DynamicCFGScheduler<T>
Models & Types · AiDotNet.Diffusion.Guidance
Dynamic Classifier-Free Guidance that adjusts scale per timestep.
Implements: IGuidanceMethod<T>
For Beginners
Instead of using the same guidance strength for every step, this uses stronger guidance at the start (to get the big picture right) and lighter guidance at the end (to keep fine details natural).
How It Works
Applies a time-dependent guidance scale that starts high for structural coherence in early (noisy) steps and decreases toward the end for fine detail preservation. This reduces over-saturation artifacts common with high static CFG scales.
Constructors
| Constructor | Summary |
|---|---|
DynamicCFGScheduler(double minScale, double maxScale) |
Initializes a new Dynamic CFG scheduler. |
Properties
| Property | Summary |
|---|---|
GuidanceType GuidanceType { get; } |
Methods
| Method | Summary |
|---|---|
Tensor<T> Apply(Tensor<T> unconditional, Tensor<T> conditional, double scale, double timestep) |