FILE 2026 workshop >>> SESC 24 Maio 2026 workshop (upcoming)
Clarissa Ribeiro
TT Tiananmen–Tehran (2026) is a technoetic art installation comprising eight generatively produced, 3D-printed masks that investigate the intersections of state violence, political dissent, algorithmic surveillance, and collective memory. Developed through neural-network training, performative generative synthesis, AI-assisted image interpretation, and digital fabrication, the work explores the emergence of anomalous facial configurations as a speculative inquiry into the politics of computational visibility.
The project originates in the computational entanglement of two photographic archives spanning 37 years: 40 facial images extracted from photographic documentation of protesters associated with the 1989 Tiananmen Square demonstrations and subsequent military crackdown in Beijing, China, and 40 facial images extracted from documentation of the 2025–2026 protests in Iran. The images were collected from the first ten results returned by Google Search for each respective historical context, introducing search-engine ranking and algorithmically mediated visibility as additional dimensions shaping the selection and construction of the generative training dataset.
1989 Tiananmen Square massacre in Beijing,
These 80 source images were processed into animated sequences using the eGIF platform and subsequently exported as MP4 videos. The videos were introduced into Autolume using an exceptionally high frames-per-second (FPS) sampling rate, expanding the number of training frames derived from the original photographic collections. This process of temporal sampling and dataset expansion established the computational foundations for training a generative neural network and, ultimately, producing the eight anomalous facial configurations subsequently translated into sculptural masks. (Note: Documenting the precise FPS setting, the total number of frames extracted, and the final number of training images makes the experiment more reproducible and clarifies the distinction between temporal frame extraction and DiffAugment-based transformations during training.)
Through the synthesis of visual information drawn from these two historical contexts, the work investigates the emergence of morphologically unstable identities that resist straightforward associations with their photographic origins. The resulting masks materialize the transformations occurring as archival images are subjected to successive processes of computational interpretation.
Beyond the photographic material used for training, TT Tiananmen–Tehran engages a broader historical constellation of pro-democracy movements, political protests, and episodes of state repression across mainland China, Hong Kong, and Iran between 1989 and 2026. Eight historical events inform this expanded conceptual framework:
1989 — Tiananmen Square, Beijing, China. The pro-democracy demonstrations and subsequent military crackdown became an enduring reference for debates concerning political memory, censorship, and state violence.
1999 — Student protests, Tehran, Iran. Demonstrations following the closure of the reformist newspaper Salaam raised questions about freedom of expression, political reform, and the repression of student movements.
2009 — Green Movement, Iran. Mass demonstrations following the disputed presidential election foregrounded the relationship between political participation, networked communication, and the circulation of protest imagery.
2011 — Jasmine Revolution-inspired mobilizations, mainland China. Online calls for demonstrations and the subsequent security response revealed the expanding role of digital monitoring, preventive policing, and censorship.
2014 — Umbrella Movement, Hong Kong. Large-scale demonstrations advocating democratic electoral reform transformed urban occupation into a visible expression of collective political agency.
2019 — Anti-extradition protests, Hong Kong. A movement that expanded into broader demands concerning political accountability, civil liberties, and the contested role of surveillance technologies in public demonstrations.
2022 — Woman, Life, Freedom movement, Iran. Protests following the death of Jina Mahsa Amini in custody brought bodily autonomy, gender, political resistance, and the visibility of dissent into an international field of debate.
2025–2026 — Protests and state repression, Iran. Demonstrations beginning amid economic grievances developed into wider expressions of political dissent, accompanied by severe repression and restrictions on communications.
These episodes are approached neither as historically equivalent nor as direct visual sources for the generative model but as establishing a wider panorama of political and technological conditions through which the work’s central concerns can be examined.
Only the photographic collections associated with Tiananmen Square in 1989 and the Iranian protests of 2025–2026 were used to construct the model’s training dataset. The eight historical events do not correspond individually to the eight masks. Their relationship is conceptual, considering that, together, they articulate a field of historical resonances concerning the vulnerability of political bodies, the mediation of collective memory, and the evolving infrastructures through which dissent becomes visible, identifiable, and subject to control.
