Filters of Identity: AR Beauty and the Algorithmic Politics of the Digital Body

Corinna Canali; Miriam Doh; Nuria Oliver

REVIEW 3 major objections 5 minor 17 references

Filters of Identity: AR Beauty and the Algorithmic Politics of the Digital Body

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read AR beauty filters are technologies of governance that enforce narrow, racialized and gendered beauty standards, not neutral tools.

desk verdict A coherent position paper that repackages the authors' prior empirical work in a governmentality frame; the governing idea is worth debating but the key statistics are not exposed and the ableist claim outruns the evidence. read the letter →

arxiv 2506.19611 v1 pith:HORXBFB2 submitted 2025-06-24 cs.HC

Miriam Doh , Corinna Canali , Nuria Oliver This is my paper

classification cs.HC

keywords ARbeautyfiltersalgorithmicgovernancebodypoliticsbiasdigitalembodimentpostfeminismself-surveillanceplatformmoderation

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This position paper argues that augmented-reality (AR) beauty filters are not neutral tools of self-expression but technologies of governance that quietly impose racialized, gendered, and ableist beauty standards. Drawing on the authors' earlier analyses of TikTok filters and on leaked platform moderation documents, it traces three channels through which the imposition happens: filter naming conventions, algorithmically biased facial modifications, and content-moderation policies that decide which bodies are visible. The paper's stake is that these filters shape digital desirability and exclusion, not just appearance. It closes by proposing transparency tools and user-driven customization as correctives, while insisting that transparency alone is not enough.

What carries the argument

The argument is carried by the concept of governmentality applied to the digital body: beauty filters govern through naming, algorithmic modification, and platform curation rather than overt coercion, with the feminized body treated as a site of aesthetic and glamour labour. Three named mechanisms do the work: filter naming conventions that encode desirability; algorithmic facial recognition and transformation that assume binary gender and Eurocentric norms; and content-moderation policies that suppress content deemed 'ugly, poor, or disabled'. The Disclaimer Block is the proposed counter-mechanism, a visualisation that exposes the pixel-level, semantic, and intensity changes a filter applies so users can see the modification they would otherwise absorb unnoticed.

What would settle it

An independent audit that applies a representative sample of AR beauty filters, including Bold Glamour, to a diverse face database and measures pixel-level and semantic transformations by gender and skin tone would settle the claim: if transformations do not systematically converge on Western, cisnormative ideals, or if users retain control over them, the governance thesis would not be supported.

Watch

Extended reading notes

Core claim

The central claim is that AR beauty filters function as an infrastructure of algorithmic governance: they encode prescriptive beauty norms into software, obscure those interventions through hyperrealistic rendering, and align with platform moderation to produce a stratified landscape of digital desirability. The case of TikTok's Bold Glamour is offered as the demonstration: it infers binary gender, modifies facial features toward lighter skin, fuller lips, smaller noses, and larger eyes, misclassifies certain faces at higher rates, and runs in tension with the platform's own stated policy against unattainable beauty ideals. The paper argues that what looks like user choice is actually a form of governmentality in which self-optimization is experienced as empowerment.

Load-bearing premise

The entire governance claim rests on the authors' prior empirical analyses of TikTok's filter ecosystem and on leaked moderation documents; if those analyses are wrong or those documents are not representative of platform practice, the claim loses its evidence.

Editorial extensions

If this is right

Platform moderation policies and beauty filters work together to decide which bodies are visible and desirable, so fixing one without the other will not achieve diversity.
Transparency tools like the Disclaimer Block would let users see the algorithmic transformations being applied, but the paper says transparency alone is insufficient.
Changes such as manual intensity controls, disabling automatic gender classification, and diversity presets would shift filters from enforcement toward user-driven representation.
The gap between policies like Effect House's ban on unattainable beauty ideals and the platforms' own promoted filters undermines stated diversity commitments.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Beyond the paper: if beauty filters are governance technologies, their body-image effects should be measurable in controlled longitudinal studies, especially comparing users who see a Disclaimer Block with those who do not.
Beyond the paper: the same logic extends to fully AI-generated faces and synthetic beauty presets, implying that transparency and regulation would need to cover generated content, not only filters applied to real faces.
Beyond the paper: the mechanism suggests a concrete design experiment—offering manual intensity controls and disabling automatic gender classification—and testing whether these changes reduce self-objectification and broaden perceived acceptability of diverse faces.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Quick take: this is a workshop position paper, not an empirical contribution. It builds a coherent governmentality framing around AR beauty filters, but the headline numbers that support the thesis are borrowed from the authors' own papers and are not reproducible from this text. The stronger version of the same argument is already in their FAccT paper; this piece is the critical-theory wrapper.

