Research map / Video / Video generation
StoryBench: A Multifaceted Benchmark for Continuous Story Visualization
Emanuele Bugliarello, Hernan Moraldo, Ruben Villegas, Mohammad Babaeizadeh, Mohammad Taghi Saffar, Han Zhang, Dumitru Erhan, Vittorio Ferrari, Pieter-Jan Kindermans, Paul Voigtlaender
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Fine-tune Phenaki in continuation mode (mask only last 6 frames) instead of generation mode (ablation) | Phenaki-Gen-ST · Oops-CSV story continuation | FID: 250 -> 155 | Improved |
| Fine-tune Phenaki in continuation mode instead of generation mode (ablation) | Phenaki-Gen-ST · Oops-CSV story continuation | FVD: 589 -> 488 | Improved |
| Continuation-mode fine-tuning vs generation-mode fine-tuning (ablation) | Phenaki · Oops-CSV story continuation human eval | text adherence preference: 21 -> 76 | Improved |
| Fine-tune on StoryBench Oops-CSV in continuation mode vs zero-shot | Phenaki-Gen-ZS · Oops-CSV story continuation human eval | background consistency preference: 16 -> 65 | Improved |
| Fine-tune on StoryBench Oops-CSV in continuation mode vs zero-shot | Phenaki-Gen-ZS · Oops-CSV story continuation human eval | visual quality preference: 37 -> 49 | Mixed |
| Multi-task fine-tuning (Oops+DiDeMo+UVO) vs zero-shot (ablation) | Phenaki-Cont · Oops-CSV story continuation human eval | text adherence preference: 46 -> 46 | No clear effect |
| Multi-task instead of single-task continuation fine-tuning (ablation) | Phenaki-Cont · Oops-CSV action execution | FID: 114 -> 140 | Hurt |
| Multi-task instead of single-task continuation fine-tuning (ablation) | Phenaki-Cont · UVO-CSV story continuation | FVD: 1125 -> 618 | Improved |
| Fine-tune on original VidLN Oops captions instead of the story-transformed captions (ablation) | Phenaki-Cont-ST · Oops-CSV story continuation human eval | text adherence preference: 58 -> 49 | Hurt |
| Fine-tune continuation model vs zero-shot on story generation | Phenaki-Gen-ZS · Oops-CSV story generation | FID: 310 -> 171 | Improved |
| Fine-tune continuation model vs zero-shot on story generation | Phenaki-Gen-ZS · Oops-CSV story generation | FVD: 614 -> 711 | Hurt |
| Fine-tune for continuation vs zero-shot on story generation (human eval) | Phenaki-Gen-ZS · story generation human eval | human preference | Hurt |
Closest papers by meaning
- MovieBench: A Hierarchical Movie Level Dataset for Long Video Generation (CVPR 2025 · similarity 0.92)
- Is Your World Simulator a Good Story Presenter? A Consecutive Events-Based Benchmark for Future Long Video Generation (CVPR 2025 · similarity 0.90)
- ViStoryBench: Comprehensive Benchmark Suite for Story Visualization (CVPR 2026 · similarity 0.90)
- EntityBench: Towards Entity-Consistent Long-Range Multi-Shot Video Generation (NeurIPS 2026 · similarity 0.89)
- T2V-CompBench: A Comprehensive Benchmark for Compositional Text-to-video Generation (CVPR 2025 · similarity 0.89)
- SeriesBench: A Benchmark for Narrative-Driven Drama Series Understanding (CVPR 2025 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