The generative process began on August 22, 2026, at 11:30 AM, using Autolume, a neural-network-based visual synthesizer. The initial dataset comprised 80 facial images, equally distributed between two photographic collections: 40 images associated with the 1989 Tiananmen Square protests and 40 extracted from images of the 2025–2026 protests in Iran. To expand the visual material available for neural-network training, the two image collections were processed through the eGIF platform, where animated GIF sequences were generated and subsequently exported as MP4 videos.
These sequences were then introduced into Autolume using an exceptionally high frames-per-second (FPS) sampling rate, enabling the extraction of an expanded set of training frames from the original photographic sources. This frame-based dataset expansion increased the number of image samples available to the model, although the resulting frames remained derived from the original 80 photographic sources.
The model was trained locally over eight days on a MacBook Pro equipped with an Apple M2 processor and 8 GB of unified memory, running macOS Tahoe 26.5. The generative process was developed using Autolume, employing the StyleGAN2 neural network architecture. The original photographic dataset comprised 80 facial images, including 40 extracted from documentation of the 1989 Tiananmen Square protests in China and 40 from the 2025–2026 protests in Iran.
To expand the training dataset, the images were processed into animated GIF sequences using the eGIF platform and subsequently exported as MP4 videos. These videos were imported into Autolume, where an exceptionally high frames-per-second (FPS) sampling rate was employed to extract additional frames, increasing the number of image samples available for training.
The model was trained for eight days, beginning on August 22, 2026, at 11:30 AM, using an image resolution of 64 × 64 pixels. The training configuration included a generator learning rate of 0.0021, a discriminator learning rate of 0.002, and a gamma value of 100 for R1 regularization. Training augmentation was implemented using DiffAugment, with translation selected as the augmentation pipeline.
The deliberately constrained source material and low-resolution training conditions constitute integral components of the artistic experiment.T
The objective was neither to pursue photorealistic reconstruction nor to preserve individual facial identities. Instead, the process investigates the generative possibilities emerging from the statistical reorganization of visual information extracted from politically and geographically distinct archives, connected by histories of state repression and the evolving technological infrastructures through which political dissent is monitored, identified, and suppressed. The 37-year interval separating Tiananmen Square in 1989 from the 2025–2026 protests in Iran reveals a profound transformation in the technological conditions of political control. While the Chinese government’s violent suppression of the 1989 demonstrations preceded the emergence of contemporary AI-powered biometric surveillance, the Iranian authorities now operate within an increasingly sophisticated environment of digital monitoring, facial recognition, and algorithmically mediated social control. This transformation acquires particular geopolitical significance through technological cooperation between China and Iran, including the reported provision of Chinese surveillance and telecommunications technologies to Iranian state institutions. Within this context, the project critically examines the transnational circulation of technologies of control and the political economies that sustain them, questioning how technological innovation can become entangled with authoritarian governance and the restriction of fundamental freedoms. By computationally merging faces drawn from these two historical moments, TT Tiananmen–Tehran (2026) transforms the generative process into a speculative confrontation between historical memory and contemporary machine vision, proposing morphological instability and facial opacity as aesthetic strategies for interrogating the expanding infrastructures of surveillance and political repression.
The neural network is thus approached as a computational environment in which photographic traces become entangled through processes of abstraction, synthesis, and morphological transformation. The resulting configurations do not constitute identifiable combinations of particular protesters. Instead, they emerge from the model’s learned distribution of facial features, producing unstable visual formations in which distinctions between the two source collections become increasingly difficult to establish.
Within this process, low-resolution imagery becomes both a technical condition and an aesthetic proposition. Computational imprecision, morphological discontinuity, and the instability of facial features open possibilities for questioning the relationship between visual representation, identity, and the classificatory operations of machine perception.
Following the eight-day training process, the model was explored through Autolume’s Perform mode, an interactive environment for manipulating generative visual transformations. Short videos were recorded during these performances, capturing continuous morphological transitions across the synthesized facial configurations. The resulting sequences revealed anomalous faces that appeared to melt, fragment, merge, and reorganize, producing unstable physiognomies situated between recognizable human features and computational abstraction.
