Image to prompt generator

Two ways this tool reads your image

To be clear about what happens, the result depends on how this site is set up.

When AI mode is available on this site, the image is resized in your browser so its longest side is at most 1024 pixels, sent as a JPEG to a vision model, and described: subject, setting, composition, camera angle and lens feel, lighting, colors, style and mood. The model is told not to identify real people by name. The answer comes back as a prompt in the format of your chosen target. Usage is limited per hour, and if the AI request fails you still get the local result.

Without AI mode, nothing leaves your browser. A local analysis measures what can be measured from pixels: the dominant colors, the average brightness, how saturated and how contrasted the image is, its orientation and the nearest of seven standard aspect ratios. It cannot recognize objects, so it leaves [describe the main subject in a few words] for you to fill in. A note under the widget tells you which mode is active.

What the local analysis actually measures

• Palette: up to five distinct dominant colors, shown as swatches with hex codes and translated into names like “deep blue” or “pale orange”.
• Lighting: dark images become “low-key, moody lighting”, bright ones “bright, high-key lighting”, the rest “balanced natural lighting”.
• Look: muted, natural or vivid colors combined with soft, medium or high contrast.
• Framing: landscape, portrait or square, plus the closest ratio among 1:1, 16:9, 9:16, 4:5, 3:2, 2:3 and 21:9, written as --ar for Midjourney.

A local result for a dark, moody street photo looks like this before you edit it:

[describe the main subject in a few words], low-key, moody lighting, deep blue, charcoal, orange color palette, natural colors, high contrast --ar 3:2

Replace the bracket with “a cyclist riding through a rainy street at night, reflections on wet asphalt” and you have a prompt that reproduces the mood and colors of the original quite closely.

Good uses of reverse prompting

Turning an image into a prompt is most useful when you like the feel of a picture and want to build on it, not copy it:

• Recreate the lighting and palette of a reference photo for your own subject.
• Keep a consistent color scheme across a series of blog or social images.
• Learn how a picture would be described in prompt vocabulary.
• Prepare an image-to-video prompt: with Veo 3 selected, the prompt describes motion, camera and an Audio: SFX: line (the local version starts with “Animate the provided image”).

Respect copyright and people's privacy: use your own images or ones you are allowed to use, and do not try to reproduce someone's likeness without consent.

Related tools

Need structured output? Image to JSON prompt returns the same analysis as JSON. To blend a subject from one picture with the style of another, use two images to prompt. If your image contains printed words you want to copy, image to text extracts them with in-browser OCR.