Repository navigation
Supported Formats
soifaxa edited this page Oct 24, 2026
·
1 revision
The app reads embedded metadata only. If a host or exporter stripped the prompt, nothing can be recovered.
| Field | Location |
|---|---|
| Parameters block | PNG tEXt / iTXt / zTXt chunk named parameters
|
Parsed into positive prompt, negative prompt (Negative prompt:), and a parameter line (Steps:, Sampler:, CFG scale:, Seed:, Size:, …).
| Field | Location |
|---|---|
| API prompt graph | PNG text key prompt (JSON) |
| UI workflow | PNG text key workflow (JSON) |
The extractor walks nodes and collects string inputs such as:
-
text(classicCLIPTextEncode) -
prompt/positive/positive_prompt system_prompt- negative fields (
negative,negative_prompt, …)
Generation size is taken from latent / scale nodes when present, otherwise from the image pixel dimensions.
| Field | Location |
|---|---|
| Comment JSON / prompt + JSON | PNG Comment
|
| Field | Location |
|---|---|
| Metadata JSON | PNG invokeai, invokeai_metadata, or sd-metadata
|
| Field | Typical keys |
|---|---|
| Free-form or A1111-style text |
UserComment, ImageDescription, Caption |
Even when the workflow has no width/height, Size is filled from the file’s pixel dimensions and shown under Parameters.