The exploration of the generative latent space became a performative investigation into the instability of computational identity, in which the face was approached not as a fixed representation but as a continuously transforming configuration. Eight screenshots were selected from each of the recorded sequences, capturing distinct moments of morphological instability. These images became the starting points for the subsequent translation of the generative performances into sculptural form.
The eight selected screenshots were introduced into ChatGPT as visual inputs for generating speculative representations as three-dimensional facial masks. This stage established a recursive relationship between generative systems since the anomalous facial configurations produced through Autolume became the visual foundations for a second process of AI-assisted image generation. The resulting representations preserved and reinterpreted the irregular features, asymmetries, fragmented surfaces, and ambiguous anatomical structures emerging from the original neural synthesis. Instead of focusing on reconstructing anatomically conventional faces, the process explored how computational anomalies could be further articulated as sculptural morphologies. These images were subsequently uploaded to Meshy AI, where image-to-3D generation translated the two-dimensional representations into three-dimensional digital geometries. The resulting models were exported in STL format and prepared for additive manufacturing.
Through 3D printing, the computational transformations acquired physical presence, materializing a sequence of translations extending from archival photography to neural-network training, performative image synthesis, generative image interpretation, three-dimensional modeling, and sculptural fabrication. The eight masks thus embody a process of recursive computational morphogenesis, in which each generative operation introduces new morphological interpretations while retaining traces of preceding transformations. The work approaches artificial intelligence not as an autonomous creative agent or a neutral technical instrument, but as a participant in a distributed artistic process involving archival selection, computational training, performative experimentation, aesthetic judgment, and material fabrication.
Developed for the 2026 edition of EmMeio, at the National Museum of the Republic in Brasília, Brazil, TT Tiananmen–Tehran presents eight 3D-printed masks mounted individually on slender black wooden sticks supported by black wooden bases. Their vertical arrangement evokes the collective presence of bodies assembled in public space while withholding the anatomical completeness of the individuals they suggest.
Positioned approximately at human face height, the masks establish a field of spectral presences, suspended between individual identity and collective political memory. The installation operates through a tension between presence and absence - faces without bodies, identities without stable recognition, and material forms emerging from photographic records of historical dissent.
Instead of representing specific individuals or historical episodes, the sculptures suggest the persistence of political presence beyond the stability of individual identification. The spatial arrangement invites viewers to encounter the masks as a collective assembly of anomalous entities, each retaining a singular morphology while participating in a shared field of computational transformation.
At the conceptual center of TT Tiananmen–Tehran (2026) is the proposition of morphogenetic opacity, considering that an aesthetic and technoetic condition in which identity is not simply concealed or erased but continuously transformed through computational processes. The project engages a paradox of contemporary technological infrastructures.
While artificial intelligence increasingly participates in systems capable of identifying, classifying, and tracking human faces, generative neural networks can also produce unstable morphologies that challenge conventional expectations of anatomical coherence and visual recognizability. The anomalies emerging from the computational entanglement of the two photographic archives become opportunities for examining the political implications of facial identification and the limits of representational certainty.
The work does not claim that these masks are technically immune to biometric recognition but investigates the possibility of approaching computational illegibility as an aesthetic condition through which the assumptions of machine vision can be critically examined.
The historical references to political movements and episodes of repression across China, Hong Kong, and Iran expand this investigation beyond the technical operations of generative AI, situating the anomalous face within a broader reflection on collective agency, political vulnerability, and the contested visibility of dissent.
Through the recursive transformation of photographic information into generative images and sculptural objects, TT Tiananmen–Tehran proposes the face as a site of continuous becoming rather than a stable marker of identity.
The resulting masks inhabit an ambiguous territory between historical documentation and computational speculation, individual memory and collective presence, technological control and the possibility of resistance. The installation proposes the anomalous face as a technoetic interface between political memory and morphogenetic becoming, where the instability of computational representation opens an aesthetic space for questioning the politics of recognition, the boundaries of identity, and the infrastructures through which human presence becomes visible, classifiable, and governable.
