Schedulers

Discover the various schedulers available for image generation.

What schedulers do

Schedulers are algorithms that control the progression of the denoising process in generative models. When generating an image, the model starts with random noise and progressively refines this noise into a coherent image. The scheduler dictates how the noise reduction happens over each iteration, impacting the quality, speed, and style of the generated image.

Supported schedulers

Although the scheduler parameter is optional (the model's own scheduler is used by default), you can specify a different scheduler from the following table.

Scheduler	Friendly name	Description
`Default`	Depends on model	The default scheduler used by the Stable Diffusion model.
`DDIMScheduler`	DDIM	Accelerates denoising with fewer steps, maintaining high quality.
`DDIMInverseScheduler`	DDIM Inverse	Inverse version of DDIM for backward processes.
`DDPMScheduler`	DDPM	Standard scheduler following the original diffusion process.
`DEISMultistepScheduler`	DEIS Multistep	Efficient denoising with multiple steps for improved quality.
`DPMSolverSinglestepScheduler`	DPM-Solver Single-step	Single-step solver for direct denoising efficiency.
`DPMSolverMultistepScheduler`	DPM-Solver Multi-step	Multi-step solver enhancing denoising accuracy and stability.
`DPMSolverMultistepInverse`	DPM-Solver Multi-step Inverse	Inverse multi-step solver for reverse processes.
`DPMSolverSDEScheduler`	DPM-Solver SDE	Combines solver techniques with SDE for efficient denoising.
`HeunDiscreteScheduler`	Heun	Uses Heun's method for precise and efficient denoising.
`KDPM2DiscreteScheduler`	KDPM2	Second-order scheduler for refined noise reduction.
`KDPM2AncestralDiscreteScheduler`	KDPM2 Ancestral	Second-order scheduler incorporating ancestral sampling.
`KarrasVeScheduler`	Karras VE	Variance-exploding scheduler by Karras for quality improvement.
`LMSDiscreteScheduler`	LMS	Linear multistep method for balanced speed and quality.
`PNDMScheduler`	PNDM	Combines numerical methods with diffusion for effective denoising.
`ScoreSdeVeScheduler`	Score SDE VE	Score-based variance-exploding scheduler for quality images.
`ScoreSdeVpScheduler`	Score SDE VP	Score-based variance-preserving scheduler for stable results.
`IPNDMScheduler`	IPNDM	Improved pseudo numerical methods for efficient diffusion.
`EulerDiscreteScheduler`	Euler	Utilizes Euler's method for simple and effective denoising.
`EulerAncestralDiscreteScheduler`	Euler Ancestral	Euler's method with ancestral sampling for better quality.
`EDMEulerScheduler`	EDM Euler	Enhanced denoising with modifications to the Euler method.
`FlowMatchEulerDiscreteScheduler`	FlowMatch Euler	Flow matching techniques combined with Euler's method for precision.
`VQDiffusionScheduler`	VQ Diffusion	Vector quantization-based scheduler for high-fidelity images.
`UniPCMultistepScheduler`	UniPC Multistep	Universal multi-step scheduler for diverse applications.
`RePaintScheduler`	RePaint	Focuses on repainting and refining image details during denoising.
`DPM++ 2M Karras`	DPM++ 2M Karras	Advanced Karras scheduler with second-order method.
`DPM++ 2M SDE Karras`	DPM++ 2M SDE Karras	Combines second-order and SDE techniques for superior denoising.
`DPM++ 2M SDE`	DPM++ 2M SDE	Second-order scheduler with SDE for robust image generation.
`DPM++ SDE Karras`	DPM++ SDE Karras	Karras method integrated with SDE for optimal denoising.
`DPM++ SDE`	DPM++ SDE	Stochastic differential equations-based scheduler for high-quality images.
`LCMScheduler`	LCM	Low-complexity scheduler designed for efficient denoising with minimal computational resources.
`EDMDPMSolverMultistepScheduler`	EDM DPM-Solver Multi-step	EDM-enhanced multi-step solver for precision and quality.
`TCDScheduler`	TCD	Time-continuous diffusion scheduler for smooth denoising processes.