World Model Task Execution on WorldTasks-Bench
60.5Task Completion RateLTX-2 (8-Step)+WMSD
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| LTX-2 (8-Step)+WMSDBase Model=LTX-2, Sampling Steps=82026.06 | 60.5 | 69.1 | 88.2 | 72.6 | 10.1 | |
| HY1.5+WMSD*Base Model=HunyuanVideo-1.5, Training=GRPO for 25 steps2026.06 | 57.4 | 63 | 82.8 | 67.3 | 112 | |
| LTX-2 (8-Step)+VLMBase Model=LTX-2, Sampling Steps=8, Augmentation=VLM reasoning2026.06 | 49.5 | 57.2 | 73.2 | 59.8 | 10.5 | |
| HY1.5Base Model=HunyuanVideo-1.52026.06 | 46.4 | 54 | 78 | 59.7 | 112 | |
| LTX-2+WMSD*Base Model=LTX-2, Training=GRPO for 25 steps2026.06 | 45.2 | 50 | 69.3 | 54.8 | 52.2 | |
| LTX-2Base Model=LTX-22026.06 | 31.5 | 39.5 | 69 | 46.7 | 52.2 | |
| LTX-2+SFTBase Model=LTX-2, Training=Supervised fine-tuning2026.06 | 29.2 | 38.9 | 68.2 | 45.4 | 52.2 | |
| LTX-2 (8-Step)Base Model=LTX-2, Sampling Steps=82026.06 | 28.5 | 39.1 | 69.4 | 45.5 | 10.1 |