ID-Crafter: VLM-Grounded Online RL for Compositional Multi-Subject Video Generation

Benchmark Model Rank Results
open-domain-subject-to-video-on-opens2v-evalID-Crafter (1.3B, Base Model + online RL)–Total Score: 0.5516Aesthetics: 0.4885Motion: 0.365FaceSim: 0.661…
open-domain-subject-to-video-on-opens2v-evalID-Crafter (1.3B, Base Model)–Total Score: 0.5433Aesthetics: 0.425Motion: 0.38FaceSim: 0.5812…
open-domain-subject-to-video-on-opens2v-evalID-Crafter (14B, Large-scale Model)–Total Score: 0.5705Aesthetics: 0.4528Motion: 0.4034FaceSim: 0.6071…