Automated LoRA Blending for WanVideo 2.2 This node automatically blends your High and Low LoRAs during generation. It calculates precise per-step weights based on your chosen strategy and passes them to WanVideoSetLoRAs. 1. Define Strategy: Choose linear, sigmoid, etc., to define how you want to transition from Low→High. 2. Visual Feedback: The node draws the curve so you can see exactly how the blend will happen. 3. Execution: The node injects a custom weight schedule into the LoRA objects. 4. Result: WanVideoSetLoRAs applies these changing weights at every step of generation. - lorahigh: High-noise LoRA (Creative phase) - loralow: Low-noise LoRA (Refinement phase) - steps: Total generation steps (MUST match your Sampler) - start/endstep: Range where blending happens - blendstrategy: Shape of the blend curve Simply connect the outputs to WanVideoSetLoRAs: If you also use a Speed LoRA (constant), you must chain the nodes in series. Do NOT split the model wire. Correct Wiring: ↓ ↓ (Output Model) ↓ (Output Model) The sampler only needs one input because the model passes through both nodes sequentially. - Green "ACTIVE BLEND": Auto-blending is enabled and ready. - Orange "PREVIEW MODE": Visualization only (no execution). Check the console to see the exact per-step weights being applied.
torch.compile settings, when connected to the model loader, torch.compile of the selected layers is attempted. Requires Triton and torch > 2.7.0 is recommended