Research map / Video / Video generation
LUVE : Latent-Cascaded Ultra-High-Resolution Video Generation with Dual Frequency Experts
Chen Zhao, Jiawei Chen, Hongyu Li, Zhuoliang Kang, Shilin Lu, Xiaoming Wei, Kai Zhang, Jian Yang, Ying Tai
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 |
|---|---|---|---|
| replace RGB/latent interpolation with the learned video latent upsampler (VLUer) (ablation) | LUVE 2K · 2K ablation | FID_patch: 51.75 (RGB) / 47.80 (latent) -> 41.03 | Improved |
| replace RGB interpolation with latent-space upsampler (ablation) | LUVE 2K · 2K ablation | latency: 40.12s -> 0.922s | Improved |
| use learned upsampler instead of plain latent interpolation (ablation) | LUVE 2K · 2K ablation | latency: 0.004s -> 0.922s | Hurt |
| add cascaded low-res motion stage over end-to-end UHR scaling (ablation) | LUVE 2K · 2K ablation | FID_patch: 54.10 -> 46.48 | Improved |
| add two standard LoRA experts (not frequency-specific) (ablation) | cascaded LUVE · 2K ablation | FID_patch: 46.48 -> 47.03 | Hurt |
| add low-frequency expert (ablation) | LUVE 2K · 2K ablation | FID_patch / Realism: FID_patch 43.86 -> 41.03; Realism 7.08 -> 7.64 | Improved |
| add high-frequency expert (ablation) | LUVE 2K · 2K ablation | FID_patch / AQ: FID_patch 44.44 -> 41.03; AQ 59.34 -> 59.78 | Improved |
| add HPS v3 data selection (ablation) | LUVE 2K · 2K ablation | FID_patch: 43.77 -> 41.03 | Improved |
| add Unsharp Masking augmentation for HFE data (ablation) | LUVE 2K · 2K ablation | FID_patch: 42.96 -> 41.03 | Improved |
| add decoder in VLUer (ablation) | VLUer · qualitative | blurriness | Improved |
| add pixel-space and frame-difference loss to latent L1 loss (ablation) | VLUer · qualitative | blocky artifacts | Improved |
| replace training-free/VSR approaches with LUVE | Wan2.1-1.3B · VBench | average score: 83.75 (UltraWan-4K) -> 84.34 (Ours-2K) | Improved |
Closest papers by meaning
- Upscale-A-Video: Temporal-Consistent Diffusion Model for Real-World Video Super-Resolution (CVPR 2024 · similarity 0.87)
- LinGen: Towards High-Resolution Minute-Length Text-to-Video Generation with Linear Computational Complexity (CVPR 2025 · similarity 0.87)
- Snap Video: Scaled Spatiotemporal Transformers for Text-to-Video Synthesis (CVPR 2024 · similarity 0.87)
- Ultra Flash: Scaling Real-Time Streaming Video Generation to High Resolutions (NeurIPS 2026 · similarity 0.86)
- UltraHR-100K: Enhancing UHR Image Synthesis with A Large-Scale High-Quality Dataset (NeurIPS 2025 · similarity 0.86)
- Lumina-Next : Making Lumina-T2X Stronger and Faster with Next-DiT (NeurIPS 2024 · 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.
