse XML the way a program does, but the tag names act as strong labels and the line breaks stop fields bleeding into each other. If your generator prefers plain text, the One-line mode gives you the same content as a single dense string.

Things to keep out of the frame. It is filled with the artefacts that actually go wrong for your subject, such as warped hands on people or illegible signage on street scenes. Most generators take it as a negative prompt; Midjourney uses --no.

Related tools