Research map / Video / Video generation
WorldCache: Content-Aware Caching for Accelerated Video World Models
Umair Nawaz, Ahmed Heakl, Ufaq Khan, ABDELRAHMAN YOUSSIEF, Salman Khan, Fahad Shahbaz Khan
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 CFC motion-adaptive thresholding (ablation) | Cosmos-Predict2.5-2B caching · PAI-Bench I2W | speedup / overall score: 1.00x -> 1.52x speedup, overall 0.8027 -> 0.8020 | Improved |
| add SWD saliency-weighted drift on top of CFC (ablation) | Cosmos-Predict2.5-2B caching · PAI-Bench I2W | speedup / overall score: 1.52x -> 1.67x speedup, overall 0.8020 -> 0.8003 | Mixed |
| add OFA optimal feature approximation on top of CFC+SWD (ablation) | Cosmos-Predict2.5-2B caching · PAI-Bench I2W | overall score: 0.8003 -> 0.8035 | Improved |
| add OFA on top of CFC+SWD (ablation) | Cosmos-Predict2.5-2B caching · PAI-Bench I2W | speedup: 1.67x -> 1.49x | Hurt |
| add ATS adaptive threshold scheduling on top of CFC+SWD+OFA (ablation) | Cosmos-Predict2.5-2B caching · PAI-Bench I2W | speedup: 1.49x -> 2.30x | Improved |
| add ATS on top of CFC+SWD+OFA (ablation) | Cosmos-Predict2.5-2B caching · PAI-Bench I2W | overall score: 0.8035 -> 0.7977 | Hurt |
| use WorldCache instead of DiCache | Cosmos-Predict2.5-2B · PAI-Bench T2W | speedup / avg score: 1.3x -> 2.1x speedup, avg 0.743 -> 0.745 | Improved |
| use WorldCache instead of DiCache | Cosmos-Predict2.5-14B · PAI-Bench T2W | speedup / avg score: 1.4x -> 2.14x speedup, avg 0.768 -> 0.771 | Improved |
| use WorldCache instead of DiCache | Cosmos-Predict2.5-2B · PAI-Bench I2W | speedup / avg score: 1.4x -> 2.3x speedup, avg 0.794 -> 0.798 | Improved |
| use WorldCache instead of DiCache | WAN2.1-1.3B · PAI-Bench T2W | speedup / overall: 1.96x -> 2.36x speedup, overall 0.7703 -> 0.7721 | Improved |
| use WorldCache instead of DiCache | WAN2.1-14B · PAI-Bench I2W | speedup / overall: 1.53x -> 2.31x speedup, overall 0.7311 -> 0.7388 | Improved |
| use WorldCache instead of DiCache | Cosmos-Predict2.5-2B · EgoDex-Eval I2V | PSNR: 12.63 -> 12.82 PSNR, 1.34x -> 1.62x speedup | Improved |
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.
- FasterCache — FasterCache: Training-Free Video Diffusion Model Acceleration with High Quality (ICLR 2025) · better on 8 of 8 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
- Model Reveals What to Cache: Profiling-Based Feature Reuse for Video Diffusion Models (ICCV 2025 · similarity 0.91)
- FasterCache: Training-Free Video Diffusion Model Acceleration with High Quality (ICLR 2025 · similarity 0.90)
- BWCache: Accelerating Video Diffusion Transformers through Block-Wise Caching (ICLR 2026 · similarity 0.90)
- Timestep Embedding Tells: It's Time to Cache for Video Diffusion Model (CVPR 2025 · similarity 0.89)
- MagCache: Fast Video Generation with Magnitude-Aware Cache (NeurIPS 2025 · similarity 0.89)
- Addressable Memory for Video World Models (NeurIPS 2026 · similarity 0.88)
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.
