png::meta tEXt · iTXt · zTXt

Extract AI Prompts & Workflow Data from Any PNG in 1 Click

Drop a PNG to see every prompt, setting and workflow hidden inside it. Prompts, seed, model and the full ComfyUI workflow, instantly visible. Works with Automatic1111, Forge, InvokeAI and JPEG/WebP EXIF too. Parsed in your browser, nothing to install.

🎥 [Demo: Upload a PNG to see AI prompts & workflow data extracted in seconds]

Drop a PNG here or click to browse · JPEG & WebP work too
ComfyUI · A1111 · Forge · InvokeAI · SwarmUI · Fooocus · any tEXt/iTXt/zTXt · JPEG/WebP EXIF

What is hidden inside an AI-generated PNG?

Tools like ComfyUI, Automatic1111 and Forge embed the full generation recipe in the PNG file itself, using text chunks defined by the PNG standard (tEXt, iTXt and zTXt). For ComfyUI that means two JSON payloads: prompt (the executed API graph) and workflow (the editable canvas graph). This viewer reads those chunks byte-for-byte, decompresses zTXt/iTXt with deflate, and shows you the prompts, seed, sampler, model and every node parameter. Read more on how PNG metadata extraction works.

Recover a lost prompt or workflow

If you still have the original PNG, the workflow is recoverable: extract it here, download the JSON, and drag it onto the ComfyUI canvas. If the file went through a screenshot, a social platform, or WhatsApp as a photo, the metadata is gone; re-encoding strips text chunks permanently.

Frequently asked questions

How do I extract AI prompts from a PNG?

Drop the PNG on pngmeta.app (or click "Upload PNG → Extract Metadata"). The tool reads the PNG's tEXt, iTXt and zTXt chunks directly in your browser and shows the positive/negative prompt, seed, sampler, model and, for ComfyUI images, the full workflow JSON. Nothing needs to be installed.

What metadata do AI-generated PNGs contain?

Most AI image tools embed the complete generation recipe as PNG text chunks. Automatic1111/Forge write a "parameters" chunk with prompt, negative prompt, steps, sampler, CFG scale, seed and model hash. ComfyUI writes two JSON chunks: "prompt" (the executed graph) and "workflow" (the editable canvas). InvokeAI, SwarmUI and Fooocus use similar keys. Standard fields like Software, Comment and creation time can also be present.

Can I reverse engineer the workflow from a PNG?

Yes, if the PNG still has its original metadata. For ComfyUI images the entire node graph is stored in the "workflow" chunk: extract it here, download the JSON and drag it onto the ComfyUI canvas (or drag the PNG itself into ComfyUI). For A1111/Forge images the "parameters" text contains every setting needed to reproduce the generation.

How to find hidden data in PNG files?

PNG files can hide data in text chunks (tEXt, iTXt, zTXt), EXIF/XMP blocks and non-standard ancillary chunks. Upload the file to pngmeta.app to list every text chunk with its key, size and decoded (and, for zTXt/iTXt, decompressed) content. If a PNG has no chunks at all it was probably screenshotted or re-encoded by a messaging app or social platform, which strips metadata.

Is there a free tool to get Stable Diffusion prompts from PNGs?

Yes. pngmeta.app is a free online PNG metadata extractor for Stable Diffusion, ComfyUI, Automatic1111, Forge, InvokeAI and SwarmUI images. Parsing happens client-side, there is no account or install, and it works on desktop and mobile.

Why does my PNG have no metadata?

Screenshots never contain generation metadata, and most messaging apps (WhatsApp, Telegram as photo, Discord mobile) re-encode images and strip text chunks. Always send AI images as files/documents to preserve the workflow.

Which tools are supported?

ComfyUI (prompt + workflow chunks), Automatic1111/Forge (parameters chunk), InvokeAI, SwarmUI, Fooocus, and any PNG with standard tEXt, iTXt or zTXt chunks.

Guides by tool

More questions? See the full PNG metadata FAQ or the developer chunk reference.