Click to enlarge ↗

Visual Quality Assessment for AI Graphics

Contracted by Contra Labs to evaluate the visual fidelity of AI-generated graphics across 80 unique prompts, each producing 4 distinct outputs. The assessment focused on three core dimensions:

Color Accuracy — How faithfully the generated output matched the intended palette, tone, and chromatic relationships specified in each prompt.

Click to enlarge ↗

Spatial Accuracy — Whether compositional elements, proportions, and spatial relationships were rendered correctly and consistently.

Click to enlarge ↗

Typography — Legibility, stylistic coherence, and structural integrity of any typographic elements within the generated outputs.

Click to enlarge ↗

Each output was scored and annotated with qualitative notes, building a structured dataset that maps where current AI image generation excels and where it breaks down. The work required a trained eye for visual detail and a systematic approach to subjective evaluation, bridging design expertise with analytical rigor.

This project sits at the intersection of design criticism and AI research: using the same principles that guide brand identity work (consistency, precision, intentionality) to stress-test machine-generated visuals.