What's new is the synthesis. Pulling together naming conventions, algorithmic bias, and content moderation into a single story about platform governance is a legitimate move, and the paper does it clearly. It engages the workshop questions directly, and the proposed interventions (manual intensity controls, disabling inferred gender classification, diversity presets) are concrete and reasonable.

The soft spots are where the evidence lives. The footnote claiming the 'Prettiest' filter had 62.2M posts and 98.66% female users gives no source, sample, or method. The Bold Glamour claims about misclassification of Black women and alignment with Western standards are presented as findings but point back to a self-cited paper without the underlying data. For a position paper this might be acceptable if the authors framed them as 'as we reported elsewhere,' but the current text asserts them more strongly. The ableist component is the weakest: the only cited evidence is the leaked moderation documents about suppressing content, not about what filters do. That conflation should be fixed.

The paper is honest about being a position paper and asks good questions. I don't see a circularity problem because it isn't fitting parameters; it's synthesizing. But the empirical basis needs to be either exposed or explicitly deferred to the prior publications with full citations.

This is worth discussing in a reading group about body politics and algorithmic fairness. I'd send it to review, not desk reject, but with the expectation that the authors either open the data or soften the claims. Cite the FAccT paper instead of this one for the empirical results.

Referee Report

3 major / 5 minor

Summary. This position paper argues that AR beauty filters are not neutral self-expression tools but "technologies of governance" that reinforce racialized, gendered, and ableist beauty standards. It synthesizes the authors' prior empirical studies of TikTok filters (Doh et al. 2024; Doh et al. 2025a), leaked platform moderation documents (Biddle et al. 2020; Hern 2020), and feminist and postfeminist theory to argue that naming conventions, algorithmic bias, and platform governance jointly construct digital desirability. The paper concludes with a set of transparency and personalization interventions (the Disclaimer Block, manual intensity controls, disabling automatic gender classification, and diversity presets) and positions itself as an invitation for workshop discussion.

Significance. If the underlying empirical claims in the cited prior work hold, the paper is a valuable theoretical synthesis: it bridges body politics in HCI, algorithmic fairness, and platform governance, and it offers concrete design directions that are testable. The paper explicitly builds on published prior work (Doh et al. 2024 in IMX; Doh et al. 2025a in FAccT) and an open-source analysis framework (OpenFilter), which is a strength. However, the central governance thesis is only as strong as its empirical evidence, and the current manuscript does not make that evidence inspectable: key numbers and effect claims are reported without methodology or even a clear pointer to the specific analyses in the cited papers. The ``ableist'' component of the thesis is particularly under-supported. These gaps matter because the paper's headline claim is an empirical generalization about how filters behave, not only a theoretical framing.

major comments (3)

[§1.1 footnote 1 and §1.2] The quantitative load-bearing claims are presented without any methodology: the statement that the "Prettiest" filter had 62.2M posts and 98.66% female users (footnote 1) and the claims that Bold Glamour has "higher misclassification rates for certain racial groups, especially Black women" and "systematically aligns facial features toward Western beauty standards" (§1.2) are asserted as established results. No sample size, sampling frame, demographic annotation protocol, error metric, or citation to the specific section of Doh et al. (2024) or Doh et al. (2025a) is given. Since these numbers are the empirical foundation for the governance thesis, the paper should either summarize the datasets and methodology or explicitly reframe these as claims drawn from the cited prior studies with enough detail (or precise pointers) for a reader to verify them.
[Abstract and §1.1/§1.2] The abstract and conclusion assert that AR beauty filters reinforce "ableist" beauty standards, but the only cited evidence related to disability is the leaked moderation documents (Biddle et al. 2020; Hern 2020), which concern content suppression on the For You feed, not the behavior of AR beauty filters themselves. The paper does not provide any empirical or theoretical link showing how the filters' algorithmic modifications encode ableist norms. Either add supporting evidence and analysis, or revise the claim to match what the cited sources actually support, for example by saying that beauty filters reinforce normative or exclusionary standards that may include ableist outcomes.
[§1.3 and Fig. 1] The text states that "Our previous work Doh et al. (2024) introduced the Disclaimer Block, Fig. 1," but the figure caption attributes the entire analysis, including the Disclaimer Block visualizations (PD, SD, TD), to Doh et al. (2025a). This creates a factual ambiguity about which prior work introduced the Disclaimer Block and which produced the Bold Glamour analysis. Please clarify the attribution and ensure that the figure credit and the reference list are consistent.

