HermanArt Bench

Experiment · active

Clockwork owl identity consistency

How consistently does one model preserve Herman’s identity across repeated unseeded trials?

Runs
2
Variable
unseeded trial variation
Prompt
ab-owl-identity-001-prompt v1.0

Protocol

Launch experiment: two first-party-tool trials using nearly identical prompts. The only prompt-text change is the trial identifier.

Frozen prompt template

A dignified clockwork owl named Herman perched on a large aged-brass skeleton key, with midnight navy enamel, exposed gears, half-moon spectacles, one subtle amber eye, and art-deco geometry.

Review rubric

  • Prompt Adherence
  • Identity
  • Material Realism
  • Mechanical Coherence

Notes

Failed, moderated, and discarded runs remain in the public data when legally releasable.

Evidence

Recorded runs

Generated specimen ab-owl-identity-001-grok-trial-01 ab-owl-identity-001-grok-trial-01
successgrok-imagine-image

Ab Owl Identity 001 Grok Trial 01

Strong owl, navy enamel, brass, key, and art-deco fan adherence. Spectacles are round rather than half-moon. Exposed gears are plausible but not a demonstrably complete functional train.

Size
1024×1024
Seed
not exposed
Method
first party tool
View experiment →
Generated specimen ab-owl-identity-001-grok-trial-02 ab-owl-identity-001-grok-trial-02
successgrok-imagine-image

Ab Owl Identity 001 Grok Trial 02

Strong owl, navy enamel, brass key, and explicit art-deco geometry adherence. One amber eye is dominant. Spectacles are round rather than half-moon; visible gears mesh locally.

Size
1024×1024
Seed
not exposed
Method
first party tool
View experiment →