Workshop held on November 7, 2026, as part of O Mundo Através da IA (The World Through AI), an exhibition at SESC 24 de Maio, São Paulo, inviting audiences to critically reflect on how artificial intelligence is transforming perception, culture, labor, creativity, and forms of social organization, while fostering a deeper understanding of the technologies shaping our present and envisioning possible futures.
There is substantial documented evidence that Chinese companies have supplied surveillance technologies, telecommunications infrastructure, and internet-control systems to Iran, including equipment associated with institutions involved in monitoring and repressing political dissent. More significantly for TT Tiananmen–Tehran (2026), a major human-rights investigation published in February 2026 explicitly connects Chinese technology transfers to the development of Iran’s surveillance and censorship infrastructure during the 2025–2026 protests.
Report: Tightening the Net: China’s Infrastructure of Oppression in Iran
Organization: ARTICLE 19
Publication: February 2026
Read the full ARTICLE 19 investigation (PDF)
This is probably the most relevant reference for the project. The investigation examines how Chinese companies and technology transfers have contributed to Iran’s increasingly sophisticated systems of digital censorship and surveillance. Importantly, the report acknowledges that it remains unclear whether Iranian authorities have specifically deployed Chinese biometric-recognition software alongside Chinese surveillance hardware. That distinction matters when attributing the origins of particular systems. Its findings include: Chinese companies have supplied equipment used in internet filtering, surveillance, and telecommunications monitoring; Chinese surveillance-camera manufacturers have established commercial relationships with Iranian distributors and state institutions; Some equipment has been supplied to organizations associated with Iran’s Islamic Revolutionary Guard Corps (IRGC); Iran’s National Information Network has developed capabilities resembling elements of China’s state-controlled internet architecture; Chinese technological cooperation has contributed to Iran’s capacity to restrict communications and monitor digital activity; The report identifies Huawei, ZTE, Tiandy, and Hikvision among the companies involved in these technological relationships.
One particularly significant case concerns Chinese manufacturer Tiandy. A 2025 investigation by cybersecurity research company Recorded Future documents reports that Tiandy entered into a five-year agreement involving Iranian government institutions, including security and military organizations. The report identifies surveillance technologies with capabilities such as facial recognition, crowd detection, movement tracking, and low-light imaging. It also discusses Iranian distributors advertising AI-enabled cameras and analytical software. The relationship extends beyond the export of ordinary telecommunications equipment. It involves technologies capable of supporting the identification, monitoring, and tracking of people in public spaces. However, the presence of such capabilities does not independently establish that a particular camera was used to identify protesters during a specific demonstration.
Recorded Future: Iran's AI Ambitions and National Security (2025, PDF)
A peer-reviewed study published through Cambridge University Press provides further evidence.
Academic article: Digital Authoritarianism and the Global Technology Industry: Evidence from Iran
The study documents the international commercial networks through which surveillance technologies reach Iranian institutions, including the role of Chinese manufacturers and local distributors. It specifically discusses Tiandy equipment supplied to Iranian military and security organizations.
Read the academic article (PDF)
A February 15, 2026 investigation by The New York Times, also published by The Irish Times, describes Iranian authorities’ use of digital surveillance to identify participants in the demonstrations that began in December 2025. According to the investigation, protesters received messages indicating that their presence at demonstrations had been detected. Researchers concluded that location data from mobile phones was a likely source of identification. The reporting also describes accounts of individuals being interrogated using information obtained through facial recognition and phone-data analysis. There is evidence of digital surveillance being used against protesters in 2026, and separate evidence that Chinese technology has contributed to Iran’s surveillance infrastructure. What remains harder to establish is precisely which Chinese-manufactured systems were involved in identifying particular protesters.
Read the February 2026 investigation
The available evidence also demonstrates that Iran’s surveillance infrastructure is not exclusively Chinese. A March 2026 investigation involving Le Monde, Forbidden Stories, and other international media organizations documented Iran’s acquisition of the Russian facial-recognition system FindFace. The investigation identified contracts and intermediaries through which this technology reached Iranian organizations, including entities connected to security institutions. This complicates any argument that Iran’s entire biometric-surveillance infrastructure was designed and supplied by China. A more defensible interpretation is that Iran’s surveillance apparatus has developed through domestic institutions and transnational technological relationships involving multiple suppliers, with Chinese companies playing a documented and significant role.
