<< All versions

Skill v1.0.0

currentAutomated scan100/100
tiger-ai-lab/rewardharness/evaluating-style-transfers
──Details
PublishedOctober 15, 2026 at 09:36 PM
Content Hashsha256:cd3d0384a2d44047...
Git SHA0c1e7051ab4f
──Files
Files (1 file, 977 B)
SKILL.md977 Bactive
SKILL.md · 17 lines · 977 B

version: "1.0.0" schema_version: 2 name: evaluating-style-transfers kind: skill description: Guidelines for evaluating artistic style transfers.


Evaluating Style Transfers

When evaluating prompts that ask for a style change (e.g., "mosaic-tile", "Michael Angelo", "Gustav Klimt"):

  1. Style Accuracy: Does the image genuinely reflect the requested style? Look for characteristic brushstrokes, textures, or color palettes.
  2. Content Preservation: Are the original subjects and composition still recognizable? A good style transfer applies the style without completely destroying the underlying content.
  3. Over-editing: Did the model replace the subject with something entirely different (e.g., replacing a bear with a generic Klimt painting)? This is a failure.
  4. Consistency: Is the style applied consistently across the entire image, or are there jarring patches of the original photo?

<!-- Release review: RewardHarness v0.4.1 (2026-08-26). -->

All versions