What PNG Info is
Most local Stable Diffusion interfaces write the settings that made an image into the image file itself. The name comes from the PNG Info tab in Automatic1111, which reads that text back so you can send the settings to txt2img and generate again. The data is plain text, not encrypted, so any tool that knows where to look can read it, including this one.
A typical record holds the positive and negative prompt, steps, sampler and scheduler, CFG scale, seed, width and height, the checkpoint name and its short hash, the VAE, any LoRAs with their weights and hashes, and extras such as hires fix and the interface version. ComfyUI goes further and stores the whole node graph.
Where each interface stores it
- Automatic1111 and Forge
- A PNG text chunk named parameters. JPEG and WebP exports put the same text in the EXIF UserComment field.
- ComfyUI (including Flux workflows)
- Two PNG text chunks: prompt, the graph that actually ran, and workflow, the full editor layout with every node.
- Fooocus
- A parameters chunk in either its own JSON format or the Automatic1111 text format, when metadata saving is turned on.
- InvokeAI
- invokeai_metadata, plus the workflow or graph that produced the image. Older versions used sd-metadata or Dream.
- NovelAI
- The prompt in a Description chunk and the settings as JSON in a Comment chunk.
PNG text chunks can be stored plain (tEXt), compressed (zTXt) or as UTF-8 (iTXt, used when a prompt has characters outside Latin-1). This viewer reads all three, plus EXIF UserComment in JPEG and WebP in both UTF-16 byte orders. It reads text metadata only; it does not decode anything hidden in the pixels.
Why JPEG, WebP and social uploads often have none
The tEXt, zTXt and iTXt chunks only exist in PNG. When you convert a PNG to JPEG or WebP in an image editor, those chunks usually have nowhere to go and are left behind. Automatic1111 and Forge avoid this for their own JPEG and WebP exports by writing EXIF UserComment instead, but a second save in another program can still drop it.
Many social networks and messaging apps resize and re-encode uploads, which throws away text chunks and most EXIF. Screenshots never carry generation data. On the other hand, file hosts, cloud drives, email attachments and sites built for sharing AI images often keep the original file byte for byte, so the prompt survives anywhere the original file is shared.
What your PNG Info gives away
Sharing an image with its metadata shares the recipe. The prompt can reveal ideas, client names, product details or reference artists you would rather keep to yourself. The seed plus the sampler, steps, CFG and size let someone reproduce a close version of the image. Model names, model hashes and LoRA names point to exactly which checkpoint and add-ons you used, including private LoRAs trained on a product, a style or a person.
ComfyUI workflows can say even more: every node, custom node packs, the names of input images you loaded and sometimes folder names from your machine. If you sell images, deliver client work or post in places where originals can be downloaded, check what is in the file first.
How to remove it
Some interfaces let you turn metadata off before you generate: Automatic1111 and Forge have a setting to stop saving generation parameters to PNG, and ComfyUI can be started with a flag that disables it. For images you already have, use the Stable Diffusion metadata remover. It deletes the parameters, prompt and workflow chunks and the EXIF UserComment field in your browser, without re-encoding, so the pixels are untouched. Then drop the cleaned file back into this viewer to confirm it reads as empty.
Questions
Is my image uploaded anywhere?
No. The file is read by JavaScript in your browser tab and never sent to a server. You can open your browser's network tab while you drop a file and see that no upload request is made.
Why does it say no metadata was found when the image came from Stable Diffusion?
The text that holds the prompt and settings is easy to lose. Converting a PNG to JPEG or WebP in an editor, taking a screenshot, or uploading to a site that re-encodes images all drop it. Some generators can also be set to save no metadata at all. Try the original file from your output folder.
Can I load a ComfyUI workflow from an image?
Yes. When a ComfyUI PNG still has its workflow chunk, this viewer shows it and lets you download it as workflow.json. You can drag that file, or the original PNG, onto the ComfyUI canvas to rebuild the graph. Custom nodes used in the workflow still need to be installed.
Can someone recreate my image from its PNG Info?
Often close to it. With the same prompt, seed, sampler, steps, CFG scale, size and model, another person can usually generate a near match, especially on the same software version. Results can drift with different hardware, drivers or extensions, but the settings give away most of the recipe.
How do I remove Stable Diffusion metadata from an image?
Use the StripShot Stable Diffusion metadata remover. It deletes the parameters, prompt and workflow chunks and the EXIF UserComment field in your browser without re-encoding the image, so the pixels stay exactly the same.