PortraitAes-Bench
PortraitAes: Intent-Conditioned Structured Portrait Aesthetics Assessment
An 11K-scale benchmark of structured records for 1.6K portraits, measuring intent-conditioned aesthetic quality across nine intents and six dimensions.
How to run
$ npm i -g lettokenburn-cli$ ltb bench info portraitaes-bench
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
Scores
8 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| Qwen3.5-VL-9B | 0.9 | N/A | Independent | medium | Oct 4, 2026 | arxiv.org |
| ArtiMuse | 0.8 | N/A | Independent | medium | Oct 4, 2026 | arxiv.org |
| PortraitCraft-4B | 0.8 | N/A | Independent | medium | Oct 4, 2026 | arxiv.org |
| Qwen3.5 Flash | 0.8 | N/A | Independent | medium | Oct 4, 2026 | arxiv.org |
| UniPercept | 0.7 | N/A | Independent | medium | Oct 4, 2026 | arxiv.org |
| Q-Align | 0.7 | N/A | Independent | medium | Oct 4, 2026 | arxiv.org |
| InternVL3-8B | 0.7 | N/A | Independent | medium | Oct 4, 2026 | arxiv.org |
| HumanAesExpert-8B | 0.5 | N/A | Independent | medium | Oct 4, 2026 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
Reproduced by the community
All runsAbout
PortraitAes-Bench comprises 11,000 expert-authored records for 1,600 human-centric images, decomposing assessment into intent routing, six specialist dimensions, and 22 secondary criteria with visible-evidence rationales.
Scoring uses Pearson and Spearman correlations against rubric-derived targets. The associated PortraitAes model achieves a Pearson correlation of 0.924 and Spearman rank correlation of 0.934 on the standard set, outperforming general-purpose MLLMs.