What "image to prompt" means
You find an image you like and want to make something in the same spirit, but you don't know how to describe it. Image to prompt (also called a reverse prompt) works backwards: you give it the picture, and it gives you the words.
In EGAKU AI this lives in the Analyze tool. Upload an image and choose Reverse Prompt. You get a prompt, a negative prompt, style tags and a guess at what kind of model made it. You can then send that prompt straight to the generator and edit it there.
Two ways it reads a prompt: exact or estimated
Analyze checks the file before it looks at the picture.
- Exact: the settings are still inside the file. Many image generators that follow the Stable Diffusion format save the settings in the PNG itself: the prompt, the negative prompt, steps, sampler, CFG scale, seed, size and model. If those are there, Analyze reads them out word for word. The result is marked as coming from the file's metadata, so you know it's the real prompt, not a guess.
- Estimated: the picture is all there is. Social networks, chat apps and screenshots strip that information, and JPG files rarely carry it. In that case an AI vision model looks at the image and writes a prompt that describes the subject, composition, lighting and style. It also suggests what kind of model it looks like (for example, an anime-style model or a photoreal one). Treat this as a well-informed starting point, not the original.
Tip: if you have the original PNG, upload that instead of a screenshot or a re-saved copy. It's the only way to get the exact prompt.
How to get a prompt from an image, step by step
- Open Analyze and keep the Image tab selected.
- Upload a PNG, JPG or WebP (up to 20 MB).
- Choose Reverse Prompt and press Analyze. Your first image analysis is free.
- Read the result: the prompt, the negative prompt, the style tags, the model guess and, if the file had them, the original settings (steps, CFG scale, seed, size).
- Copy the prompt, or send it to the generator. Change the subject to your own, keep the parts that describe the look, and generate.
Example. From a photo of a ceramic mug on a wooden table by a window, you might get something like: "handmade ceramic mug with a speckled glaze on a weathered oak table, soft morning window light from the left, shallow depth of field, warm muted tones, lifestyle product photography". Swap the mug for your own product and the style carries over.
From a video: get a text-to-video prompt and a motion prompt
Switch to the Video tab to work the same way with a clip. Upload an MP4, MOV or WebM (up to 100 MB). Analyze takes five evenly spaced frames from the video and returns:
- a short summary of what happens,
- a text-to-video prompt to recreate a similar clip from scratch,
- an image-to-video motion prompt: the movement on its own, to use when you already have a starting image,
- style tags and a negative prompt.
The motion prompt is the useful part when you want the same kind of camera move or action on a different picture. Pair it with an image in the image-to-video tab of the generator. For more on that, see how to turn a photo into a video.
Other things Analyze can do with an image
- Photo Critique: notes on composition, lighting, color and story, with three concrete suggestions.
- Ask Anything: ask a question about the image, such as how it was lit or what material something is.
- Style Transfer: pick a target style (watercolor, ukiyo-e woodblock print, 70s film photo and more) and get three ready-to-copy prompts that re-render the image in that style.
- Video Concept: turns a still into a 15-second video idea: a shot breakdown, a motion prompt and a music suggestion.
- Tool Chain Guess: an estimate of which kinds of tools (generator, upscaler, editor) were likely used to make the image.
Tips for better prompts
- Crop to what you care about. If you only want the lighting of a portrait, a tighter crop gives a tighter prompt.
- Keep the style, change the subject. The most reusable parts of a reverse prompt are the camera, light and color words. Replace the subject with your own.
- Clean up the negative prompt. Keep the items that fix problems you actually see. Our negative prompt guide explains what each kind of term does.
- Tighten the wording. A reverse prompt describes everything it sees. Removing details you don't need usually helps. See how to write AI image prompts.
- Learn from it rather than copy it. Use reverse prompts to understand a look and build your own. Don't use them to reproduce someone else's artwork or a real person.
Credit costs and limits
Your first image analysis is free. After that, each analysis costs:
| Mode | Image (credits) | Video (credits) |
|---|---|---|
| Reverse Prompt | 5 | 8 |
| Photo / Video Critique | 7 | 10 |
| Ask Anything | 7 | 10 |
| Style Transfer | 7 | — |
| Video Concept | 8 | — |
| Tool Chain Guess | 8 | 12 |
- These are standard prices. In some regions the amount shown in the app differs.
- Images: PNG, JPG or WebP up to 20 MB. Videos: MP4, MOV or WebM up to 100 MB.
- Only PNGs that still carry their generation settings give the exact prompt. Everything else is an estimate.
- Your analyses stay in your account. They aren't shown in any public gallery.
Have an image you've been wondering about? Open Analyze and try it with your free analysis.