ERA OF MINDS

How to get a prompt from an image for Nano Banana, ChatGPT Images, Seedream and Midjourney

By Era of Minds ·

Short answer

To get a prompt from an image, first check whether the original file still has its prompt saved in the metadata (Stable Diffusion and ComfyUI images often do). If not, upload the image to ChatGPT or Gemini and ask for a detailed text-to-image prompt, use Midjourney's Describe tool, or use a dedicated image to prompt app such as Reprompt on Android. The result is plain text, so you can paste it into whichever generator you use now, such as Nano Banana in Gemini, ChatGPT Images, Seedream, FLUX or Midjourney. Except for metadata, every method gives you a close reconstruction, not the exact original prompt.

Can you get the exact prompt from an AI image?

Sometimes, but usually not. The only way to recover the exact original prompt is if it was saved inside the image file. Tools like Automatic1111, Forge and ComfyUI write the prompt, seed and settings into PNG text chunks, and free viewers such as siutil's PNG Info viewer can read them back. Screenshots, social media uploads and re-saving usually strip that data, and some generators never store the prompt at all.

Everything else is reverse image prompting: an AI looks at the picture and writes a new prompt that should produce something similar. That is still very useful. You get the style words, lighting and composition terms you might not have thought of.

How to get a prompt from an image with ChatGPT or Gemini

Both ChatGPT and Gemini can read images, so this is the quickest method if you already use one of them.

  1. Open a new chat and upload or paste the image.
  2. Ask for a prompt, not a description. For example: "Write a detailed text-to-image prompt that would recreate this image. Cover subject, art style, lighting, color palette, camera angle and composition. Keep it under 80 words."
  3. Name the generator you'll use ("format it for Nano Banana", "for ChatGPT Images", "for Seedream 5.0" or "for Midjourney V8, with parameters") so the wording fits.
  4. Run the prompt, compare the result with the original, and ask the chatbot to adjust whatever is off.

The catch: a plain "describe this image" request gives you a paragraph written for people, not a prompt. Asking for prompt structure makes a big difference. If you're choosing between paid chatbot plans for this kind of work, our ChatGPT, Claude and Gemini plan calculator compares prices and limits.

How to get a prompt from an image for Nano Banana, ChatGPT Images or Seedream

None of today's popular generators need a special prompt format to accept text, so any method on this page works for them. What changes is the style of prompt each one likes:

  • Nano Banana (Google Gemini): Google's image models are now branded Nano Banana, with Nano Banana 2.1 and Nano Banana Pro as the current versions. They read full sentences well, so ask Gemini for a natural-language prompt that describes the scene, rather than a list of comma-separated tags. Since Gemini can see your image too, you can extract and generate in the same chat.
  • ChatGPT Images: OpenAI retired DALL·E from ChatGPT and replaced it with ChatGPT Images, which now runs on Images 2.5. Upload the reference, ask for a prompt, then say "now generate it" in the same conversation.
  • Seedream (ByteDance): the current release is Seedream 5.0 Pro, available in Dreamina and other platforms. It is strong at text and layouts, so ask your extraction tool to quote any visible words in the image exactly.
  • FLUX (Black Forest Labs): FLUX 3 Image launched in October 2026 and, like the others, takes descriptive plain-English prompts.

The simple rule: a good text prompt is portable. Extract it once, then paste it into any of these and adjust a few words for each model's taste.

Does Midjourney /describe still work for image to prompt?

Yes. As of October 2026, Midjourney's Describe help page says the tool was updated to give longer, more detailed prompts for its V8.1 and V8.2 models. On the website, drag an image onto the Imagine bar and drop it on "Drop image to describe", or right-click an image and choose "Describe Image". In Discord, type /describe and upload an image or paste a link.

Either way you get four prompt suggestions, and you can run one or all of them. Midjourney is upfront that the suggestions "won't precisely copy your image" and that running Describe again gives a different set. You need a Midjourney subscription to use it.

What is the best image to prompt app for Android?

If you want to do this from your phone, a dedicated app saves you writing the instructions yourself. One option is Reprompt, a free Android app that pulls an AI prompt from a photo. Its Google Play listing names Midjourney, DALL·E and Stable Diffusion. The output is a plain text prompt, so you can also paste it into other generators yourself. It also has a library of 300+ prompts you can browse and copy.

Reprompt is an independent tool. It is not affiliated with Midjourney, OpenAI or Google. There's also a web version at reprompt.org if you're on a computer — the website tools also support Flux.

How to get a prompt from an image on your phone with Reprompt

  1. Install Reprompt from Google Play.
  2. Upload a photo or an AI-generated image.
  3. Let the app analyze it. It looks at the subject, style, lighting, colors, camera angle and composition.
  4. Copy the prompt, or rewrite it in the app to make variations, then paste it into your image generator.

