Research map / Video / Video generation
UnityVideo: Unified Multi-Modal Multi-Task Learning for Enhancing World-Aware Video Generation
Jiehui Huang, Yuechen Zhang, Xu He, Yuan Gao, Zhi Cen, Bin Xia, Yan Zhou, Xin Tao, Pengfei Wan, Jiaya Jia
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 |
|---|---|---|---|
| train jointly on optical flow (ablation) | RGB video baseline · Modality ablation (text-to-video, VBench) | Subject consistency: 96.51 -> 97.82 | Improved |
| train jointly on depth (ablation) | RGB video baseline · Modality ablation (text-to-video, VBench) | Subject consistency: 96.51 -> 98.13 | Improved |
| unified multi-modality instead of depth only (ablation) | UnityVideo · Modality ablation (text-to-video, VBench) | Subject consistency: 98.13 -> 98.01 | No clear effect |
| unified multi-modality instead of depth only (ablation) | UnityVideo · Modality ablation (text-to-video, VBench) | Imaging quality: 69.09 -> 69.18 | No clear effect |
| train only controllable generation (ablation) | RGB video baseline · Task ablation (depth-guided, VBench) | Background consistency: 96.06 -> 95.58 | Hurt |
| unified multi-task instead of ControlGen only (ablation) | UnityVideo · Task ablation (depth-guided, VBench) | Background consistency: 95.58 -> 96.08 | Improved |
| unified multi-task instead of JointGen only (ablation) | UnityVideo · Task ablation (depth-guided, VBench) | Background consistency: 97.24 -> 97.18 | Mixed |
| add In-Context Learner (ablation) | UnityVideo · Architecture ablation (depth-conditioned T2V, VBench) | Subject consistency: 96.51 -> 97.92 | Improved |
| add Modality Switcher (ablation) | UnityVideo · Architecture ablation (depth-conditioned T2V, VBench) | Subject consistency: 96.51 -> 97.94 | Improved |
| combine In-Context Learner and Modality Switcher (ablation) | UnityVideo · Architecture ablation (depth-conditioned T2V, VBench) | Subject consistency: 97.94 -> 98.31 | Improved |
| UnityVideo vs Wan2.1 on overall consistency | UnityVideo · VBench text-to-video | Overall consistency: 21.53 -> 23.57 | Improved |
| UnityVideo vs SeC on segmentation | UnityVideo · Video segmentation (UniBench) | mIoU: 65.52 -> 68.82 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.
- DepthCrafter — DepthCrafter: Generating Consistent Long Depth Sequences for Open-world Videos (CVPR 2025) · better on 1 of 1 shared result
- SAMWISE — SAMWISE: Infusing Wisdom in SAM2 for Text-Driven Video Segmentation (CVPR 2025) · better on 1 of 1 shared result
- Aether — Aether: Geometric-Aware Unified World Modeling (ICCV 2025) · better on 1 of 1 shared result
- full-dit — FullDiT: Video Generative Foundation Models with Multimodal Control via Full Attention (ICCV 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Divide and Conquer: Decoupled Representation Alignment for Multimodal World Models (ECCV 2026 · similarity 0.90)
- UniReal: Universal Image Generation and Editing via Learning Real-world Dynamics (CVPR 2025 · similarity 0.89)
- WorldWeaver: Generating Long-Horizon Video Worlds via Rich Perception (NeurIPS 2025 · similarity 0.88)
- Learning World Models for Interactive Video Generation (NeurIPS 2025 · similarity 0.87)
- MVBench: A Comprehensive Multi-modal Video Understanding Benchmark (CVPR 2024 · similarity 0.87)
- WorldReasonBench: Human-Aligned Stress Testing of Video Generators as Future World-State Predictors (NeurIPS 2026 · similarity 0.86)
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.
