CROSS-Bench
Harbor: pendingA benchmark of 28,000 decision problems evaluating candidate-bound visual contribution through matched invalidation and rebinding tests on vision-language models.
- Samples
- 28,000
- Released
- 2026
- Scores
- 4
EmoGenie
EmoGenie: Benchmarking and Evaluating Expressed and Evoked Emotion in T2I Generation with MLLMs
A benchmark of 6,400 AI-generated images and 91k QA pairs evaluating expressed/evoked emotion perception and assessment in T2I models and MLLMs.
How to run
ltb · emogenie
$ npm i -g lettokenburn-cli$ ltb bench info emogenie
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
13 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| GPT-Image-2 | 83.0 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| Qwen-Image | 79.8 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| Z-Image-Base | 75.0 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| Seedream-5.0 | 74.6 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| FLUX.2 Max | 72.9 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| PixArt-Σ | 63.9 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| SD3.5 | 63.8 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| Qwen3-VL-8B-Instruct | 62.5 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| Claude Sonnet 4.6 | 62.0 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| InternVL3.5-8B-Instruct | 61.3 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| Qwen2.5-VL-7B | 57.8 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| LLaVA-OneVision-1.5-8B | 57.3 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| SD3 | 53.5 | N/A | Independent | medium | Oct 13, 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 runsLoading community runs…
About
EmoGenie comprises 6,400 images generated by eight T2I models from 800 emotion-targeted prompts. It includes EmoGenie-QA (90k closed-set and 1,057 open-ended questions) and EmoGenie-Assess (13,175 human ratings for valence, arousal, dominance, and aesthetics).
Scoring uses accuracy for QA tasks and correlation with human Mean Opinion Scores (MOS) for continuous predictions. The associated EmoProbe model achieves state-of-the-art performance in unified coarse-to-fine affective understanding.