Skip to main content
  • Pricing
AI image

AI Image Model Counting Test for Repeated Icons

By Oakgen TeamPublished 1 min read

Share
AI Image Model Counting Test for Repeated Icons

A chart or card can look balanced with the wrong number of marks. If the symbols carry a count, a missing or extra one changes the message. Assess the placements numerically and keep illustration quality separate from the ability to follow exact quantity and position directions.

Build the test input

Make a plain brief for a blank landing-page graphic with seven outlined stars in one row. The third and sixth stars should be filled. Exclude other decorative shapes in that row. Keep approved spacing reference marks on a separate overlay to avoid confusing them with the intended output.

Run the proposed comparison

Send the same text brief and the same supported aspect ratio to each candidate. In a second condition, supply a simple composition reference if every candidate supports it. Keep text-only and reference-led results distinct. Save unedited files and record how many marks are legible without zooming.

Decide what passes

Count all outlines first. Then count the filled marks and identify their ordinal positions. Classify merged half-stars as ambiguous rather than choosing the generous reading. Require the intended count and order to pass if this graphic represents a genuine feature count or dashboard state.

Act on the outcome

A pretty output that fails the ordering can be reused as a style board for manual icon assembly. For a repeated ad template, use a deterministic layout and delegate only decorative assets to the model. Try a second count before treating one seven-star pass as evidence of arbitrary counting skill.

Model references

Share