Research map / Video / Video generation
Your Data Manifold is Secretly a Reward Model: Shell-LCC for Text-to-Video Generation
Shihao Zhang, Yunzhi Li, Yuguang Yan, Junzhe Zhang, WEI ZHAO, Bohan Wang, Hanwang Zhang
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 Shell-LCC manifold reward to SFT model | private 4.5B T2V · VBench | Imaging Quality: 0.7509 -> 0.7631 | Improved |
| add Shell-LCC to Wan-T2V-1.3B | Wan-T2V-1.3B · VBench | Imaging Quality: 0.6629 -> 0.7037 | Improved |
| add Shell-LCC to UltraWan-T2V-1.3B | UltraWan-T2V-1.3B · VBench | Aesthetic Quality: 0.5744 -> 0.6299 | Improved |
| add DPO to SFT baseline | private 4.5B T2V · VBench | Imaging Quality: 0.7509 -> 0.7397 | Hurt |
| add DPO to SFT baseline | private 4.5B T2V · VBench | Aesthetic Quality: 0.6724 -> 0.6784 | Improved |
| add Shell-LCC to SFT baseline | private 4.5B T2V · VBench | Aesthetic Quality: 0.6724 -> 0.6735 | No clear effect |
| add Shell-LCC to SFT baseline | private 4.5B T2V · VBench | Motion Smoothness / Subject Consistency: 0.9891 -> 0.9900; 0.9680 -> 0.9682 | No clear effect |
| add Shell-LCC | Wan-T2V-1.3B · VBench | Overall Consistency: 0.2271 -> 0.2292 | No clear effect |
| add Shell-LCC | UltraWan-T2V-1.3B · VBench | Overall Consistency: 0.2288 -> 0.2236 | No clear effect |
| drop the locality constraint in LCC (reconstruction only) (ablation) | LCC manifold · distinguishability | accuracy: -9.5% | Hurt |
| use Shell-LCC with scale vector instead of plain LCC (ablation) | manifold model · distinguishability | accuracy: 92.7 -> 91.4 | Hurt |
| vary anchors M, tau1, tau2 (ablation) | Shell-LCC · distinguishability | accuracy: M 90.8/91.4/91.8; tau changes <0.3% | No clear effect |
Closest papers by meaning
- InstructVideo: Instructing Video Diffusion Models with Human Feedback (CVPR 2024 · similarity 0.87)
- Inference-Time Text-to-Video Alignment with Diffusion Latent Beam Search (NeurIPS 2025 · similarity 0.86)
- T2V-Turbo: Breaking the Quality Bottleneck of Video Consistency Model with Mixed Reward Feedback (NeurIPS 2024 · similarity 0.86)
- VideoMLA: Low-Rank Latent KV Cache for Minute-Scale Autoregressive Video Diffusion (NeurIPS 2026 · similarity 0.86)
- Align-A-Video: Deterministic Reward Tuning of Image Diffusion Models for Consistent Video Editing (CVPR 2025 · similarity 0.86)
- VideoCrafter2: Overcoming Data Limitations for High-Quality Video Diffusion Models (CVPR 2024 · similarity 0.85)
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.
