That Stare
Paired feature review
Review one unfamiliar dog using clear originals and two existing, independently generated Studio portraits. Lock the source evidence before looking at either output.
Photos stay on this device, in memory. No uploads, saved browser data, or generation calls. Export records document your observations; they do not approve artwork or launch.
Load the existing files
Use single-frame PNG, JPEG or WebP files at their original resolution. No screenshots or enlarged thumbnails.
Establish the source feature
Choose one specific, named feature. Inspect every original and mark its boundary with three or more taps. For an absent feature, outline the region you checked for absence.
Compare Output A and Output B
The output order is shuffled and filenames are hidden. Keep the feature judgment tied to the locked originals. Mark the corresponding output region, including the inspected region if a feature is absent.
Original reference only. Source labels are locked. Use “Restart source review” to revise them; this clears both output judgments and marks the revised source review as unblinded research.
Export the review bundle
Each record binds your judgment to the raw file hashes. Boundary points use decoded, EXIF-oriented native image pixels. Uncertain evidence is valid and remains unknown.
Complete the evidence and both explicit reviews to export.