Analysis of Classifier-Free Guidance Weight Schedulers

Classifier-Free Guidance (CFG) enhances the quality and condition adherence of text-toimage diffusion models. It operates by combining the conditional and unconditional predictions using a fixed weight. However, recent works vary the weights throughout the diffusion process, reporting superior resul...

Full description

Saved in:
Bibliographic Details
Published in:Transactions on Machine Learning Research Journal
Main Authors: Wang, Xi, Dufour, Nicolas, Andreou, Nefeli, Cani, Marie-Paule, Fernández Abrevaya, Victoria, Picard, David, Kalogeiton, Vicky
Format: Journal Article
Language:English
Published: [Amherst Massachusetts]: OpenReview.net, 2022 2024
Subjects:
Artificial Intelligence
Computer Science
Generative Model
Diffusion Model
ISSN:2835-8856
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Abstract Classifier-Free Guidance (CFG) enhances the quality and condition adherence of text-toimage diffusion models. It operates by combining the conditional and unconditional predictions using a fixed weight. However, recent works vary the weights throughout the diffusion process, reporting superior results but without providing any rationale or analysis. By conducting comprehensive experiments, this paper provides insights into CFG weight schedulers. Our findings suggest that simple, monotonically increasing weight schedulers consistently lead to improved performances, requiring merely a single line of code. In addition, more complex parametrized schedulers can be optimized for further improvement, but do not generalize across different models and tasks.
AbstractList Classifier-Free Guidance (CFG) enhances the quality and condition adherence of text-toimage diffusion models. It operates by combining the conditional and unconditional predictions using a fixed weight. However, recent works vary the weights throughout the diffusion process, reporting superior results but without providing any rationale or analysis. By conducting comprehensive experiments, this paper provides insights into CFG weight schedulers. Our findings suggest that simple, monotonically increasing weight schedulers consistently lead to improved performances, requiring merely a single line of code. In addition, more complex parametrized schedulers can be optimized for further improvement, but do not generalize across different models and tasks.
Author Dufour, Nicolas
Fernández Abrevaya, Victoria
Cani, Marie-Paule
Kalogeiton, Vicky
Picard, David
Andreou, Nefeli
Wang, Xi
Author_xml – sequence: 1
  givenname: Xi
  surname: Wang
  fullname: Wang, Xi
  organization: Laboratoire d'informatique de l'École polytechnique [Palaiseau]
– sequence: 2
  givenname: Nicolas
  orcidid: 0000-0002-1903-5110
  surname: Dufour
  fullname: Dufour, Nicolas
  organization: Laboratoire d'informatique de l'École polytechnique [Palaiseau]
– sequence: 3
  givenname: Nefeli
  surname: Andreou
  fullname: Andreou, Nefeli
  organization: University of Cyprus [Nicosia]
– sequence: 4
  givenname: Marie-Paule
  orcidid: 0000-0001-7752-9031
  surname: Cani
  fullname: Cani, Marie-Paule
  organization: Laboratoire d'informatique de l'École polytechnique [Palaiseau]
– sequence: 5
  givenname: Victoria
  orcidid: 0000-0002-9829-4929
  surname: Fernández Abrevaya
  fullname: Fernández Abrevaya, Victoria
  organization: Max Planck Institute for Intelligent Systems [Tübingen]
– sequence: 6
  givenname: David
  orcidid: 0000-0002-6296-4222
  surname: Picard
  fullname: Picard, David
  organization: Laboratoire d'Informatique Gaspard-Monge
– sequence: 7
  givenname: Vicky
  orcidid: 0000-0002-7368-6993
  surname: Kalogeiton
  fullname: Kalogeiton, Vicky
  organization: Laboratoire d'informatique de l'École polytechnique [Palaiseau]
BackLink https://hal.science/hal-04823146$$DView record in HAL
BookMark eNotjk9LwzAYh3OY4Jz7AN569dD6Jnnz71iK24SCBxW9laRNbKC20myDfXsrenrg98CP54asxmn0hNxRKFALAQ92_ojngiFgQTkgrMiaaS5yrYW8JtuUogMBihtEuiayHO1wSTFlU8iqwS46RD_nu9n7bH-KnR1bn737-Nkfs5e2991p8HO6JVfBDslv_7khb7vH1-qQ18_7p6qsc0uZPuZOOApCUYRgtFXYdoxLaZToeAugjALWhc4xtMZZlM65JVt51MH4pTLwDbn_--3t0HzP8cvOl2aysTmUdfO7AWrGKcoz5T-R7Eqa
ContentType Journal Article
Copyright Distributed under a Creative Commons Attribution 4.0 International License
Copyright_xml – notice: Distributed under a Creative Commons Attribution 4.0 International License
