Free, private, like PNG Info

AI Image Prompt Extractor

Read the prompt and settings an image generator saved in the file, and copy them or load the workflow again.

Last updated: November 2026

Prompt

And negative prompt

Settings

Model, seed, steps

ComfyUI

Workflow download

Private

No uploads

Drop your image here

PNG, JPG and WebP from Stable Diffusion, ComfyUI, NovelAI and more.

Your file never leaves this device

How to get the prompt from an AI image

01

Add the original image

Drop or choose the PNG, JPG or WebP as the generator saved it. Copies from social media or chat apps have usually lost the prompt.

02

See the prompt and settings

The prompt, negative prompt, model, seed, steps, sampler and other settings appear as the generator saved them.

03

Copy or download

Copy the prompt alone, copy everything in the Stable Diffusion WebUI layout, or download a ComfyUI workflow as JSON.

04

Reuse it

Paste the settings back into your generator, or load the workflow into ComfyUI, to recreate or vary the image.

Where generators save the prompt

Image generators you run yourself save their settings inside the image, so you can make it again later. Each one does it its own way; the extractor reads all of these, from each tool's own source code:

  • Stable Diffusion WebUI and Forge

    A PNG text field named "parameters": the prompt, "Negative prompt:" and a line of settings. JPG and WebP files keep the same text in their EXIF comment.

  • ComfyUI

    Two PNG text fields: "prompt", the graph that was run, and "workflow", the full editor layout. WebP files keep them in their EXIF fields.

  • InvokeAI

    A PNG text field named "invokeai_metadata" with the prompt, model and settings as JSON, and in newer versions the workflow too.

  • NovelAI

    PNG text fields with the prompt and a JSON comment of settings, and the same data hidden in the transparency channel.

  • Fooocus and SwarmUI

    Their settings as JSON in the PNG "parameters" field.

  • Other tools

    Some tools fill in the IPTC AI prompt field in the photo metadata, which the extractor reads as well.

Why some AI images have no prompt

ChatGPT and Google Gemini don't save the prompt in their images. We checked images straight from both in November 2026: they carry Content Credentials saying they were made with AI, but no prompt. The extractor tells you when an image says it was made with AI without saving its prompt.

Images that had a prompt lose it easily: a screenshot, a converter, an editor that saves a new file, or a social network that strips metadata on upload. If you only have such a copy, the prompt is gone from the file.

This tool never invents a prompt. Image to prompt tools ask an AI to describe a picture, which gives a new prompt that may produce something similar, not the one that was used.

Reuse the settings

Stable Diffusion WebUI and Forge: Copy prompt and settings gives the text in the WebUI's own layout, the prompt, then "Negative prompt:", then one line of settings. Paste it into the prompt box and use the button that reads generation parameters from the prompt to fill in the model, seed, steps and the rest.

ComfyUI: Download ComfyUI workflow saves the full node graph as JSON. Drop it onto ComfyUI's canvas to load it.

NovelAI: its prompt and settings are also hidden in the image's transparency channel, so the extractor can read them even from a PNG whose text fields were removed, as long as the pixels weren't re-saved.

Common mistakes

Using a copy from social media

Most platforms remove metadata on upload. Ask for the original file, or check the page it was posted on.

Converting the image first

Converting or re-saving usually drops the prompt. Check the file before you change it.

Expecting a prompt from ChatGPT or Gemini

They don't save it in the image. Their images carry Content Credentials instead, which the Content Credentials Checker shows.

Forgetting the model and seed

The same prompt gives a different image with another model, seed or sampler. Copy prompt and settings keeps them together.

Your images are read in your browser and never uploaded. Prompts are often other people's work: credit them when you reuse one. To check what else an image says about AI, use the AI Image Checker.

Why use ImgTweak?

✓No upload, no account
✓Seven generators' formats
✓ComfyUI workflow download
✓Copy in the WebUI layout
✓Many images at once
✓Works on phones

Frequently asked questions

How do I get the prompt from an AI image?

Add the image above. If the generator saved its prompt in the file, as Stable Diffusion WebUI, ComfyUI, InvokeAI, NovelAI, Fooocus and SwarmUI do, the prompt and settings appear with a Copy button.

Why is no prompt found?

Either the generator didn't save one, or it was removed. ChatGPT, Gemini and most online generators don't save the prompt in their images, and screenshots, re-saving and most social media remove it from images that had one.

Is this the same as PNG Info in Stable Diffusion?

It reads the same "parameters" text that PNG Info shows, and also ComfyUI, InvokeAI, NovelAI, Fooocus and SwarmUI metadata, the comment in JPG and WebP files, and IPTC prompt fields, without installing anything.

Can it guess the prompt of any image?

No. It reads only what the generator saved. Tools that write a new prompt from what they see in a picture, often called image to prompt, are a different thing, and their prompt won't be the one actually used.

How do I load a ComfyUI workflow from an image?

Select Download ComfyUI workflow, then drop the JSON file onto ComfyUI's canvas. ComfyUI saves the whole node graph in its PNG and WebP images, so the workflow comes back exactly as it was.

Can I paste the settings back into Stable Diffusion WebUI?

Yes. Copy prompt and settings gives the text in the WebUI's own layout: the prompt, then "Negative prompt:", then the settings. Paste it into the prompt box and use the button that reads generation parameters from the prompt.

What is the NovelAI hidden data?

NovelAI also stores its prompt and settings in the lowest bit of each pixel's transparency, so they survive when text fields are stripped. The extractor reads it from PNG files that keep their exact pixels.

Which files work?

PNG, JPG and WebP, which is where these generators save their settings. Other formats are read too, but rarely hold a prompt.

Are my images uploaded?

No. Everything is read in your browser, and nothing is sent anywhere.

Does it check whether an image is AI generated?

It shows the prompt where one was saved, which only image generators do. For a full check, including Content Credentials and AI labels, use the AI Image Checker.

Ready to get the prompt?

Add an AI image above to see what its generator saved.