Skip to main content

​
Automated Rubric Evaluation

Every image generated by Defect Generation, Variation Generator, or Sim-to-Real is automatically evaluated by a vision-language judge against the job’s selected Evaluation Rubric Set.
  • 1–5 Scoring Scale: Each rubric metric is scored from 1 (severe issue) to 5 (excellent).
  • Attention Threshold (< 3): Any image with at least one metric scoring below 3 is highlighted in the Augment History so reviewers can inspect it first.

​
Reviewing & Regenerating Images

  1. Filter by Score or Status: In the Augment History panel, filter outputs by review status (Unreviewed, Approved, Rejected) or low rubric scores.
  2. Approve or Reject: Mark individual outputs as Approved or Rejected.
  3. In-Place Regeneration: Click Regenerate on any image to re-run that single task (with the same or an adjusted preset/prompt). When the new image finishes, it automatically replaces the previous image in the output dataset (superseded = true).