Like every reverse image prompt tool, it gives you a strong starting point rather than the original prompt. Expect to tweak a word or two.

Which image to prompt method should you use?

  • You have the original Stable Diffusion or ComfyUI file: read the metadata. It's the only exact answer.
  • You already pay for ChatGPT or Gemini: upload the image and ask for a structured prompt.
  • You generate in Nano Banana or ChatGPT Images: extract the prompt in Gemini or ChatGPT and generate in the same chat.
  • You generate in Midjourney: use Describe, since its suggestions are written in the style Midjourney expects.
  • You're on Android or want something free and quick: use an image to prompt app like Reprompt.

How to make an extracted prompt work better

Treat the result as a draft. Keep the subject and style words, cut vague filler like "beautiful" or "stunning", and add the details that matter to you, such as aspect ratio or a specific lens. Run it a few times: image models are random, so one bad result doesn't mean the prompt is wrong. And be thoughtful about whose work you are recreating. Copying a style for practice is different from passing off a near-copy of someone else's art.

Want more AI explainers like this? See how Anthropic is handling visuals in Claude Dashboards and Motion, or browse the Era of Minds blog.

Key facts

  • The exact prompt can only be recovered from an image's metadata, which screenshots and social uploads usually strip.
  • ChatGPT and Gemini can turn an uploaded image into a text-to-image prompt if you ask for prompt structure.
  • Current popular image generators include Google Nano Banana 2.1 and Nano Banana Pro, ChatGPT Images 2.5, ByteDance Seedream 5.0 Pro, FLUX 3 Image and Midjourney V8.2 (checked Oct 13, 2026).
  • OpenAI retired the DALL·E GPT from ChatGPT; ChatGPT Images replaced it.
  • A text prompt extracted from an image can be pasted into any of these generators.
  • Midjourney Describe returns four prompt suggestions per image and was updated for V8.1 and V8.2 (per Midjourney's docs, checked Oct 13, 2026).
  • Reprompt is a free Android app that pulls an AI prompt from a photo; its Play listing names Midjourney, DALL·E and Stable Diffusion, and its plain text output can be pasted elsewhere. It has a library of 300+ prompts. The reprompt.org website tools also support Flux.
  • Reprompt is independent and not affiliated with Midjourney, OpenAI or Google.

FAQ

How do I get a prompt from an image?

Check the original file's metadata first, since Stable Diffusion and ComfyUI images often store the prompt. If there's none, upload the image to ChatGPT or Gemini and ask for a detailed text-to-image prompt, use Midjourney's Describe tool, or use an image to prompt app like Reprompt.

How do I get a prompt from an image for Nano Banana?

Upload the image to Gemini and ask for a detailed natural-language prompt that recreates it, then ask Gemini to generate with that prompt. Any other extraction method works too, since Nano Banana accepts plain text prompts.

Can I use an extracted prompt in ChatGPT Images or Seedream?

Yes. An extracted prompt is plain text, so you can paste it into ChatGPT Images, Seedream, FLUX, Midjourney or Nano Banana and tweak the wording for each model.

Is DALL·E still available in ChatGPT?

No. OpenAI retired the DALL·E GPT from ChatGPT and points users to ChatGPT Images instead.

Can I find the exact prompt used for an AI image?

Only if it's saved in the file's metadata. Otherwise any image to prompt tool writes a new prompt that recreates something similar, not the original wording.

How do I get a prompt from an image in ChatGPT?

Upload the image and ask: write a detailed text-to-image prompt that recreates this image, covering subject, style, lighting, colors, camera angle and composition. Name the generator you plan to use so the format fits.

How do I get a prompt from an image in Gemini?

The same way: upload the image to Gemini and ask for a structured text-to-image prompt rather than a description, then refine it after testing.

Is Midjourney /describe still available?

Yes. As of October 2026, Midjourney's docs list Describe on the website (drag an image to the Imagine bar or right-click Describe Image) and as /describe in Discord. It returns four prompt suggestions.

Is there a free image to prompt app for Android?

Yes. Reprompt is a free Android app on Google Play that extracts AI prompts from photos; its listing names Midjourney, DALL·E and Stable Diffusion, and the plain text output can be pasted into other generators. It includes a library of 300+ prompts. The reprompt.org website tools also support Flux. It is independent and not affiliated with Midjourney, OpenAI or Google.

What is a reverse image prompt?

A reverse image prompt is a text prompt written by analyzing an existing image, so you can generate similar images. It works backwards from picture to prompt instead of prompt to picture.

Sources

Keep reading