admin You found an amazing AI image and want to make something similar. The original prompt is rarely shared, but you can rebuild a very close one in under a minute. This guide shows the full process, with a worked example at the end.
Step 1: Save a clear copy of the image
Download the largest version you can find. Screenshots work, but crop away buttons, captions, watermarks and borders so the AI only sees the picture. If the image is very dark or tiny, the description will be vague, so pick a clean, well-lit copy whenever possible.
Step 2: Upload it to an image to prompt generator
Open the free ph-imetix image to prompt tool, drop the image in and choose 5 variants. The tone changes how the prompt is written:
- Vivid for art, illustrations and fantasy scenes
- Direct or Formal for photos, products and realistic images
- Simplified when you want a short prompt to build on
Using Midjourney or Stable Diffusion? Start from the Midjourney or Stable Diffusion version of the tool.
Step 3: Compare the variants
Read the five prompts side by side. Notice the words that appear in most of them, usually the medium, the lighting and the colour palette. Those repeated words are the “core” of the style. The words that change between variants are the parts the AI was less sure about, and the ones you can most freely edit.
Step 4: Edit before you generate
- Change the subject if you want the same style with something new (“a red fox” instead of “a cat”).
- Remove details you don’t care about. Shorter prompts are easier to control.
- Add generator settings: in Midjourney, aspect ratio like
--ar 2:3; in Stable Diffusion, a negative prompt. - Keep the style words in the same order. Moving them can change their weight.
Step 5: Generate and refine
Run the prompt, compare with the original and adjust one thing at a time: light, colour, camera angle or style words. Two or three rounds usually get you very close. Keep notes of what changed, because you’ll build your own prompt vocabulary quickly. Our 10 prompt-writing tips explain what to tweak first.
Why you can’t get the exact original prompt
An image doesn’t contain its prompt, and generators add randomness (the “seed”), their own style and hidden settings. An image to prompt tool describes what it actually sees. In practice that often works better than the original prompt, because it’s written to match the final image instead of the creator’s first idea.
Different generators also read prompts differently. See Midjourney vs Stable Diffusion vs Flux for how to adapt one prompt for each.
Worked example
Reference: a flat illustration of a sunset with a round yellow sun on a dark horizon. One of the generated variants:
“Minimal flat vector sunset poster, glowing yellow sun sitting on a dark horizon, amber-to-violet sky gradient, geometric minimalism, symmetrical composition, calm evening mood.”
Edits for a new image in the same style:
- Subject swap: “a lighthouse silhouette on a cliff” instead of the plain horizon
- Kept: flat vector, amber-to-violet gradient, geometric minimalism, calm mood
- Added in Midjourney:
--ar 2:3for a poster format
Quick checklist
- Clear, cropped image
- 5 variants, right tone
- Keep the repeated style words
- Change the subject, add settings
- Refine one detail at a time
Related articles
- 10 Tips to Write Better AI Image Prompts (With Examples)
- Midjourney vs Stable Diffusion vs Flux vs ChatGPT: How to Write Prompts for Each
- AI Product Photography Prompts: Turn One Photo into Many Scenes
Try it free: Image to Prompt Generator · Midjourney · Stable Diffusion & Flux · Stock keywords · Bulk