Optimized WAN 2.1 Image-to-Video (I2V) 720P workflow using SageAttention and TeaCache and Torch.compile for significantly improved speed and efficiency.
Benchmark Results (RTX 4090 | 64 Frames | 16 FPS | 20 Steps)
β±οΈ Base: 1002s
β‘ Optimized (SageAttention + TeaCache + Torch.compile): 516s
π₯ Twice as fast!
β¨ Key Features:
β
Drastically Faster Inference β Cut generation time in half!
β
SageAttention Optimization β More efficient attention mechanisms
β
TeaCache Acceleration β Reduces redundant computations
β Torch Compile β Compiler optimizations
Workflow file attached!
Try it instantly with no setup:
π Run the workflow here