ComfyUI Metadata Viewer Extract the Workflow from a PNG or MP4

What a ComfyUI metadata viewer actually reads

ComfyUI embeds the entire generation recipe inside the file it saves. A PNG stores it in the text chunks; an MP4 stores it in the container metadata when the saving node writes it. This ComfyUI workflow extractor parses those fields and shows you:

• workflow — the editable node graph, exactly as it appeared on the canvas.
• prompt — the resolved API-format graph with every value already filled in.
• Positive and negative prompt text as it was sent to the encoder.
• Seed, steps, CFG, sampler and scheduler from the sampler node.
• Checkpoint, UNet, VAE and LoRA names with their strengths.
• Resolution and frame count for video outputs.

How to extract a ComfyUI workflow from an image or video

1. Open the extractor — the drop zone loads in any browser.
2. Drag the file in — the original PNG or MP4 that ComfyUI wrote.
3. Read the breakdown — prompt, seed and sampler up top, full JSON below.
4. Rebuild it — copy the workflow JSON and drop it onto the ComfyUI canvas to load the graph.

Why an image sometimes has no workflow

The metadata is fragile, and almost every step that touches the file strips it. If the extractor finds nothing, one of these usually happened:

• It was re-encoded — converting PNG to JPG or WebP drops the text chunks entirely.
• It went through a social platform — Discord, Reddit, X, WhatsApp and Instagram all re-compress uploads.
• It is a screenshot — a screenshot is a new file with none of the original data.
• The saving node did not embed it — some custom save nodes and video encoders write no metadata.
• It was cropped or edited — most editors rewrite the file without the original chunks.

The rule of thumb: only the untouched file straight out of the ComfyUI output folder is guaranteed to still carry the workflow.

Who uses this

• Reproducing a result — you saved a good generation weeks ago and want the exact settings back.
• Learning from a shared image — someone posted the original PNG and you want the graph behind it.
• Auditing a render — confirming which model, LoRA and seed a batch actually used.
• Recovering a lost workflow — the .json is gone but an output file survived.
• Debugging a video pipeline — checking frame count, fps and sampler on an MP4 output.