Multi Lora Weight Composition And Switching

via “inference with multi-lora application and dynamic weight scheduling”

Using Low-rank adaptation to quickly fine-tune diffusion models.

Unique: Implements per-step and per-layer LoRA weight scheduling during inference, enabling dynamic concept influence across diffusion timesteps. Caches composed weights to avoid redundant computation while supporting real-time weight adjustment.

vs others: Enables fine-grained control over concept interaction during generation (unlike static composition) while maintaining inference efficiency through weight caching; supports temporal concept evolution.