https://github.com/cloneofsimo/consistency_models\n

https://twitter.com/RiversHaveWings/status/1634038603247661062

\n\n

Does anyone know of any other one?

\n","updatedAt":"2023-03-16T11:06:22.699Z","author":{"_id":"624bebf604abc7ebb01789af","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1649143001781-624bebf604abc7ebb01789af.jpeg","fullname":"Apolinário from multimodal AI art","name":"multimodalart","type":"user","isPro":true,"isHf":true,"isHfAdmin":false,"isMod":false,"followerCount":5237,"isUserFollowing":false}},"numEdits":1,"editors":["multimodalart"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1649143001781-624bebf604abc7ebb01789af.jpeg"],"reactions":[{"reaction":"👍","users":["Wauplin","julien-c"],"count":2}],"isReport":false}},{"id":"641ca2a98357e6ae9b1c61c2","author":{"_id":"60d4f63701fc5a5e75a85a5d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674029591255-60d4f63701fc5a5e75a85a5d.jpeg","fullname":"Maina Denis","name":"kinyugo","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":2,"isUserFollowing":false},"createdAt":"2023-03-23T19:04:09.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"I have implemented a library for training consistency models. https://github.com/Kinyugo/consistency_models","html":"

I have implemented a library for training consistency models. https://github.com/Kinyugo/consistency_models

\n","updatedAt":"2023-03-23T19:04:09.299Z","author":{"_id":"60d4f63701fc5a5e75a85a5d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674029591255-60d4f63701fc5a5e75a85a5d.jpeg","fullname":"Maina Denis","name":"kinyugo","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":2,"isUserFollowing":false}},"numEdits":0,"editors":["kinyugo"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1674029591255-60d4f63701fc5a5e75a85a5d.jpeg"],"reactions":[{"reaction":"❤️","users":["julien-c","multimodalart"],"count":2},{"reaction":"👍","users":["admarcosai"],"count":1}],"isReport":false}},{"id":"6436ca59197d3e05c1ca7397","author":{"_id":"624bebf604abc7ebb01789af","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1649143001781-624bebf604abc7ebb01789af.jpeg","fullname":"Apolinário from multimodal AI art","name":"multimodalart","type":"user","isPro":true,"isHf":true,"isHfAdmin":false,"isMod":false,"followerCount":5237,"isUserFollowing":false},"createdAt":"2023-04-12T15:12:25.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"OpenAI open sourced their implementation as well! https://github.com/openai/consistency_models","html":"

OpenAI open sourced their implementation as well! https://github.com/openai/consistency_models

\n","updatedAt":"2023-04-12T15:12:25.250Z","author":{"_id":"624bebf604abc7ebb01789af","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1649143001781-624bebf604abc7ebb01789af.jpeg","fullname":"Apolinário from multimodal AI art","name":"multimodalart","type":"user","isPro":true,"isHf":true,"isHfAdmin":false,"isMod":false,"followerCount":5237,"isUserFollowing":false}},"numEdits":0,"editors":["multimodalart"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1649143001781-624bebf604abc7ebb01789af.jpeg"],"reactions":[],"isReport":false}},{"id":"649d6f872a4ae4c471a4f980","author":{"_id":"5dfcb1aada6d0311fd3d5448","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1584435275418-5dfcb1aada6d0311fd3d5448.jpeg","fullname":"Patrick von Platen","name":"patrickvonplaten","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":810,"isUserFollowing":false},"createdAt":"2023-06-29T11:48:23.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"Model is being merged to Diffusers as well and will be in the next 0.18.0 release! \nhttps://github.com/huggingface/diffusers/pull/3492","html":"

Model is being merged to Diffusers as well and will be in the next 0.18.0 release!
https://github.com/huggingface/diffusers/pull/3492

\n","updatedAt":"2023-06-29T11:48:23.686Z","author":{"_id":"5dfcb1aada6d0311fd3d5448","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1584435275418-5dfcb1aada6d0311fd3d5448.jpeg","fullname":"Patrick von Platen","name":"patrickvonplaten","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":810,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9614415764808655},"editors":["patrickvonplaten"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1584435275418-5dfcb1aada6d0311fd3d5448.jpeg"],"reactions":[],"isReport":false}},{"id":"6665674ed507e6a951560db0","author":{"_id":"6186ddf6a7717cb375090c01","avatarUrl":"/avatars/716b6a7d1094c8036b2a8a7b9063e8aa.svg","fullname":"Julien BLANCHON","name":"blanchon","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":176,"isUserFollowing":false},"createdAt":"2024-06-09T08:26:54.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"# Unleashing Speed: Consistency Models for Fast Generative AI\n\nhttps://cdn-uploads.huggingface.co/production/uploads/6186ddf6a7717cb375090c01/-hCgmyRwM-0eSwukFljnA.mp4 \n\n## Links 🔗:\n👉 Subscribe: https://www.youtube.com/@Arxflix\n👉 Twitter: https://x.com/arxflix\n👉 LMNT (Partner): https://lmnt.com/\n\n\nBy Arxflix\n![9t4iCUHx_400x400-1.jpg](https://cdn-uploads.huggingface.co/production/uploads/6186ddf6a7717cb375090c01/v4S5zBurs0ouGNwYj1GEd.jpeg)","html":"

\n\t\n\t\t\n\t\n\t\n\t\tUnleashing Speed: Consistency Models for Fast Generative AI\n\t\n

\n\n

\n\t\n\t\t\n\t\n\t\n\t\tLinks 🔗:\n\t\n

👉 Subscribe: https://www.youtube.com/@Arxflix
👉 Twitter: https://x.com/arxflix
👉 LMNT (Partner): https://lmnt.com/

