Visual World Modelling on MagicBrush
8.14GPT-4o ScoreBAGEL-14B
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| BAGEL-14B2026.02 | 8.14 | 0.44 | 0.95 | |
| UNIWORLD-V12026.02 | 7.42 | 0.4 | 0.91 | |
| OMNIGEN22026.02 | 7.04 | 0.48 | 0.92 | |
| OMNIGEN2026.02 | 6.86 | 0.48 | 0.94 | |
| SMARTEDIT2026.02 | 6.71 | — | — | |
| SWIRL (ITERATIVE)Method Strategy (SFT/Bootstrap/SWIRL)=SWIRL (ITERATIVE)2026.02 | 6.62 | 0.3 | 0.92 | |
| SWIRL (IDM → FWM)Method Strategy (SFT/Bootstrap/SWIRL)=SWIRL (IDM → FWM)2026.02 | 6.48 | 0.26 | 0.92 | |
| LIQUID-SFTTest-time Verification (N)=82026.02 | 6.18 | 0.16 | 0.92 | |
| LIQUID-SFTTest-time Verification (N)=42026.02 | 6.1 | 0.14 | 0.93 | |
| SWIRL (ITER. + SHARE)Method Strategy (SFT/Bootstrap/SWIRL)=SWIRL (ITER. + SHARE)2026.02 | 6 | 0.28 | 0.92 | |
| GOT2026.02 | 5.96 | — | — | |
| LIQUID-SFTTest-time Verification (N)=22026.02 | 5.71 | 0.14 | 0.93 | |
| LIQUID-SFTMethod Strategy (SFT/Bootstrap/SWIRL)=SFT2026.02 | 5.46 | 0.28 | 0.92 | |
| LIQUID-BOOTSTRAPMethod Strategy (SFT/Bootstrap/SWIRL)=Bootstrap2026.02 | 5.27 | 0.3 | 0.89 | |
| BLIP3O-NEXT2026.02 | 3.42 | 0.7 | 0.74 | |
| CHAMELEON-BOOTSTRAP2026.02 | 3.27 | — | — | |
| InstructPix2Pix2026.02 | 3.12 | — | — | |
| CHAMELEON-SFT2026.02 | 2.52 | — | — |