Generation Recipe Reproducibility

How to use Generation Recipe Reproducibility

Record what an image-generation delivery claims to reproduce. Compare declared recipe references and fingerprints without treating a matching seed or parameter label as a deterministic provider guarantee.

1. Record each generation’s actual model, version, seed, dimensions and explicit parameter strings under a stable recipe ID.
2. Enter output asset IDs and their independently obtained fingerprint references against those recipe IDs.
3. Investigate different fingerprints under the same complete declared recipe and fill missing metadata before exporting the reproducibility register.

Good to know

• Complete refers only to the displayed entered schema. Fingerprints are references, not hashes calculated or independently authenticated here.
• No generation is run. Matching recipes do not establish vendor determinism, equivalent images, permissions, model weights or a complete execution environment.
• Exact displayed CSV headers; at most100 data rows and8,000 UTF-16 characters per input. Plain decimals have at most9 fractional places; display rounds half-up to12 places. Unknown evidence stays unknown. CSV exports preserve summaries, settings, every calculated detail/finding, limitations and scoped input records; spreadsheet-like text is apostrophe-prefixed.
• Browser-local supplied-record review. No AI, network requests, provider connection, file upload, stored input or automatic delivery. Use neutral identifiers instead of personal information or secrets.