Research map / Video / Video generation
Vivid-VR: Distilling Concepts from Text-to-Video Diffusion Transformer for Photorealistic Video Restoration
Haoran Bai, Xiaoxu Chen, Canqian Yang, Zongyao He, Sibin Deng, Ying Chen
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 concept distillation (100K T2V-generated training videos) (ablation) | Vivid-VR · UGC50 | MUSIQ: 57.36 -> 67.61 | Improved |
| add concept distillation (ablation) | Vivid-VR · UGC50 | NIQE: 5.364 -> 4.361 | Improved |
| use Qwen2.5-VL captioner instead of concept distillation (ablation) | Vivid-VR · UGC50 | MUSIQ: 67.61 -> 60.88 | Hurt |
| add control feature projector (ablation) | Vivid-VR · UGC50 | MUSIQ: 63.06 -> 67.61 | Improved |
| fine-tune VAE encoder instead of projector (ablation) | Vivid-VR · UGC50 | MUSIQ: 67.61 -> 64.31 | Hurt |
| add cross-attention branch to MLP connector (ablation) | Vivid-VR · UGC50 | MUSIQ: 59.78 -> 67.61 | Improved |
| keep MLP branch instead of skip connection (ablation) | Vivid-VR · UGC50 | MUSIQ: 63.91 -> 67.61 | Improved |
| use dual-branch connector instead of ZeroSFT (ablation) | Vivid-VR · UGC50 | MUSIQ: 61.21 -> 67.61 | Improved |
| remove MLP branch entirely (ablation) | Vivid-VR · UGC50 | convergence | Hurt |
| increase generated training videos from 0 to 100K (ablation) | Vivid-VR · UGC50 | MUSIQ: 57.36 -> 67.61 | Improved |
| increase generated training videos from 100K to 150K (ablation) | Vivid-VR · UGC50 | MUSIQ: 67.61 -> 67.19 | No clear effect |
| train on generated videos only (no source videos) (ablation) | Vivid-VR · UGC50 | MUSIQ: 67.19 -> 53.77 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- SeedVR2-7B — SeedVR: Seeding Infinity in Diffusion Transformer Towards Generic Video Restoration (CVPR 2025) · better on 0 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
- VD3D: Taming Large Video Diffusion Transformers for 3D Camera Control (ICLR 2025 · similarity 0.87)
- One-Step Diffusion for Detail-Rich and Temporally Consistent Video Super-Resolution (NeurIPS 2025 · similarity 0.86)
- SeedVR: Seeding Infinity in Diffusion Transformer Towards Generic Video Restoration (CVPR 2025 · similarity 0.86)
- SeeClear: Semantic Distillation Enhances Pixel Condensation for Video Super-Resolution (NeurIPS 2024 · similarity 0.86)
- Upscale-A-Video: Temporal-Consistent Diffusion Model for Real-World Video Super-Resolution (CVPR 2024 · similarity 0.86)
- VideoElevator: Elevating Video Generation Quality with Versatile Text-to-Image Diffusion Models (AAAI 2025 · 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.
