Wan Workbench
Category

The Wan Bench

Structured Wan2.2 video generation experiments with fixed variables, controlled comparisons, seeds, resolutions, sampling settings, and output evidence.

2 articles

Page 1 of 1 | Total 2 articles

My Reproducible Wan2.2 Baseline: Versions, Seeds, Nodes, and Settings
The Wan Bench |

My Reproducible Wan2.2 Baseline: Versions, Seeds, Nodes, and Settings

This post establishes the first reproducible Wan2.2 baseline workflow on consumer hardware like the RTX 3090. By locking down explicit versions, seed policies, core ComfyUI node inventories, and resolution settings, it provides a reliable, fixed reference point for future controlled experiments and systematic video generation comparisons.

5
Wan2.2 at 480p vs. 720p: Quality, Speed, and Failure Rate
The Wan Bench |

Wan2.2 at 480p vs. 720p: Quality, Speed, and Failure Rate

This post compares Wan2.2 performance at 480p and 720p using a controlled baseline on a 24GB RTX 3090. By isolating resolution as the sole variable, it evaluates trade-offs in spatial detail, wall-clock render times, peak VRAM pressure, and temporal failure rates, proving that 720p increases visibility of artifacts rather than fixing them.

2