DBID 1XC
VOOES
DOI 10.48550/arXiv.2404.13040
DatabaseName Hyper Article en Ligne (HAL)
Hyper Article en Ligne (HAL) (Open Access)
DatabaseTitleList
DeliveryMethod fulltext_linktorsrc
Discipline Computer Science
ExternalDocumentID oai:HAL:hal-04823146v1
GroupedDBID 1XC
ALMA_UNASSIGNED_HOLDINGS
GROUPED_DOAJ
M~E
VOOES
ID FETCH-LOGICAL-a128t-b5b1057140f98a74cd2366975d3c0079702dfdb24a9ba46bbb4047e48f9eb05f3
ISSN 2835-8856
IngestDate Sun Oct 19 06:21:50 EDT 2025
IsDoiOpenAccess true
IsOpenAccess true
IsPeerReviewed true
IsScholarly true
Keywords Generative Model
Diffusion Model
Language English
License Distributed under a Creative Commons Attribution 4.0 International License: http://creativecommons.org/licenses/by/4.0
LinkModel OpenURL
MergedId FETCHMERGED-LOGICAL-a128t-b5b1057140f98a74cd2366975d3c0079702dfdb24a9ba46bbb4047e48f9eb05f3
ORCID 0000-0002-7368-6993
0000-0001-7752-9031
0000-0002-1903-5110
0000-0002-6296-4222
0000-0002-9829-4929
OpenAccessLink http://dx.doi.org/10.48550/arXiv.2404.13040
ParticipantIDs hal_primary_oai_HAL_hal_04823146v1
PublicationCentury 2000
PublicationDate 2024-00-00
PublicationDateYYYYMMDD 2024-01-01
PublicationDate_xml – year: 2024
  text: 2024-00-00
PublicationDecade 2020
PublicationTitle Transactions on Machine Learning Research Journal
PublicationYear 2024
Publisher [Amherst Massachusetts]: OpenReview.net, 2022
Publisher_xml – name: [Amherst Massachusetts]: OpenReview.net, 2022
SSID ssib050739441
Score 2.3289576
SecondaryResourceType preprint
Snippet Classifier-Free Guidance (CFG) enhances the quality and condition adherence of text-toimage diffusion models. It operates by combining the conditional and...
SourceID hal
SourceType Open Access Repository
SubjectTerms Artificial Intelligence
Computer Science
Title Analysis of Classifier-Free Guidance Weight Schedulers
URI https://hal.science/hal-04823146
hasFullText 1
inHoldings 1
isFullTextHit
isPrint
journalDatabaseRights – providerCode: PRVAON
  databaseName: DOAJ Directory of Open Access Journals
  issn: 2835-8856
  databaseCode: DOA
  dateStart: 20220101
  customDbUrl:
  isFulltext: true
  dateEnd: 99991231
  titleUrlDefault: https://www.doaj.org/
  omitProxy: false
  ssIdentifier: ssib050739441
  providerName: Directory of Open Access Journals
– providerCode: PRVHPJ
  databaseName: ROAD: Directory of Open Access Scholarly Resources
  issn: 2835-8856
  databaseCode: M~E
  dateStart: 20220101
  customDbUrl:
  isFulltext: true
  dateEnd: 99991231
  titleUrlDefault: https://road.issn.org
  omitProxy: false
  ssIdentifier: ssib050739441
  providerName: ISSN International Centre
link http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwtV1Lj9MwELaqhQMX3ojnKkLs0dA6Thwfy-5WRZRqJYroLbITW41U0lWbVnviF_CjmbHrpNvTcuBiRY4T2ZmJ_Xlm_A0hHxI9YEykmhqbMMoznlEdg0DiwsYiLqTtF-6g8ERMp9l8Lq96vT_hLMxuKeo6u7mR1_9V1FAHwsajs_8g7valUAHXIHQoQexQ3knwhzQjLuVlZWHpo6O1QYWoSndG4KeziCIHJyw1y30UfACpsy6HuHMmfHMBlyZwsXbhesGZ3Vnl_cwxrzp8bKFJq3KqRfAukHK1dXeMNcv2iXOfZAoPEVWGYuCiObRMsM4meZZ8Hv5y8BUab6DDi-3GNM3mLLlAOweGynjPB5pf8JXw9IGREyngaJZ5wvHjKd4xsOECtp5Xu4-ARzgmtPaUT7fptMfD7_nVxSiffJl-vX33IAZxPJxAuVBLClMZYF2e7mAvfY8J2GhhIOjvyzA7JejT5C4VattB7yZ3Xfp03CEAK4tgnHdgZfaYPNyLJRp67XhCeqZ-Sh6FDB7RfkJ_RtKgLNHKRkfKEgVlibyyRJ2yPCc_Rpez8zHdZ9KgCvBHQzX8kgDMYTNtZaYEL0oWp6kUSRkXABKl6LPSlppxJbXiqdYahiEMz6w0MHIbvyAn9ao2L0lUDliqrJZCSUxPz_TAlsjwCotDIftq8Iq8h2Hn154rJUf2cvjGOdZ1X_j1XRq9IQ9Qq7y96y05adZb847cL3ZNtVmfOuPJqZPQX2N6X1k
linkProvider ISSN International Centre
openUrl ctx_ver=Z39.88-2004&ctx_enc=info%3Aofi%2Fenc%3AUTF-8&rfr_id=info%3Asid%2Fsummon.serialssolutions.com&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.atitle=Analysis+of+Classifier-Free+Guidance+Weight+Schedulers&rft.jtitle=Transactions+on+Machine+Learning+Research+Journal&rft.au=Wang%2C+Xi&rft.au=Dufour%2C+Nicolas&rft.au=Andreou%2C+Nefeli&rft.au=Cani%2C+Marie-Paule&rft.date=2024&rft.pub=%5BAmherst+Massachusetts%5D%3A+OpenReview.net%2C+2022&rft.issn=2835-8856&rft_id=info:doi/10.48550%2FarXiv.2404.13040&rft.externalDBID=HAS_PDF_LINK&rft.externalDocID=oai%3AHAL%3Ahal-04823146v1
thumbnail_l http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/lc.gif&issn=2835-8856&client=summon
thumbnail_m http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/mc.gif&issn=2835-8856&client=summon
thumbnail_s http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/sc.gif&issn=2835-8856&client=summon