By Tool

Recover the prompt from a ComfyUI PNG

ComfyUI embeds the full workflow and prompt inside every output PNG as text chunks — as long as the file is the original, unconverted PNG, the prompt is fully recoverable.

Try it now: AI Metadata Inspector

Free, no login — upload an image and read its embedded metadata in your browser.

What to know

When ComfyUI saves a PNG, it writes the entire workflow graph and the flattened execution prompt into the file as PNG text chunks (a "workflow" key and a "prompt" key). That means the prompt, seed, model, sampler, and node graph all travel inside the image itself — nothing is stored elsewhere.

The catch is fragility: those chunks only survive while the file stays a PNG. Converting to JPEG or WebP, or re-saving through an editor that drops custom chunks, deletes the workflow. If you still have the original PNG, recovery is reliable; if you only have a converted copy, the embedded data is gone.

How to recover it

  1. 1

    Confirm you have the original PNG

    Recovery works from the file ComfyUI saved. A JPEG/WebP copy, a screenshot, or a social-media download will have lost the chunks.

  2. 2

    Drag the PNG back into ComfyUI

    Dropping the original PNG onto the ComfyUI canvas reconstructs the full workflow, including the prompt and every node parameter.

  3. 3

    Or read the text chunks with a metadata viewer

    Upload the PNG to the AI Metadata Inspector to read the embedded "prompt" and "workflow" chunks without opening ComfyUI.

  4. 4

    Save the recovered prompt somewhere durable

    Copy the recovered prompt and parameters into your library so a future format conversion cannot lose them again.

Importing ComfyUI output into a catalog that extracts the workflow on arrival means the prompt is preserved centrally — so a later JPEG export can never be the only copy.

Frequently asked questions

Frequently Asked Questions

Quick answers to the most common questions.

Related recovery guides

Never lose a prompt again

Numonic extracts the prompt, model, seed, and settings from every AI image on import and keeps them searchable — so recovery stops being a rescue mission and the prompt is always attached to the work.