Read the prompt and settings from an AI image

What the viewer reads

Many Stable Diffusion tools write a record of how an image was made into the image file itself, next to the pixels. The viewer looks for the formats below and shows what it finds as plain text. It does not guess: if a file holds no generation data, it says so.

AUTOMATIC1111 and Forge

The AUTOMATIC1111 Stable Diffusion web UI, and Forge, which is built on it, save a block of text that the code calls the infotext. In a PNG it is stored in a text chunk named parameters. In a JPEG or WebP it is stored in the EXIF UserComment field. The first lines are the prompt. A line starting with Negative prompt: holds the negative prompt. The last line lists the settings, for example Steps: 20, Sampler: Euler a, CFG scale: 7, Seed: 965400086, Size: 512x512, Model hash: 45dee52b, followed by the model name and any extras such as LoRA hashes. The web UI option Write infotext to metadata of the generated image is on by default, as are the options that add the model name and hash.

The viewer splits the infotext the same way the web UI does and shows each setting on its own line. It also keeps the original text, so Copy full infotext gives you exactly what the file contains.

ComfyUI

The standard Save Image node in ComfyUI writes two PNG text chunks, both JSON. prompt is the graph that was run, with every node and its inputs. workflow is the full editor graph, including node positions, links and groups. The viewer finds the text nodes wired to a sampler's positive and negative inputs and shows them as the prompt and negative prompt, then lists the seed, steps, CFG, sampler, scheduler, denoise and the model, LoRA and VAE file names. Download workflow.json saves the workflow chunk exactly as it is stored, after checking that it is valid JSON.

NovelAI

NovelAI images carry PNG text chunks such as Title, Description, Software, Source and Comment. The description is the prompt, and the comment is JSON with the settings, including the negative prompt (uc), steps, scale, seed and sampler. The viewer reads these when Software says NovelAI. NovelAI also says it stores the same information in the image's alpha channel. That copy is part of the pixels, and the viewer does not read it.

Other text the file contains is listed under All metadata with a short preview. Compressed text chunks are detected but not unpacked, so their content is not shown.

How to read a prompt from an image, step by step

1. Save the original image file to your device. Use the download option of the site or app where you found it, rather than a screenshot, so the file stays unchanged.
2. Choose the file in the viewer at the top of this page, or drop it onto the box. PNG, JPEG and WebP files work, up to 20 at a time.
3. Read the result. The prompt comes first, then the negative prompt, then the settings: seed, steps, sampler, CFG scale, size and model, for example.
4. Press Copy prompt or Copy negative prompt to put that text on your clipboard. For AUTOMATIC1111 and Forge images, Copy full infotext copies the whole original block, settings included.
5. For a ComfyUI image, press Download workflow.json to save the node graph as a separate file.
6. If nothing shows up, read the next section. The generation data was probably removed before the file reached you.

Everything runs in your browser. The image is not sent to SaveMingo, and closing the tab discards it. Prompts and settings are shown as plain text, never as web page code.

Why some images show nothing

The generation data lives in metadata, not in the pixels, so anything that rewrites the file can drop it. The usual reasons are:

• The image was uploaded somewhere that removes metadata. Many image hosts, social apps and chat apps re-encode images or remove metadata when you upload them. The copy you download from them may no longer hold the prompt, even if the original did.
• It was converted to another format. The parameters, prompt and workflow chunks are a PNG feature. When a PNG is turned into a JPEG or WebP, the new file keeps the prompt only if the converting app copies it into another field, such as EXIF, which many apps do not.
• It is a screenshot. A screenshot is a new image of what was on the screen. It contains none of the original file's metadata.
• Metadata was turned off when it was made. In AUTOMATIC1111 and Forge, the creator can switch off Write infotext to metadata of the generated image. ComfyUI has a --disable-metadata start-up option that stops it saving the prompt and workflow in files.
• It was cleaned on purpose. Some creators remove the data before they share, for example with a metadata remover.
• The app that made it did not save a prompt. Not every generator writes its prompt into the file, and some use formats this viewer does not read. The viewer only reads the formats listed above.

If you need the prompt, the most reliable route is to ask the creator, or to look for the original file where they first posted it.

Using the prompt in AUTOMATIC1111, Forge and ComfyUI

AUTOMATIC1111 and Forge

Both web UIs have a PNG Info tab. Drop the original image onto its Source box to see the infotext, then use the Send to txt2img, Send to img2img, Send to inpaint or Send to extras buttons to carry the prompt and settings over. That needs the image file. If you only have the text, copy it with Copy full infotext here, paste it into the txt2img prompt box and press the ↙️ button in the tools next to Generate. Its tooltip reads Read generation parameters from prompt or last generation if prompt is empty into user interface. The web UI then splits the text into the prompt, negative prompt and settings fields. Dropping the image file straight onto the prompt box also fills it with the infotext.

You can paste just the prompt or negative prompt too, if you only want the wording and not the settings.

ComfyUI

Drag the original PNG onto the ComfyUI window to load the workflow that made it. The official ComfyUI examples rely on this: they say their images can be loaded, or dragged onto the window, to get the full workflow. If you downloaded workflow.json here, drag that file onto the window, or open it with Open in the File menu. The workflow refers to model and custom node names, so to run it you need the same files installed. The ComfyUI examples list the model files each workflow needs.

Privacy: your own images can leak your prompt

Everything this viewer can read, anyone else can read too. If you share an original PNG from AUTOMATIC1111, Forge or ComfyUI, the person who gets it can see your prompt, seed and model, and with ComfyUI your whole workflow. Prompts and workflows can also contain names, notes and the file names of models and images on your computer.

To share an image without that data, use the AI metadata remover. It makes a copy without the prompt, settings and workflow and leaves the pixels untouched. On this page, the same clean copy is under All metadata and a clean copy without it for each file. For camera photos and location data, see the image metadata remover.

Respect the creator

A prompt or a workflow can be someone's creative work, built over many attempts. If you reuse one, credit the person who made it, and follow any terms they posted with the image. Our copyright page explains how we handle rights questions.

Questions

Is this the same as PNG Info? It reads the same infotext that the PNG Info tab in AUTOMATIC1111 and Forge reads, without installing anything. It also reads ComfyUI and NovelAI data.

Can I get the ComfyUI workflow from an image? Yes, if the PNG still contains its workflow chunk. Press Download workflow.json, or drag the PNG onto ComfyUI.

Does it work with JPEG and WebP? Yes, for AUTOMATIC1111 and Forge images, which keep the infotext in EXIF. ComfyUI and NovelAI data is read from PNG files.

Why is the prompt different from what I see in another viewer? The viewer shows the text exactly as stored in the file. Other tools may tidy it up or read a different field.

Is my image uploaded? No. The browser reads the file on your device. See our privacy policy.

Sources

• AUTOMATIC1111 Stable Diffusion web UI: modules/images.py (PNG parameters, EXIF UserComment, reading an image dropped on the prompt box), javascript/dragdrop.js, modules/infotext_utils.py (infotext parsing, send-to buttons), modules/shared_options.py (defaults), modules/ui.py (PNG Info tab) and modules/ui_toprow.py (the ↙️ button)
• Forge: modules/ui.py and modules/ui_toprow.py
• ComfyUI: nodes.py (Save Image writes prompt and workflow), comfy/cli_args.py (--disable-metadata), the ComfyUI examples and the ComfyUI frontend (loading PNG and JSON files, File menu Open)
• NovelAI: novelai-image-metadata (alpha channel) and the naipng reader (PNG text chunks)