The invokeai_metadata chunk
Modern InvokeAI (3.x/4.x/5.x) writes a tEXt chunk named invokeai_metadata containing JSON with fields like positive_prompt, negative_prompt, seed, steps, cfg_scale, scheduler, model (an object with key/name/base), loras, width/height and, for canvas/control workflows, control_layers. Older releases used sd-metadata or Dream. PNGMeta summarises the common fields and shows the full JSON pretty-printed.
Optional invokeai_graph / workflow
If "save workflow" is enabled InvokeAI also embeds invokeai_workflow (the node editor graph) and sometimes invokeai_graph. Download them here as JSON and import them via the InvokeAI workflow editor.
Reproducing an image
Copy the prompt and negative prompt, then set seed, steps, CFG and scheduler in InvokeAI. Use the model name/base shown in the summary to select the same checkpoint. LoRAs are listed with their weights in the raw JSON.
Frequently asked questions
How do I see the metadata of an InvokeAI image?
Upload the PNG to pngmeta.app. The generation summary shows prompt, negative prompt, seed, steps, CFG, scheduler and model; the raw invokeai_metadata chunk is listed below it.
Does InvokeAI metadata work in Automatic1111?
Not directly — the formats differ. Copy the prompt and settings from the summary into the A1111 prompt box manually.