Megadose AI progress, ranked and analyzed.

ShotPlan: Cinematic Video Generation with Learnable Planning Token

· HF Daily Papers ·
ShotPlan adds explicit, frame-level shot planning to video diffusion for multi-shot cinematic generation.

The paper says current models are strong at single shots but weaker at coherent multi-shot sequences. Its planning tokens capture transition cues and control when shot changes happen. FRoPE lets those transitions be modeled at the frame level. The authors report stronger inter-shot consistency and more flexible shot management than existing cinematic video generation methods. HF Daily Papers' note

score 5

Categories: Research