Skip to main content
  • Pricing
AI image

Nano Banana Animal Reference Proportion Test

By Oakgen TeamPublished 1 min read

Share
Nano Banana Animal Reference Proportion Test

Photoreal output does not ensure a pet or character still looks like the subject. Use a small review sheet with relative measurements before choosing a model for a consistent reference series. Keep the requested environment change simple so shape differences can be traced easily.

Build the test input

Use a permitted pet image or an original fictional quadruped with an unusually broad muzzle, one notched ear, and a narrow tail. Request the same pose in a new indoor scene. Choose a shot where the original landmarks are visible at comparable angles. Save the base image for reference.

Run the proposed comparison

Test the same image and prompt at a shared aspect ratio. Trace the ear outline, muzzle width relative to the face, and tail curve. Avoid directly comparing ratios across angles unless intentional perspective change is part of the task. Keep coat color and anatomy fidelity as two separate observations.

Decide what passes

Pass if the recognizable profile landmarks remain and no extra legs or ears appear. A high-quality answer that smooths the notched ear into an ordinary pet still fails a character-reference brief. An angle that hides a landmark can be marked unobservable rather than scored as either present or omitted.

Act on the outcome

When a candidate misses one landmark but generally captures pose and fur, simplify the source or use a stronger identity image if supported. For a matching series, only count outcomes with enough visible evidence to approve the intended character. Save the acceptable views before testing an entirely different indoor environment.

Model references

Share