Research map / Video / Video generation
Product of Experts for Visual Generation
Yunzhi Zhang, Carson Murtuza-Lanier, Zizhang Li, Yilun Du, Jiajun Wu
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 conditional sampling of generative experts (ablation) | Ours · Image object insertion (graphics engine input, 30 scenes) | MSE background: 1.511 -> 1.429 | Improved |
| add conditional sampling of generative experts (ablation) | Ours · Image object insertion (graphics engine input, 30 scenes) | GPT-4o controllability: 0.727 -> 0.827 | Improved |
| add conditional sampling of generative experts (ablation) | Ours · Image object insertion (graphics engine input, 30 scenes) | GPT-4o text alignment: 0.675 -> 0.817 | Improved |
| add conditional sampling of generative experts (ablation) | Ours · Image object insertion (graphics engine input, 30 scenes) | Aesthetic: 0.599 -> 0.596 | Hurt |
| increase particle count from L=1 to L=32 | Ours · MIG-Bench layout control (50 scenes) | VQAScore: 0.567 -> 0.904 | Improved |
| increase particle count from L=1 to L=32 | Ours · MIG-Bench layout control (50 scenes) | mIoU: 0.675 -> 0.728 | Improved |
| add SMC resampling at L=32 (ablation) | Ours · MIG-Bench layout control (50 scenes) | VQAScore: 0.854 -> 0.904 | Improved |
| add SMC resampling at L=32 (ablation) | Ours · MIG-Bench layout control (50 scenes) | mIoU: 0.691 -> 0.728 | Improved |
| add SMC resampling at L=4 (ablation) | Ours · MIG-Bench layout control (50 scenes) | mIoU: 0.715 -> 0.695 | Hurt |
| add SMC resampling at L=8 (ablation) | Ours · MIG-Bench layout control (50 scenes) | mIoU: 0.735 -> 0.715 | Hurt |
| add SMC resampling at L=8 (ablation) | Ours · MIG-Bench layout control (50 scenes) | VQAScore: 0.807 -> 0.879 | Improved |
| compose experts with Ours instead of Image2V | Wan2.1 14B · Physics-simulator video generation (object-centric simulation, 12 scenes) | GPT-4o controllability: 0.708 -> 0.708 | No clear effect |
Closest papers by meaning
- UniControl: A Unified Diffusion Model for Controllable Visual Generation In the Wild (NeurIPS 2023 · similarity 0.88)
- DuoGen: Towards Autonomous Interleaved Multimodal Generation (CVPR 2026 · similarity 0.87)
- ThinkGen: Generalized Thinking for Visual Generation (CVPR 2026 · similarity 0.86)
- Vision Transformers as Probabilistic Expansion from Learngene (ICML 2024 · similarity 0.86)
- Imagine Before You Draw: Visual Prompt Engineering for Image Generation (NeurIPS 2026 · similarity 0.86)
- Physics-Informed Video Generation via Mixture-of-Experts Latent Alignment (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.