Within this context, TT Tiananmen–Tehran (2026) examines the transnational circulation of technologies of control and the political economies through which technological innovation becomes entangled with authoritarian governance and restrictions on fundamental freedoms.
Tiananmen and Tehran are connected not only through the historical experience of political repression but also through the evolution and international circulation of technologies capable of transforming political bodies into identifiable, trackable data.
A documented relationship gives TT Tiananmen–Tehran (2026) a precise geopolitical dimension while allowing its generative anomalies to remain open to the more speculative questions of technoetic aesthetics and morphogenetic opacity.
The core dimension of the project is the transformation in surveillance technologies across the 37 years separating the two photographic archives. In 1989, the Chinese authorities’ suppression of the Tiananmen demonstrations occurred before contemporary AI-powered facial recognition and automated biometric monitoring. By 2026, Iranian authorities had access to a much more sophisticated infrastructure of digital monitoring, internet censorship, mobile-phone tracking, and algorithmically assisted identification. Chinese technology transfers are one documented component of this transformation. Consequently, the relationship between Tiananmen and Tehran is not exclusively symbolic or historical. It can also be examined through the international circulation of technological systems that facilitate state surveillance. This creates a particularly interesting conceptual tension: the same broad field of computational technologies that enables increasingly sophisticated identification and surveillance also provides the generative methods through which the artwork produces anomalous, morphologically unstable faces. That relationship is not evidence that the masks evade facial-recognition systems. It is, however, a substantial foundation for investigating the politics of computational visibility.
The objective was neither to pursue photorealistic reconstruction nor to preserve individual facial identities. Instead, the process investigates the generative possibilities emerging from the statistical reorganization of visual information extracted from politically and geographically distinct photographic archives, connected by histories of state repression and the evolving technological infrastructures through which political dissent is monitored, identified, and suppressed. The 37-year interval separating the 1989 Tiananmen Square demonstrations from the 2025–2026 protests in Iran reveals a profound transformation in the technological conditions of political control. While the Chinese government’s military suppression of the 1989 demonstrations preceded the emergence of contemporary AI-powered biometric surveillance, the Iranian authorities now operate within an increasingly sophisticated infrastructure of digital monitoring, facial recognition, mobile-device tracking, and algorithmically mediated social control.
This historical relationship acquires particular geopolitical significance through documented technological cooperation between China and Iran. Investigations by ARTICLE 19 (2026), alongside academic research into digital authoritarianism and international technology transfers, identify Chinese companies, including Huawei, ZTE, Tiandy, and Hikvision, as suppliers of telecommunications and surveillance technologies to Iranian institutions. These relationships have contributed to the development of infrastructures supporting internet censorship, communications monitoring, and the expansion of state surveillance capabilities. The political implications of these technological exchanges became particularly visible during the 2025–2026 Iranian protests, when digital monitoring, restrictions on internet access, and the reported identification of demonstrators through mobile-device data and facial-recognition systems formed part of the authorities’ response to political dissent.
By computationally entangling facial images extracted from these two historical moments, the project transforms generative neural synthesis into a speculative confrontation between historical memory and contemporary machine vision. The anomalous facial configurations emerging from the generative process challenge the aesthetic conventions of biometric legibility, proposing morphogenetic opacity not as a demonstrably effective means of escaping surveillance, but as a technoetic strategy for interrogating the increasingly automated infrastructures through which human identity becomes visible, classifiable, and governable.
ARTICLE 19 (2026). Tightening the Net: China’s Infrastructure of Oppression in Iran. Particularly relevant for Chinese technology transfers, internet-control systems, and the political consequences of China–Iran technological cooperation.
Digital Authoritarianism and the Global Technology Industry: Evidence from Iran. Peer-reviewed research available through Cambridge University Press, documenting surveillance technology transfers and the role of international suppliers in Iran’s technological infrastructure.
Recorded Future (2025). Iran’s AI Ambitions: Balancing Economic Isolation with National Security Imperatives. Relevant for the documented capabilities of Chinese-manufactured surveillance systems and their commercial relationships with Iranian institutions.