Multi-subject in-context image generation on MSIC
0.339CLIP-T ScoreOracle (caption)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Oracle (caption)input=ground-truth captions2026.04 | 0.339 | — | — | |
| MUSIC*test-time scaling=true2026.04 | 0.33 | 0.631 | 0.822 | |
| MUSIC2026.04 | 0.322 | 0.622 | 0.812 | |
| OmniGen2026.04 | 0.3 | 0.525 | 0.714 | |
| UNO2026.04 | 0.296 | 0.541 | 0.721 | |
| MS-Diffusion2026.04 | 0.29 | 0.532 | 0.714 | |
| MIP-Adapter2026.04 | 0.288 | 0.497 | 0.715 | |
| Subject Diffusion2026.04 | 0.287 | 0.513 | 0.702 |