minor comments (5)

[Header] The running header on page 1 reads "AR BEAUTY AND THEALGORITHMIC POLITICS OF THEDIGITALBODY" with missing spaces; this should be fixed to "AR Beauty and the Algorithmic Politics of the Digital Body."
[References] The reference for "Leeat et al. (2019)" is incomplete and inconsistently formatted (e.g., "R. Z. Leeat" and the trailing description of the Harvard Kennedy School Gender Action Portal); please convert it to a standard citation format with accurate author names and publication details.
[§1.1 footnote 1] The footnote reporting the "Prettiest" filter statistics does not cite any previous work, even though the surrounding sentence references Doh et al. (2024). Please add an explicit citation so the reader can locate the source of the statistic.
[§1.1 and §1.2] The name "Bold Glamour" is sometimes italicized and sometimes not; please use consistent formatting throughout the paper.
[§1.1] The phrase "For Youfeed" should be "For You feed" (with a space), and the platform name "TikTok" should be consistently capitalized as "TikTok."

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity; the position paper's claims rest on prior empirical work but do not reduce to their inputs by construction.

full rationale

This is a position paper that synthesizes the authors' prior empirical studies (Doh et al. 2024, 2025a) and external reporting (Biddle et al. 2020; Hern 2020) to argue that AR beauty filters function as technologies of governance. The central claim is an interpretive claim supported by evidence, not a derived prediction. No equations or fitted parameters appear; no quantity is defined in terms of another and then 'predicted'; no uniqueness theorem is invoked; no ansatz is smuggled in via citation. The reliance on self-cited prior work is transparent ('Building on our previous research') and the cited studies are external empirical analyses that are, in principle, reproducible and falsifiable. The lack of methodological detail in the present paper (e.g., the 'Prettiest' usage figures in footnote 1, Bold Glamour misclassification rates in Sec. 1.2) is an evidentiary and reproducibility concern, not a circularity concern under the stated criteria. The 'ableist' component is supported by leaked moderation documents, not by the filter analyses, which is a scope/evidence gap but again not circularity. Therefore, no circular step can be exhibited, and the score reflects only a minor, non-circular self-citation burden.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The central argument assumes the validity of critical-theory frameworks and the accuracy of prior empirical studies. No numeric free parameters are fitted. The proposed Disclaimer Block is a design concept from earlier work, not a newly postulated entity here.

assumptions (4)

domain assumption Objectification theory (Fredrickson and Roberts 1997) applies to AR beauty filters and explains their self-surveillance effects.
Invoked in Sec 1.1 to support the claim that filters amplify internalization of an external gaze, but no empirical test is presented in this paper.
domain assumption Foucault's governmentality framework is a valid lens for analyzing platform-filter interactions.
Used in Sec 1.1.1 to characterize filters as 'conduct of conduct'; this is an interpretive commitment, not a demonstrated fact.
domain assumption The empirical findings of Doh et al. 2025a, Doh et al. 2024, and the whistleblower reports (Biddle 2020; Hern 2020) are accurate and representative.
The core evidence for algorithmic bias and moderation suppression is cited rather than presented; the argument depends on these external sources.
domain assumption TikTok's Effect House policies and their contradiction with practice are accurately described.
Stated in Sec 1.2 without evidence beyond a news reference, yet it supports the criticism of platform governance.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Filters of Identity: AR Beauty and the Algorithmic Politics of the Digital Body." pith.science (2026). https://pith.science/paper/HORXBFB2

@misc{pith2026250619611,
  author       = {Pith},
  title        = {Pith review of: Filters of Identity: AR Beauty and the Algorithmic Politics of the Digital Body},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HORXBFB2}},
  note         = {Machine review of arXiv:2506.19611}
}

read the original abstract

This position paper situates AR beauty filters within the broader debate on Body Politics in HCI. We argue that these filters are not neutral tools but technologies of governance that reinforce racialized, gendered, and ableist beauty standards. Through naming conventions, algorithmic bias, and platform governance, they impose aesthetic norms while concealing their influence. To address these challenges, we advocate for transparency-driven interventions and a critical rethinking of algorithmic aesthetics and digital embodiment.

