isGenAI.com · Free, reads what the generator stored
Image to prompt: AI image prompt extractor
This extractor reveals the exact words a generator stored inside an AI picture, word for word. Upload the original file to get the instructions, the negative terms, the seed and settings, and the model name, ready to copy and pick up where the creator left off. When a picture stores no prompt, it writes a new image-to-prompt description of what is visible, clearly labelled as an approximation rather than the original.
- Free · No sign-up
- AUTOMATIC1111 · ComfyUI · InvokeAI
- Reads the exact saved prompt, no AI model call
- PNG, JPEG, WebP · up to 20 MB
- Private by default
Free. No account. Up to 20 MiB. Private by default.
See supported filesLanguages: English
Image prompt extractor
Read your image.
Find the saved prompt.
Drop an original image here or choose a file. Reading starts the moment you pick it.
PNG, JPEG or WebP · Up to 20 MBNo image handy? Try an example.
Same picture, different saved details. Read each file to see what it contains.
We drew this picture in code and added sample generation details to one copy. See the files and method.
Quick answer
How do I extract the prompt from an AI image?
isGenAI (isgenai.com) is a free image prompt extractor that reads the prompt a generator saved inside the file, word for word. Upload the original PNG, JPEG or WebP: it returns the saved prompt, negative prompt, seed, sampler, steps and model from AUTOMATIC1111, ComfyUI (including Flux), InvokeAI and NovelAI records. If nothing is saved, it says so instead of guessing.
- AUTOMATIC1111-style parameters are read from PNG, JPEG and WebP; ComfyUI and InvokeAI records from PNG.
- Screenshots and re-saved copies usually lose the prompt, so use the creator’s original export.
- No reconstruction model is used: the reader returns only saved text.
- Free, up to 20 MiB, no account, and available through the API.
Last updated by isGenAI. Methodology and limits
Use the original.
Choose the file your generator exported, in its original format. That is where the details live.
Get the real instructions.
isGenAI reads the wording stored inside the file, exactly as the generator wrote it. No AI model, no guessing.
Make it your starting point.
Copy the instructions in one click. Open the extra details for the negative terms and model name.
Read the seed and generation settings.
Generation settings are the numbers a generator stores next to its instructions. The reader shows the seed, steps, sampler, scheduler, CFG scale, dimensions and denoising strength stored in the file. The fields you see are the fields the file holds, read exactly as stored, never padded with defaults or invented values.
PNG files: start with the original export
A PNG carries both the visible picture and separate text records, and that is where the instructions live. Download the original from the generator rather than taking a screenshot, and keep a copy before editing, resizing or converting it, because an export keeps the pixels while discarding the record.
How do I pull generation details from a file?
Upload the file, copy the returned instructions and review the negative terms, model name and generation settings alongside them. The fields describe exactly what this copy stores. A stored seed is meaningful with the same model, software version and workflow; change any of those and the output changes with it. For ComfyUI, open the workflow viewer when you want the stored graph rather than one extracted string.
What if the file has nothing readable?
Grab the creator’s original download and compare it with the shared copy. A missing record means the metadata was removed, the workflow stores it differently or there is no single answer to return, and the reader tells you so instead of inventing instructions. Use the metadata comparison to see precisely which records survived an export. Stored records are editable, so keep the file and its source together when documenting your findings.
Extract a Stable Diffusion prompt from a PNG (A1111 PNG Info online)
AUTOMATIC1111’s PNG Info tab shows the generation parameters the web UI saves in a PNG text record. This page reads the same record online, with nothing to install: upload the original PNG and copy the instructions, negative terms, seed, steps, sampler, CFG scale and dimensions. Files from other Stable Diffusion tools work whenever they save the same parameters record, and compatible records in JPEG and WebP are read too. This is a generation-settings reader; for every PNG metadata field, use the EXIF viewer.
AUTOMATIC1111’s PNG info documentation describes the stored text. For camera dates and GPS instead, use the photo EXIF data viewer.
Read a ComfyUI workflow from a PNG
ComfyUI saves its workflow inside the PNG it exports, including Flux workflows, and isGenAI follows the steps connected to the output to find its text node. Complex workflows combine several text or picture inputs, so when there is no single answer to copy the reader says so and hands you the full graph instead of choosing one for you.
Need the nodes rather than the wording? Inspect and download the full ComfyUI workflow as editable or API JSON.
Which files can I read?
You can read original exports that still carry a supported record: original PNG files with AUTOMATIC1111-style parameters, ComfyUI, InvokeAI or NovelAI records, plus compatible parameter records in JPEG and WebP.
InvokeAI
Use an original PNG carrying its creation settings. Instructions, negative terms and model details are read directly. When main and style texts differ, the reader points you to the full workflow in InvokeAI rather than returning half of it.
NovelAI
Use the original PNG you downloaded from NovelAI. The reader checks that its Software record says NovelAI, then reads the JSON Comment record: prompt, undesired content (negative prompt), seed, steps, sampler, scale and size, including V4 and V4.5 base captions. Images with several character captions hold more than one prompt, so the reader says no single prompt can be recovered. The alpha-channel stealth copy is not decoded. More on NovelAI metadata.
Why was nothing found?
Usually the copy you uploaded is not the original export. Screenshots and social previews carry only the picture. An export can also drop its details, or use a format outside this reader. The original file from its creator is the one that holds the record.
An empty result says nothing about whether a person or AI made the picture. For the full evidence, run the AI detector.
Try a sample before your own file
A sample is the quickest way to see both results. Download our sample with generation text and the same picture without one, then upload each above. One returns the stored text; the other returns nothing to copy.
These are controlled examples: we drew a 640 × 400 picture in code and added one AUTOMATIC1111-style parameters record (Steps: 20, Sampler: Euler, CFG scale: 7, Seed: 42) to the 13,352-byte copy. Read how the example works.
Can I use it in my own workflow?
Yes. Read the same records through the API with no reconstruction call and no API key. See the API example for a runnable Node.js script.
Prompt extractor questions
Is this an image-to-prompt generator?
No, and that is the point. It reads the exact instructions the generator stored inside the file, word for word, with no guessing from the pixels. If the file holds none, the result says so plainly instead of inventing any. Stored text is editable, so the report tells you precisely what this copy of the file contains.
How do I extract a prompt from a Stable Diffusion PNG?
Upload the original PNG from AUTOMATIC1111 or any tool that saves the same parameters record. The extractor returns the stored instructions and, when present, the negative text, seed, steps, sampler, CFG scale and dimensions, instantly. Use the original download: screenshots and re-encoded copies lose this record.
Can it read NovelAI images?
Yes, from the original PNG download. It reads NovelAI’s JSON Comment record and shows the prompt, the undesired content (negative prompt), seed, steps, sampler, scale and size. Screenshots and JPEG copies lose this record, and the alpha-channel stealth copy is not decoded.
Can it read ComfyUI and InvokeAI images?
Yes. For ComfyUI it follows the stored graph to the text node connected to the output, and for workflows that combine several inputs it points you to the full graph instead of picking one. InvokeAI instructions, negative text and model details are read directly. Open the ComfyUI workflow viewer for the complete node graph.
How do I remove a saved prompt before sharing a picture?
Use the metadata remover. It builds a new PNG in your browser with the text chunks gone, so generation instructions and ComfyUI workflows stay out of the download, along with EXIF and XMP records. Keep your original if you want those details later.