Consistency Models
📰 OpenAI News
Consistency models are a new family of generative models that enable fast one-step generation and zero-shot data editing, outperforming existing diffusion models and non-adversarial generative models on standard benchmarks
Action Steps
- Understand the limitations of diffusion models and their iterative sampling process
- Explore the concept of consistency models and their ability to directly map noise to data
- Investigate the training methods for consistency models, including distillation of pre-trained diffusion models and standalone training
- Evaluate the performance of consistency models on standard benchmarks, such as CIFAR-10 and ImageNet 64x64
Who Needs to Know This
Researchers and engineers working on generative models, computer vision, and AI can benefit from this new approach to generate high-quality samples and perform data editing tasks, such as image inpainting and super-resolution
Key Insight
💡 Consistency models can generate high-quality samples and perform data editing tasks without requiring explicit training on these tasks
Share This
💡 Consistency models enable fast one-step generation and zero-shot data editing, outperforming existing diffusion models and non-adversarial generative models #AI #GenerativeModels
Key Takeaways
Consistency models are a new family of generative models that enable fast one-step generation and zero-shot data editing, outperforming existing diffusion models and non-adversarial generative models on standard benchmarks
Full Article
# Consistency Models | OpenAI
[Skip to main content](https://openai.com/index/consistency-models#main)
[](https://openai.com/)
* [Research](https://openai.com/research/index/)
* Products
* [Business](https://openai.com/business/)
* [Developers](https://openai.com/api/)
* [Company](https://openai.com/about/)
* [Foundation(opens in a new window)](https://openaifoundation.org/)
[Try ChatGPT(opens in a new window)](https://chatgpt.com/)
* Research
* Products
* Business
* Developers
* Company
* [Foundation(opens in a new window)](https://openaifoundation.org/)
[Try ChatGPT(opens in a new window)](https://chatgpt.com/)
OpenAI
June 20, 2024
[Publication](https://openai.com/research/index/publication/)
# Consistency models
[Read paper (opens in a new window)](https://arxiv.org/abs/2303.01469)
Loading…
Share
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.
* [Generative Models](https://openai.com/research/index/?tags=generative-models)
* [Software & Engineering](https://openai.com/research/index/?tags=software-engineering)
## Authors
Yang Song, Prafulla Dhariwal, Mark Chen, Ilya Sutskever
Our Research
* [Research Index](https://openai.com/research/index/)
* [Research Overview](https://openai.com/research/)
* [Research Residency](https://openai.com/residency/)
* [OpenAI for Science](https://openai.com/science/)
* [Economic Research](https://openai.com/signals/)
Latest Advancements
* [GPT-5.3 Instant](https://openai.com/index/gpt-5-3-instant/)
* [GPT-5.3-Codex](https://openai.com/index/introducing-gpt-5-3-codex/)
* [GPT-5](https://openai.com/gpt-5/)
* [Codex](https://openai.com/index/introducing-gpt-5-3-codex/)
Safety
* [Safety Approach](https://openai.com/safety/)
* [Security & Privacy](https://openai.com/security-and-privacy/)
* [Trust & Transparency](https://openai.com/trust-and-transparency/)
ChatGPT
* [Explore ChatGPT(opens in a new window)](https://chatgpt.com/overview)
* [Business](https://chatgpt.com/business/business-plan)
* [Enterprise](https://chatgpt.com/business/enterprise)
* [Education](https://chatgpt.com/business/education)
* [Pricing(opens in a new window)](https://chatgpt.com/pricing)
* [Download(opens in a new window)](https://chatgpt.com/download)
Sora
* [Sora Overview](https://openai.com/sora/)
* [Features](https://openai.com/sora/#features)
* [Pricing](https://openai.com/sora/#pricing)
* [Sora log in(opens in a new window)](https://sora.com/)
API Platform
* [Platform Overview](https://openai.com/api/)
* [Pricing](https://openai.com/api/pricing/)
* [API log in(opens in a new window)](https://platform.openai.com/login)
* [Documentation(opens in a new window)](https://developers.openai.com/api/docs)
* [Developer Forum(opens in a new window)](https://c
[Skip to main content](https://openai.com/index/consistency-models#main)
[](https://openai.com/)
* [Research](https://openai.com/research/index/)
* Products
* [Business](https://openai.com/business/)
* [Developers](https://openai.com/api/)
* [Company](https://openai.com/about/)
* [Foundation(opens in a new window)](https://openaifoundation.org/)
[Try ChatGPT(opens in a new window)](https://chatgpt.com/)
* Research
* Products
* Business
* Developers
* Company
* [Foundation(opens in a new window)](https://openaifoundation.org/)
[Try ChatGPT(opens in a new window)](https://chatgpt.com/)
OpenAI
June 20, 2024
[Publication](https://openai.com/research/index/publication/)
# Consistency models
[Read paper (opens in a new window)](https://arxiv.org/abs/2303.01469)
Loading…
Share
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.
* [Generative Models](https://openai.com/research/index/?tags=generative-models)
* [Software & Engineering](https://openai.com/research/index/?tags=software-engineering)
## Authors
Yang Song, Prafulla Dhariwal, Mark Chen, Ilya Sutskever
Our Research
* [Research Index](https://openai.com/research/index/)
* [Research Overview](https://openai.com/research/)
* [Research Residency](https://openai.com/residency/)
* [OpenAI for Science](https://openai.com/science/)
* [Economic Research](https://openai.com/signals/)
Latest Advancements
* [GPT-5.3 Instant](https://openai.com/index/gpt-5-3-instant/)
* [GPT-5.3-Codex](https://openai.com/index/introducing-gpt-5-3-codex/)
* [GPT-5](https://openai.com/gpt-5/)
* [Codex](https://openai.com/index/introducing-gpt-5-3-codex/)
Safety
* [Safety Approach](https://openai.com/safety/)
* [Security & Privacy](https://openai.com/security-and-privacy/)
* [Trust & Transparency](https://openai.com/trust-and-transparency/)
ChatGPT
* [Explore ChatGPT(opens in a new window)](https://chatgpt.com/overview)
* [Business](https://chatgpt.com/business/business-plan)
* [Enterprise](https://chatgpt.com/business/enterprise)
* [Education](https://chatgpt.com/business/education)
* [Pricing(opens in a new window)](https://chatgpt.com/pricing)
* [Download(opens in a new window)](https://chatgpt.com/download)
Sora
* [Sora Overview](https://openai.com/sora/)
* [Features](https://openai.com/sora/#features)
* [Pricing](https://openai.com/sora/#pricing)
* [Sora log in(opens in a new window)](https://sora.com/)
API Platform
* [Platform Overview](https://openai.com/api/)
* [Pricing](https://openai.com/api/pricing/)
* [API log in(opens in a new window)](https://platform.openai.com/login)
* [Documentation(opens in a new window)](https://developers.openai.com/api/docs)
* [Developer Forum(opens in a new window)](https://c
DeepCamp AI