Research map / Video / Video generation
REDUCIO! Generating 1K Video within 16 Seconds using Extremely Compressed Motion Latents
Rui Tian, Qi Dai, Jianmin Bao, Kai Qiu, Yifan Yang, Chong Luo, Zuxuan Wu, Yu-Gang Jiang
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 content frame fusion via cross-attention in the VAE decoder (ablation) | Reducio-VAE · Pexels reconstruction | PSNR: 27.91 -> 35.88 | Improved |
| use cross-attention fusion instead of add or linear fusion (ablation) | Reducio-VAE · Pexels reconstruction | PSNR: 30.86 -> 35.88 (linear), 34.17 -> 35.88 (add) | Improved |
| use middle frame rather than a random frame as content frame (ablation) | Reducio-VAE · Pexels reconstruction | PSNR: 31.72 -> 35.88 | Improved |
| increase temporal downsampling from 2 to 4 at fs=32 (ablation) | Reducio-VAE · Pexels reconstruction | PSNR: 36.79 -> 35.88 | Hurt |
| shift compression from temporal to spatial at identical total factor (ablation) | Reducio-VAE · Pexels reconstruction | PSNR: 34.14 -> 35.88 | Improved |
| increase latent channels from 4 to 16 (ablation) | Reducio-VAE · Pexels reconstruction | PSNR: 33.17 -> 35.88 | Improved |
| increase latent channels from 16 to 32 (ablation) | Reducio-VAE · Pexels reconstruction | PSNR: 35.88 -> 35.24 | No clear effect |
| use 3D convolution instead of 2D convolution in the VAE (ablation) | Reducio-VAE · Pexels reconstruction | PSNR: 30.22 -> 35.51 | Improved |
| use joint 3D attention instead of factorized 2D+1D attention (ablation) | Reducio-DiT · UCF-101 FVD | FVD: 382.2 -> 337.6 | Improved |
| use both semantic and content image conditions rather than semantic only (ablation) | Reducio-DiT · UCF-101 FVD | FVD: 472.1 -> 337.6 | Improved |
| add semantic features to content-only conditioning (ablation) | Reducio-DiT · UCF-101 FVD / IS | FVD: 363.7 -> 337.6 | Improved |
| use patch size 1 with high VAE compression instead of temporal patch size 2 (ablation) | Reducio-DiT · UCF-101 FVD | FVD: 351.0 -> 337.6 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 3 on most shared results.
- Lavie — LaVie: High-Quality Video Generation with Cascaded Latent Diffusion Models (IJCV 2024) · better on 3 of 3 shared results
- DynamiCrafter — DynamiCrafter: Animating Open-domain Images with Video Diffusion Priors (ECCV 2024) · better on 1 of 2 shared results
- Show-1 — Show-1: Marrying Pixel and Latent Diffusion Models for Text-to-Video Generation (IJCV 2024) · 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
- LinGen: Towards High-Resolution Minute-Length Text-to-Video Generation with Linear Computational Complexity (CVPR 2025 · similarity 0.86)
- Efficient Video Diffusion Models via Content-Frame Motion-Latent Decomposition (ICLR 2024 · similarity 0.86)
- LongDiff: Training-Free Long Video Generation in One Go (CVPR 2025 · similarity 0.86)
- From Slow Bidirectional to Fast Autoregressive Video Diffusion Models (CVPR 2025 · similarity 0.86)
- Masked Generative Nested Transformers with Decode Time Scaling (ICML 2025 · similarity 0.86)
- VideoMAR: Autoregressive Video Generation with Continuous Tokens (NeurIPS 2025 · 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.
