Multi-Human Image Generation on MultiHuman TestBench (1-5 People)
90.7Count AccuracyGPT-Image-1
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| GPT-Image-1Source Type=Proprietary2025.10 | 90.7 | 83.7 | 81 | 33.2 | 96.2 | 77 | |
| DISCOBase Model=Flux, Source Type=Ours2025.10 | 86.6 | 94.3 | 88.7 | 33.3 | 88.9 | 78.4 | |
| DISCOBase Model=Krea, Source Type=Ours2025.10 | 83.8 | 80.1 | 84.1 | 32.9 | 92.3 | 74.6 | |
| Qwen-ImageSource Type=Open-Source2025.10 | 80.3 | 47.9 | 50.6 | 33.2 | 94.5 | 61.3 | |
| DreamOSource Type=Open-Source2025.10 | 79.1 | 39 | 50.4 | 31.8 | 88.6 | 57.8 | |
| OmniGen2Source Type=Open-Source2025.10 | 74.8 | 45.7 | 36.5 | 33.5 | 88.2 | 55.7 | |
| Gemini-NanobananaSource Type=Proprietary2025.10 | 74 | 67.7 | 59.7 | 31.9 | 98.3 | 66.3 | |
| Krea-DevSource Type=Open-Source2025.10 | 67.3 | 52.2 | 55 | 31.2 | 92.6 | 59.7 | |
| Flux-DevSource Type=Open-Source2025.10 | 61.8 | 56.5 | 51.2 | 31.4 | 88.5 | 57.9 | |
| HiDreamSource Type=Open-Source2025.10 | 61.1 | 44.8 | 22.4 | 32.6 | 93.6 | 50.9 |