Figures

Figures reproduced from arXiv: 2506.19611 by the authors.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

17 extracted references · 16 canonical work pages

[1]

Biddle, P

S. Biddle, P. V . Ribeiro, and T. Dias. 2020.Invisible Censorship. The Intercept. https://theintercept.com/ 2020/03/16/tiktok-app-moderators-users-discrimination/[Accessed: Nov 2024]. 4 Filters of Identity: AR Beauty and the Algorithmic Politics of the Digital Body Wendy Brown

work page 2020
[5]

"My Kind of Woman": Analysing Gender Stereotypes in AI through The Averageness Theory and EU Law

"My Kind of Woman": Analysing Gender Stereotypes in AI through The Averageness Theory and EU Law. arXiv:2407.17474 [cs.CY]https://arxiv.org/abs/2407.17474 Ana Elias, Rosalind Gill, and Christina Scharff. 2017.Aesthetic labour: Beauty politics in neoliberalism. Springer. Silvia Federici. 2004.Caliban and the Witch. Autonomedia. Michel Foucault. 2007.Securi...

work page Pith review arXiv 2017
[10]

tried to filter out videos from ugly, poor or disabled users

TikTok “tried to filter out videos from ugly, poor or disabled users”.The Guardian(2020). https: //tinyl.io/BjRa[Accessed: Nov 2024]. Clara Isakowitsch

work page 2020
[11]

InIrish Conference on Artificial Intelligence and Cognitive Science

How augmented reality beauty filters can affect self-perception. InIrish Conference on Artificial Intelligence and Cognitive Science. Springer, 239–250. TERESA DE LAURETIS. 1987.Technologies of Gender: Essays on Theory, Film, and Fiction. Indiana University Press.http://www.jstor.org/stable/j.ctt16gzmbr R. Z. Leeat, N. Shnabel, and P. Glick

work page 1987
[13]

2014.Governmentality

The Chicago Face Database: A Free Stimulus Set of Faces and Norming Data.Behavior Research Methods47 (2015), 1122–1135.https://doi.org/10.3758/s13428-014-0532-5 Ole Jacob Madsen. 2014.Governmentality. Springer New York, New York, NY , 814–816. https://doi.org/10. 1007/978-1-4614-5583-7_126 Angela McRobbie

doi:10.3758/s13428-014-0532-5 2015
[15]

Piera Riccio, Bill Psomas, Francesco Galati, Francisco Escolano, Thomas Hofmann, and Nuria Oliver

Mirror, Mirror on the Wall, Who Is the Whitest of All? Racial Biases in Social Media Beauty Filters.Social Media+ Society10, 2 (2024), 20563051241239295. Piera Riccio, Bill Psomas, Francesco Galati, Francisco Escolano, Thomas Hofmann, and Nuria Oliver

work page 2024
[16]

beauty filters

OpenFilter: a framework to democratize research access to social media AR filters.Advances in Neural Information Processing Systems35 (2022), 12491–12503. A. Ruggeri. 2023.The problems with TikTok’s controversial “beauty filters”. BBC Online. https://tinyl.io/BjRX [Accessed: Nov 2024]. 5 Filters of Identity: AR Beauty and the Algorithmic Politics of the D...

work page 2022
[17]

Elizabeth Wissinger

Brand intimacy, female friendship and digital surveillance networks.New Formations84, 84-85 (2015), 228–245. Elizabeth Wissinger. 2015.This year’s model: Fashion, media, and the making of glamour. NYU Press, New York, USA. Shoshana Zuboff

work page 2015

Show all 17 references

[1997]

Rebecca Fribourg, Etienne Peillard, and Rachel Mcdonnell

Objectification theory: Toward understanding women’s lived experiences and mental health risks.Psychology of women quarterly21, 2 (1997), 173–206. Rebecca Fribourg, Etienne Peillard, and Rachel Mcdonnell

1997
[2004]

Piera Riccio, Julien Colin, Shirley Ogolla, and Nuria Oliver

Post-feminism and popular culture.Feminist media studies4, 3 (2004), 255–264. Piera Riccio, Julien Colin, Shirley Ogolla, and Nuria Oliver

2004
[2015]

Undoing the demos: Neoliberalism’s stealth revolution. (2015). Juraj Cug, Alina T˘anase, Cristian Ionu¸ t Stan, and Tan¸ ta Camelia Chitc˘a

2015
[2017]

Donna Haraway

Governance of and by platforms.SAGE handbook of social media(2017), 254–278. Donna Haraway

2017
[2019]

prescriptive beauty norm

The “prescriptive beauty norm” reflects a desire to enhance gender hierarchy and contributes to social policing of women and employment discrimination practices known as the “beauty tax.”.Journal of Personality and Social Psychology(2019). Available at: Harvard Kennedy School ...

