Research map / Video / Video generation
VIVA: VLM-Guided Instruction-Based Video Editing with Reward Optimization
Xiaoyan Cong, Haotian Yang, Angtian Wang, Yizhi Wang, Yiding Yang, Canyu Zhang, Chongyang Ma
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 |
|---|---|---|---|
| add VLM instructor conditioning (ablation) | vanilla DiT editor · VIE-Bench Add | Avg VLM score: 4.57 -> 6.86 | Improved |
| add VLM instructor conditioning (ablation) | vanilla DiT editor · VIE-Bench Remove | Avg VLM score: 2.62 -> 7.65 | Improved |
| add masked flow-matching loss (ablation) | VIVA + VLM instructor · VIE-Bench Add | Avg VLM score: 6.86 -> 8.14 | Improved |
| add masked flow-matching loss (ablation) | VIVA + VLM instructor · VIE-Bench Replace | Avg VLM score: 6.50 -> 6.51 | No clear effect |
| mix image-editing data into training (ablation) | VIVA + V + M · VIE-Bench Replace | Avg VLM score: 6.51 -> 8.82 | Improved |
| mix image-editing data into training (ablation) | VIVA + V + M · VIE-Bench Add | Avg VLM score: 8.14 -> 7.91 | Mixed |
| add Edit-GRPO post-training (ablation) | VIVA + V + M + I · VIE-Bench Add | Avg VLM score: 7.91 -> 8.86 | Improved |
| add Edit-GRPO post-training (ablation) | VIVA + V + M + I · VIE-Bench Replace | Avg VLM score: 8.82 -> 8.86 | No clear effect |
| add Edit-GRPO post-training (ablation) | VIVA + V + M + I · VIE-Bench Add | PickScore: 21.03 -> 20.90 | Hurt |
| use VIVA vs commercial Runway Gen-4 Aleph | Runway Gen-4 Aleph · VIE-Bench hybrid edit | Avg VLM score: 7.85 -> 5.88 | Hurt |
| use VIVA vs Runway on reference-based replace | Runway Gen-4 Aleph · VIE-Bench reference-based Replace | Avg VLM score: 8.07 -> 8.74 | Improved |
Closest papers by meaning
- Transforming Image Editors into Video Editors (NeurIPS 2026 · similarity 0.89)
- Señorita-2M: A High-Quality Instruction-based Dataset for General Video Editing by Video Specialists (NeurIPS 2025 · similarity 0.89)
- Text-Vision Co-Instructed Image Editing (NeurIPS 2026 · similarity 0.88)
- VicEdit: Learning to Edit Videos from Visual In-Context Examples (NeurIPS 2026 · similarity 0.88)
- VINCIE-NExT: Unlocking Video Editing from Images via In-Context Modeling (NeurIPS 2026 · similarity 0.87)
- InsightEdit: Towards Better Instruction Following for Image Editing (CVPR 2025 · similarity 0.87)
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.
