BWhat changed
Measured FID variability on a two-axis panel of training and generation seeds for hundreds of SiT image models and introduced per-cell guidance tuning and multi-training-seed reporting recommendations.
Scientific scope
Measured FID variability on a two-axis panel of training and generation seeds for hundreds of SiT image models and introduced per-cell guidance tuning and multi-training-seed reporting recommendations. Empirical findings concern the studied ImageNet256 SiT settings. The reported approximately1.3percent coefficient of variation is not a universal significance threshold or noise floor. Guidance optimization adds evaluation cost; this does not invent FID.
Evidence
The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Empirical findings concern the studied ImageNet256 SiT settings. The reported approximately1.3percent coefficient of variation is not a universal significance threshold or noise floor. Guidance optimization adds evaluation cost; this does not invent FID.
Contributors
- Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.
- The FID Lottery: Quantifying Hidden Randomness in Generative-Model Evaluation
- Kyutai scientific publications index
Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.
Full scientific record · Browse contribution grades · Request a correction