Consistency Models

By Yang Song, Prafulla Dhariwal, Mark Chen, Ilya Sutskever. 2023

2,043 citations recorded by the source corpus.

Diffusion models have significantly advanced the fields of image, audio, and video generation, but they depend on an iterative sampling process that causes slow generation. To overcome this limitation, we propose consistency models, a new family of models that generate high quality samples by directly mapping noise to data. They support fast one-step generation by design, while still allowing multistep sampling to trade compute for sample quality. They also support zero-shot data editing, such as image inpainting, colorization, and super-resolution, without requiring explicit training on these tasks. Consistency models can be trained either by distilling pre-trained diffusion models, or as standalone generative models altogether. Through extensive experiments, we demonstrate that they outperform existing distillation techniques for diffusion models in one- and few-step sampling, achieving the new state-of-the-art FID of 3.55 on CIFAR-10 and 6.20 on ImageNet 64x64 for one-step generation. When trained in isolation, consistency models become a new family of generative models that can outperform existing one-step, non-adversarial generative models on standard benchmarks such as CIFAR-10, ImageNet 64x64 and LSUN 256x256.

Subfields: Computer Networks and Communications, Information Systems

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Networks and Communications

In Information Systems

Citations in the corpus

Latent Consistency Models: Synthesizing High-Resolution Images with Few-Step Inference (2023, methodology): “LCM aims to predict the solution of the PF-ODE by minimizing the consistency distillation loss (Song et al., 2023): Here, ˆ z Ψ t n is an estimation of the evolution of the PF-ODE from t n +1 → t n using ODE solver Ψ : where the solver Ψ( · , · , · , · ) is used to approximate the integration from…”
Fast High-Resolution Image Synthesis with Latent Adversarial Diffusion Distillation (2024, positive): “Consistency Distillation [58] is another recent and popular approach for distillation.”
Improved Techniques for Training Consistency Models (2023, methodology): “By choosing the appropriate weighting function, noise embedding layers, and dropout, we significantly improve the sample quality of consistency models using the squared ℓ 2 metric, closing the gap with the original CT in Song et al. (2023) that relies on LPIPS (see Fig.”
DeepCache: Accelerating Diffusion Models for Free (2023, methodology): “The Consistency Model [64] converts random noise to the initial images with only one model evaluation.”
DiffusionDet: Diffusion Model for Object Detection (2022, methodology): “For instance, the most recent consistency models [85] have proposed a fast one-step generation method for the diffusion model.”
InstaFlow: One Step is Enough for High-Quality Diffusion-Based Text-to-Image Generation (2023, methodology): “Consistency models [22] are a new family of generative models that naturally operate in a one-step manner, but their performance on large-scale text-to-image generation is still unclear.”
SinSR: Diffusion-Based Image Super-Resolution in a Single Step (2023, neutral): “Several algorithms are proposed for general diffusion models [22, 28, 37, 38] (a) The inference of SR3 [33] starts from a pure noise, which requires a large number of inference steps (T= 100 after using DDIM [37]).”
Consistency Trajectory Models: Learning Probability Flow ODE Trajectory of Diffusion (2023, methodology): “Recent developments have introduced various consistency concepts, showing their ability to enhance sample quality (Daras et al., 2023; Li et al., 2023), accelerate sampling speed (Song et al., 2023), and improve density estimation in diffusion modeling (Lai et al., 2023a).”
Vista: A Generalizable Driving World Model with High Fidelity and Versatile Controllability (2024, neutral): “Potential solutions may include faster sampling techniques [89, 146] and training-based distillations [91, 93, 101, 103, 110, 124].”
Adversarial Diffusion Distillation (2023, methodology): “Consistency models [66] address the latter issue by enforcing a consistency regularization on the ODE trajectory and demonstrate strong performance for pixel-based models in the few-shot setting.”
Improved Distribution Matching Distillation for Fast Image Synthesis (2024, methodology): “During inference, at each step, we alternate between denoising and noise injection steps, following the consistency model [9], to improve sample quality.”
Seed-TTS: A Family of High-Quality Versatile Speech Generation Models (2024, methodology): “Secondly, we employ consistency distillation [Song et al., 2023] and a modified flow matching algorithm Esser et al. [2024] to reduce the computation cost of the diffusion model.”
DreamLLM: Synergistic Multimodal Comprehension and Creation (2023, methodology): “To enhance inference efficiency, potential strategies could include the adoption of Consistency Models (Song et al., 2023), or the implementation of model compression techniques such as quantization (Yao et al.”
Mean Flows for One-step Generative Modeling (2025, neutral): “While related to previous one-step generative models [46, 43, 14, 31, 49, 23, 12, 52], our method provides a more principled framework.”
SiT: Exploring Flow and Diffusion-based Generative Models with Scalable Interpolant Transformers (2024, result): “We also note that our models are compatible with other samplers [36, 44] specifically tuned to diffusion models as well as sampling distillation [54, 63].”
Retrieval-Augmented Generation for AI-Generated Content: A Survey (2024, neutral): “In this section, we introduce RAG foundations and outline enhancement methods that further improve the effectiveness.”