虛擬攝影的四向度閱讀：以《LUMI》為案例

Four-Dimensional Reading of Virtual Photography: A Case Study of LUMI

本文提出一套分析虛擬攝影的四向度閱讀框架，以Abelardo Gil-Fournier與Jussi Parikka的錄像散文《LUMI》（2024）為主要分析對象。虛擬攝影的「現實基礎」已從物理場景轉向可運算的資料集與潛在客觀世界，傳統攝影理論中以索引性與再現為核心的觀看前提因此受到挑戰，影像如何在演算法結構中生產意義與執行功能也成為迫切的問題。為此，本文將超現實攝影理論中既有的三項視覺邏輯（摹仿性、前攝影性、謎樣性）與操作性影像的論述加以整合，建構出四向度閱讀框架，並透過《LUMI》加以考察。摹仿性說明影像如何透過統計學習維持可信的視覺外觀；前攝影性追問資料集的前置處理如何預先決定影像的生成條件；謎樣性標示符號結構與技術結構之間無法完全疊合的不可解層面；操作性則指向影像在計算流程與自動化決策中的功能性作用。這套框架是一種批判性的分析工具，可延伸應用於監控影像、氣候視覺化或AI新聞攝影等不同脈絡。虛擬攝影是與資料治理、演算法決策及環境感知深度交織的影像形式；理解虛擬攝影，不只是辨識一種新的影像類型，更是在追問當代影像如何重新形塑我們與世界的關係。

The rapid proliferation of artificial intelligence (AI)-generated imagery has fundamentally altered the conditions under which photographs are produced, circulated, and interpreted. Virtual photography refers to computer-generated images that visually simulate photographic effects without relying on optical imaging mechanisms. It has emerged across domains including computer graphics, cinematic visual effects, video games, and digital art. As Ali Shobeiri (2024) argues, virtual photography originates from a virtual core that possesses full reality in the Deleuzian sense. Unlike analog and digital photography, which are born from the realization of light upon photosensitive material, virtual photography exists through the continuous actualization of data. This ontological shift means that the ground of reality for such images has migrated from physical scenes to computable datasets and algorithmic structures. Consequently, photography theory’s longstanding reliance on indexicality and representation as the foundations of visual meaning has destabilized, and the question of how images produce meaning and perform functions within algorithmic frameworks has become more pressing.
This paper addresses this challenge by proposing a four-dimensional analytical framework for reading virtual photography, using the video essay LUMI (2024) by Abelardo Gil-Fournier and Jussi Parikka as its primary case study. LUMI combines AI‑generated imagery and sound design to explore cryosphere as a planetary‑scale reflective system, employing a first‑person narrative voice that speaks as an AI trained on historical mountain and glacier photography from the mid‑19th to the mid‑20th centuries. The central research questions are: How can we develop a systematic reading method that accounts for both the semiotic and the computational dimensions of virtual photography? How does LUMI, as a specific artistic practice, reveal the tensions among photographic appearance, data preprocessing, symbolic indeterminacy, and operational functionality in AI-generated images?
This paper draws on several intersecting theoretical frameworks. David Bate’s Photography and Surrealism (2004) provides the foundational analytical categories of mimetic, prophotographic, and enigmatic - three modes through which surrealist photography can disrupt the naturalized transparency of photographic representation. Bate conceptualizes surrealism not as a fixed image type, but as“a mode of treating signs,”or a symptomatic reading strategy that attends to contradictions and ruptures within images. Jussi Parikka’s Operational Images (2023) and What is Media Archaeology? (2012) offer a related yet distinct perspective, foregrounding how images function as components of technical and automated processes rather than as objects of human contemplation. Gil-Fournier and Parikka’s Living Surfaces (2024) conceptualizes earth surfaces as progressively decomposed into analyzable image environments through optical instruments, cartographic systems, and data visualization. Amanda Wasielewski’s concept of“the latent objective world”(2024) elucidates how AI‑generated photography reveals not the physical world, but the object world of the dataset, insofar as training images have already objectified reality, and generative AI further displaces referentiality from the photographed scene to the data that produce it. Mark Polizzotti’s Why Surrealism Matters (2024) and David Lomas’s article on surrealist automatism (2012) provide historical grounding for the parallel between surrealist sign operations and AI’s generative logic.
The paper methodologically extends Bate’s three-category analysis of surrealist photographic signifiers by integrating Parikka’s operational image discourse, thereby constructing a four-dimensional reading framework. The four dimensions of mimetic, prophotographic, enigmatic, and operational are treated as co-present and simultaneously active analytical axes rather than as a linear sequence. This framework is then applied to a close reading of LUMI, examining specific visual sequences, narrative strategies, and computational procedures within the work. The approach combines semiotic analysis with attention to technical infrastructure, drawing on media archaeology’s commitment to tracing intersecting yet distinct historical lineages. The paper does not claim historical continuity between surrealist photography and virtual photography; rather, it mobilizes surrealism’s analytical tools as a means of addressing the structural parallels in how both practices destabilize the relationship between signifier and signified.
The analysis reveals several key findings. First, the mimetic dimension shows that virtual photography sustains visual credibility not through indexical reference to physical reality, but through statistical consistency with existing photographic conventions. In LUMI, the AI model learns both physical parameters (luminance distribution, reflective patterns) and cultural conventions (composition, tonal range, grain texture) from its training dataset of historical photographs, producing images that viewers initially recognize as“real”mountain photography.
Second, the prophotographic dimension shifts from concerns with how the scene was arranged before the shutter to how features are selected and encoded before generation. The systematic erasure of human figures from glacier landscapes in LUMI, as they are marked, classified, and removed by the algorithm, shows that data preprocessing functions as a form of data governance. They determine what can enter the image and what is excluded as noise.
Third, the enigmatic dimension identifies an irreducible gap between symbolic and technical structures. LUMI’s generated glacier images simultaneously function as credible landscape photographs and as artifacts of statistical recombination. They produce a fundamental instability in both the signifier-signified relationship and the categorical identity of the image itself, in that they look like photography, but are not records of any optical event.
Fourth, the operational dimension traces how images function within computational and institutional systems. In LUMI, virtual photography operates as a surface for calibrating predictive models, shaping risk imagination, and producing visual evidence for climate discourse. The work simultaneously exposes the paradox that algorithmic capacity to generate glacier images co-exists with the ongoing physical dissolution of actual glaciers.
The four-dimensional framework is proposed as a critical analytical tool rather than a closed interpretive model. While this paper focuses on LUMI as an art practice where operationality manifests primarily through the reshaping of environmental imagination, the framework’s applicability extends to other contexts. In surveillance systems, social media filters, in‑game photography, and emerging uses of generative imagery in journalism, the configuration and relative weight of the four dimensions will differ. Understanding virtual photography thus entails interrogating the broader infrastructure of data governance, algorithmic decision-making, and environmental perception through which contemporary images reorganize our relation to the world.