UniPCScheduler<T>

Models & Types · AiDotNet.Diffusion.Schedulers

UniPC (Unified Predictor-Corrector) scheduler for fast, high-quality diffusion sampling.

Inheritance: NoiseSchedulerBase<T> → UniPCScheduler<T>

For Beginners

UniPC is like a “guess and check” approach to denoising.

Most schedulers just predict the next step (predict only):

  • Step 1: Predict the cleaner image → use as-is

UniPC adds a correction step for better accuracy:

  • Step 1: Predict the cleaner image (predictor step)
  • Step 2: Check how good the prediction was and improve it (corrector step)

This two-phase approach means:

  • Better quality at the same number of steps (e.g., 10-step UniPC ≈ 15-step DDIM)
  • Or same quality with fewer steps (faster generation)

Key characteristics:

  • Combines predictor-corrector methodology with multi-step methods
  • Supports orders 1-3 (higher = more accurate per step)
  • Deterministic by default
  • Particularly effective at very low step counts (5-15 steps)

Used by:

  • ComfyUI, A1111, and many Stable Diffusion UIs as an alternative scheduler
  • Effective with SD 1.5, SDXL, and other diffusion models

How It Works

UniPC combines predictor and corrector steps into a unified framework, achieving superior sampling quality with fewer function evaluations compared to pure predictor methods like DDIM or DPM-Solver++.

Reference: Zhao et al., “UniPC: A Unified Predictor-Corrector Framework for Fast Sampling of Diffusion Models”, NeurIPS 2023

Constructors

Constructor Summary
UniPCScheduler(SchedulerConfig<T> config, int solverOrder, bool useCorrectorStep) Initializes a new instance of the UniPC scheduler.

Methods

Method Summary
Dictionary<string, object> GetState()
Vector<T> Step(Vector<T> modelOutput, int timestep, Vector<T> sample, T eta, Vector<T> noise) Performs one UniPC step: the UniC corrector refines the previous predictor output with the new model evaluation, then the UniP predictor advances to the next timestep.
void LoadState(Dictionary<string, object> state)
void SetTimesteps(int inferenceSteps) Sets up the inference timesteps and computes lambda/sigma/alpha schedules.