Informazioni
October 2026 Kombai paper evaluating originality, variation across product domains, and requirement satisfaction in LLM-generated interfaces. It compares 2,400 HTML designs from 15 models across 160 prompts with human reference designs, separating acceptable output from diverse design.
Sintesi
Why This Benchmark Adds New Signal:
1. Separate creativity measures: Distinguishes cross-model originality, creative range across paired briefs, and acceptance-criteria coverage
2. Structural comparison: Combines rendered UI segments and a separate color score to measure design repetition
3. Observed diversity gap: Models satisfy more than 90% of criteria on average yet vary less across briefs than the human reference
4. Evaluation boundary: Tests desktop HTML with default sampling and no diversity-seeking prompts; human designers adapt public inspiration references
1. Separate creativity measures: Distinguishes cross-model originality, creative range across paired briefs, and acceptance-criteria coverage
2. Structural comparison: Combines rendered UI segments and a separate color score to measure design repetition
3. Observed diversity gap: Models satisfy more than 90% of criteria on average yet vary less across briefs than the human reference
4. Evaluation boundary: Tests desktop HTML with default sampling and no diversity-seeking prompts; human designers adapt public inspiration references
Tag
ui-generationdesign-diversitycreativity-evaluationhuman-baselinebenchmark