2019
[2020]

The beguiling: Glamour in/as platformed cultural production.Social Media+ Society6, 1 (2020), 2056305119898779. A. Hern

2020
[2021]

In2021 IEEE International Symposium on Mixed and Augmented Reality (ISMAR)

Mirror, mirror on my phone: Investigating dimensions of self-face perception induced by augmented reality filters. In2021 IEEE International Symposium on Mixed and Augmented Reality (ISMAR). IEEE, 470–478. R. Gill. 2007.Gender and the Media. Polity. Tarleton Gillespie

2007
[2022]

Miriam Doh, Corinna Canali, and Anastasia Karagianni

Beauty filters for physical attractiveness: Idealized appearance and imagery, visual content and representations, and negative behaviors and sentiments.Journal of Research in Gender Studies12, 2 (2022), 33–47. Miriam Doh, Corinna Canali, and Anastasia Karagianni

2022
[2024]

InProceedings of the 2024 ACM International Conference on Interactive Media Experiences

Pixels of Perfection and Self-Perception: Deconstructing AR Beauty Filters and Their Challenge to Unbiased Body Image. InProceedings of the 2024 ACM International Conference on Interactive Media Experiences. ACM, Stockholm, Sweden, 349–353. Miriam Doh, Corinna Canali, and Nuri...

2024 arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

Biddle, P

S. Biddle, P. V . Ribeiro, and T. Dias. 2020.Invisible Censorship. The Intercept. https://theintercept.com/ 2020/03/16/tiktok-app-moderators-users-discrimination/[Accessed: Nov 2024]. 4 Filters of Identity: AR Beauty and the Algorithmic Politics of the Digital Body Wendy Brown

work page 2020

[2] [5]

"My Kind of Woman": Analysing Gender Stereotypes in AI through The Averageness Theory and EU Law

"My Kind of Woman": Analysing Gender Stereotypes in AI through The Averageness Theory and EU Law. arXiv:2407.17474 [cs.CY]https://arxiv.org/abs/2407.17474 Ana Elias, Rosalind Gill, and Christina Scharff. 2017.Aesthetic labour: Beauty politics in neoliberalism. Springer. Silvia Federici. 2004.Caliban and the Witch. Autonomedia. Michel Foucault. 2007.Securi...

work page Pith review arXiv 2017

[3] [10]

tried to filter out videos from ugly, poor or disabled users

TikTok “tried to filter out videos from ugly, poor or disabled users”.The Guardian(2020). https: //tinyl.io/BjRa[Accessed: Nov 2024]. Clara Isakowitsch

work page 2020

[4] [11]

InIrish Conference on Artificial Intelligence and Cognitive Science

How augmented reality beauty filters can affect self-perception. InIrish Conference on Artificial Intelligence and Cognitive Science. Springer, 239–250. TERESA DE LAURETIS. 1987.Technologies of Gender: Essays on Theory, Film, and Fiction. Indiana University Press.http://www.jstor.org/stable/j.ctt16gzmbr R. Z. Leeat, N. Shnabel, and P. Glick

work page 1987

[5] [13]

2014.Governmentality

The Chicago Face Database: A Free Stimulus Set of Faces and Norming Data.Behavior Research Methods47 (2015), 1122–1135.https://doi.org/10.3758/s13428-014-0532-5 Ole Jacob Madsen. 2014.Governmentality. Springer New York, New York, NY , 814–816. https://doi.org/10. 1007/978-1-4614-5583-7_126 Angela McRobbie

doi:10.3758/s13428-014-0532-5 2015

[6] [15]

Piera Riccio, Bill Psomas, Francesco Galati, Francisco Escolano, Thomas Hofmann, and Nuria Oliver

Mirror, Mirror on the Wall, Who Is the Whitest of All? Racial Biases in Social Media Beauty Filters.Social Media+ Society10, 2 (2024), 20563051241239295. Piera Riccio, Bill Psomas, Francesco Galati, Francisco Escolano, Thomas Hofmann, and Nuria Oliver

work page 2024

[7] [16]

beauty filters

OpenFilter: a framework to democratize research access to social media AR filters.Advances in Neural Information Processing Systems35 (2022), 12491–12503. A. Ruggeri. 2023.The problems with TikTok’s controversial “beauty filters”. BBC Online. https://tinyl.io/BjRX [Accessed: Nov 2024]. 5 Filters of Identity: AR Beauty and the Algorithmic Politics of the D...

work page 2022

[8] [17]

Elizabeth Wissinger

Brand intimacy, female friendship and digital surveillance networks.New Formations84, 84-85 (2015), 228–245. Elizabeth Wissinger. 2015.This year’s model: Fashion, media, and the making of glamour. NYU Press, New York, USA. Shoshana Zuboff

work page 2015