By Arxflix
$\"9t4iCUHx_400x400-1.jpg\"$

\n","updatedAt":"2024-06-09T08:26:54.182Z","author":{"_id":"6186ddf6a7717cb375090c01","avatarUrl":"/avatars/716b6a7d1094c8036b2a8a7b9063e8aa.svg","fullname":"Julien BLANCHON","name":"blanchon","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":176,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.5111569166183472},"editors":["blanchon"],"editorAvatarUrls":["/avatars/716b6a7d1094c8036b2a8a7b9063e8aa.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2303.01469","authors":[{"_id":"64122a6e982a341736be4f84","name":"Yang Song","hidden":false},{"_id":"64122a6e982a341736be4f85","name":"Prafulla Dhariwal","hidden":false},{"_id":"64122a6e982a341736be4f86","name":"Mark Chen","hidden":false},{"_id":"64122a6e982a341736be4f87","name":"Ilya Sutskever","hidden":false}],"publishedAt":"2023-03-02T18:30:16.000Z","title":"Consistency Models","summary":"Diffusion models have made significant breakthroughs in image, audio, and\nvideo generation, but they depend on an iterative generation process that\ncauses slow sampling speed and caps their potential for real-time applications.\nTo overcome this limitation, we propose consistency models, a new family of\ngenerative models that achieve high sample quality without adversarial\ntraining. They support fast one-step generation by design, while still allowing\nfor few-step sampling to trade compute for sample quality. They also support\nzero-shot data editing, like image inpainting, colorization, and\nsuper-resolution, without requiring explicit training on these tasks.\nConsistency models can be trained either as a way to distill pre-trained\ndiffusion models, or as standalone generative models. Through extensive\nexperiments, we demonstrate that they outperform existing distillation\ntechniques for diffusion models in one- and few-step generation. For example,\nwe achieve the new state-of-the-art FID of 3.55 on CIFAR-10 and 6.20 on\nImageNet 64x64 for one-step generation. When trained as standalone generative\nmodels, consistency models also outperform single-step, non-adversarial\ngenerative models on standard benchmarks like CIFAR-10, ImageNet 64x64 and LSUN\n256x256.","upvotes":8,"discussionId":"64122a73982a341736be50a3","ai_summary":"Consistency models achieve high sample quality through fast one-step generation, outperforming diffusion model distillation and standalone generative models on benchmarks.","ai_keywords":["diffusion models","consistency models","adversarial training","fast one-step generation","few-step sampling","zero-shot data editing","image inpainting","colorization","super-resolution","FID","CIFAR-10","ImageNet 64x64","LSUN 256x256"]},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"639c3ca2db7c5f350040e4b8","avatarUrl":"/avatars/392bd13bcb1ee0d90e07dd8e8e711c39.svg","isPro":false,"fullname":"Thomas Boyer","user":"thethomasboyer","type":"user"},{"_id":"5f7fbd813e94f16a85448745","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1649681653581-5f7fbd813e94f16a85448745.jpeg","isPro":true,"fullname":"Sayak Paul","user":"sayakpaul","type":"user"},{"_id":"64dc0516a34448aee68a3294","avatarUrl":"/avatars/612d8feb74179870094652e65eef1d9a.svg","isPro":false,"fullname":"loukili m","user":"portero","type":"user"},{"_id":"64bd7456b567ae97c3398076","avatarUrl":"/avatars/f27a1df7a374dd40bad53fb7019828dc.svg","isPro":false,"fullname":"Andy Wang","user":"wangdelp","type":"user"},{"_id":"630b77f68b327c7b8b98c409","avatarUrl":"/avatars/2fe95c9ac95f34dbb031f2ec018f68b0.svg","isPro":false,"fullname":"Minghui Hu","user":"h1t","type":"user"},{"_id":"64b695dcd3df8086e5ed7c89","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64b695dcd3df8086e5ed7c89/ZULwnPE8Q5q7bxKJsJ7v5.jpeg","isPro":false,"fullname":"Adam Fields","user":"adamelliotfields","type":"user"},{"_id":"63107b18e87051f3e3e0f598","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63107b18e87051f3e3e0f598/R9onir4Y0MZuq1jEWCZ2-.jpeg","isPro":false,"fullname":"Unchun Yang","user":"ucyang","type":"user"},{"_id":"659edbd3e6d28c70c08766d4","avatarUrl":"/avatars/805faaf9d79bebe7b981b2081617f688.svg","isPro":false,"fullname":"rubbyninja","user":"rubbyninja","type":"user"}],"acceptLanguages":["*"]}">

Papers

arxiv:2303.01469

Consistency Models

Published on Mar 2, 2023

Upvote

Authors:

Abstract

Consistency models achieve high sample quality through fast one-step generation, outperforming diffusion model distillation and standalone generative models on benchmarks.

AI-generated summary

Diffusion models have made significant breakthroughs in image, audio, and video generation, but they depend on an iterative generation process that causes slow sampling speed and caps their potential for real-time applications. To overcome this limitation, we propose consistency models, a new family of generative models that achieve high sample quality without adversarial training. They support fast one-step generation by design, while still allowing for few-step sampling to trade compute for sample quality. They also support zero-shot data editing, like image inpainting, colorization, and super-resolution, without requiring explicit training on these tasks. Consistency models can be trained either as a way to distill pre-trained diffusion models, or as standalone generative models. Through extensive experiments, we demonstrate that they outperform existing distillation techniques for diffusion models in one- and few-step generation. For example, we achieve 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 as standalone generative models, consistency models also outperform single-step, non-adversarial generative models on standard benchmarks like CIFAR-10, ImageNet 64x64 and LSUN 256x256.