AI Paper Atlas

Research map / Video / Video generation

Show-1: Marrying Pixel and Latent Diffusion Models for Text-to-Video Generation

David Junhao Zhang, Jay Zhangjie Wu, Jiawei Liu, Rui Zhao, Lingmin Ran, Yuchao Gu, Difei Gao, Mike Zheng Shou

IJCV 2024 · Video › Video generation · approach: Compositional text-to-video diffusion · 275 citations (OpenAlex)

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 changedWhereMeasuredEffect
use pixel-based instead of latent (f=8) VDM for 64x40 keyframes (ablation)Show-1 · Keyframe stage ablationCLIPSIM: 0.2441 -> 0.3096Improved
use pixel-based instead of latent (f=8) VDM for 64x40 keyframes (ablation)Show-1 · UCF-101FVD: 584 -> 383Improved
generate low-resolution 64x40 keyframes instead of high-resolution 512x320 with latent VDM (f=8) (ablation)Show-1 · Keyframe stage ablationCLIPSIM: 0.2793 -> 0.3096Improved
use latent final super-resolution instead of pixel final super-resolution (ablation)Show-1 · Inference memorymax memory: 72GB -> 15GBImproved
lower latent compression ratio (f=8 -> f=2) at 64x40 keyframes (ablation)latent keyframe VDM · Keyframe stage ablationCLIPSIM: 0.2441 -> 0.2742Improved
increase resolution of latent keyframes from 256x160 to 512x320 (f=8) (ablation)latent keyframe VDM · Keyframe stage ablationCLIPSIM: 0.2874 -> 0.2793Hurt
change input to noised-denoised (SDEdit) low-res video in final super-resolution (ablation)SDx4 with temporal layers · UCF-101FVD: 459 -> 423Improved
add expert finetuning (timesteps 0-900) to expert translation (ablation)Final super-resolution module · UCF-101FVD: 423 -> 383Improved
add expert finetuning (timesteps 0-900) to expert translation (ablation)Final super-resolution module · UCF-101IS: 33.83 -> 35.67Improved
finetune whole pipeline on square videos instead of resizing outputsShow-1 · UCF-101FVD: 383.46 -> 369.33Improved
use Show-1 hybrid pipeline instead of reproduced Make-A-VideoShow-1 · Inference speedtotal time: 256s -> 178sImproved
use Show-1 latent final super-res instead of Make-A-Video pixel final super-resShow-1 · Inference memoryfinal SR memory: 54GB -> 15GBImproved

Later papers that compared with this method

5 later accepted papers report Show-1 as a baseline in their results tables; 5 of them report a better number on most of the results they share with it.

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

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

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy