This article investigates the impact of artifcial intelligence on contemporary audiovisual creation through three emblematic case studies. It demonstrates how generative systems reconfgure creative processes without compromising artistic authenticity, establishing a dialogue between advanced computational techniques and poetic expression. The analysis reveals that algorithmic curation, when guided by rigorous aesthetic principles, enhances the emergence of hybrid languages in immersive narratives. The results point to a paradigm of expanded authorship, where human-machine collaboration transcends mere technological instrumentalization to constitute a new feld of aesthetic and epistemological experimentation.
The technological revolution of recent years has radically redefned artistic practices in the audiovisual feld. The ability of artifcial intelligence systems to generate original content – from synthetic images to algorithmic musical compositions – raises fundamental questions about authorship, creativity, and the very status of artwork in the digital era. This study adopts an interdisciplinary perspective, articulating contributions from computer science, media studies, and critical theory to examine how generative models are transforming not only production tools but also the epistemological foundations of artistic creation.
The emergence of Generative Artifcial Intelligence (GAI) systems has caused signifcant
tensions in cultural practices developed in contexts where computational devices assume tasks
traditionally associated with human creativity. As Santaella [
The present study enters this debate by analyzing three audiovisual projects that employ computational intelligence systems at diferent levels of their creative processes. More than isolated examples, these cases constitute paradigmatic manifestations of a broader movement that is redefning the boundaries between human creation and algorithmic processing. The central hypothesis guiding this investigation is that, far from representing a threat to artistic authenticity, generative systems can enhance poetic expression when subjected to rigorous aesthetic curation.
Rather than adopting a deterministic stance - either utopian or dystopian - we follow a constructivist view of authorship, recognizing the role of both human intentionality and algorithmic agency (Di Dio et al., 2023). In this paradigm, collaboration between humans and computational systems is not reduced to mere technological instrumentalization but constitutes a fertile feld for aesthetic and epistemological experimentations that challenge traditional categories of artistic thought.
The feld of computational creativity, as defned by Wiggins [
Computational creativity in the contemporary audiovisual context signifcantly transcends the
limits of automation and reproduction of preexisting aesthetic patterns. Cheng [
The Schema theory, as applied by Cheng [
Technical advances in generative algorithms have been fundamental to this evolution.
Generative Adversarial Networks (GANs), proposed by Goodfellow et al. [
This technical evolution, however, should not be understood merely in terms of instrumental
capabilities. As Santaella [
In the specifc context of audiovisual production, this dialectic manifests in the growing
integration between traditional narratives and adaptive systems. Rodrigues et al. [
The question of authorship in AI-mediated creative environments has been the subject of
intense academic debate. Prass et al. [
This new paradigm does not, however, imply a complete dissolution of human agency in the creative process. On the contrary, as we will demonstrate in the following case studies, successful integration between generative systems and artistic expression fundamentally depends on rigorous aesthetic curation, where poetic and conceptual principles established by the human artist guide the algorithmic exploration of creative possibilities. It is in this productive tension between computational autonomy and human intentionality that the most innovative forms of contemporary audiovisual creation emerge.
The selection of these three cases—spanning experimental flmmaking, interactive simulation, and generative projection—was guided by their methodological diversity and their relevance to the central hypothesis of this paper. Each project exemplifes a unique mode of integrating generative AI systems into the creative process, allowing for a comparative exploration of hybrid authorship across media formats
The "SOMOS: A Criação de Ybirá-Ubuntu" project exemplifes the integration between traditional aesthetic principles and advanced generative techniques. The use of ComfyUI Stable Difusion workfow for visual generation allowed an unprecedented approach to nature representation, where deep learning algorithms were employed not as mere reality simulators, but as mediators of a new visual language. The creative process involved meticulous curation of latent spaces, ensuring that the stochastic properties of the model served the narrative without dominating it.
The work represents a paradigmatic case of how generative artifcial intelligence can be
employed to transcend the limitations of conventional representational approaches. Rather than
simply reproducing photorealistic images of nature, the project explored the capabilities of
difusion models to create a visual language that captures the interconnected essence of forest
ecosystems. As Santaella [
The workfow implemented in ComfyUI using Stable Difusion, allowed a systematic
exploration of the latent space of visual possibilities, where each iteration represented not just an
aesthetic variation, but a conceptual reconfguration of the relationship between natural elements.
This approach directly dialogues with the concept of "transformational creativity" proposed by
Wiggins [
While the visual aesthetic may resemble common outputs from Stable Difusion, what distinguishes these frames is the recursive compositional strategy informed by a feedback loop between textual prompts and visual outputs. The fnal selections were further validated through dialogues with guest artists and curators, who highlighted the coherence between narrative intent and generative form.
A particularly innovative aspect of the project was the implementation of a feedback system
between the generated visual outputs and subsequent textual prompts, creating a hermeneutic
cycle where each new image informed the direction of subsequent explorations. This iterative
process exemplifes what Colton and Wiggins [
The conceptual dimension of "SOMOS: A Criação de Ybirá-Ubuntu" also deserves highlighting,
as it establishes a dialogue between Brazilian indigenous cosmovisions and African philosophies,
mediated by cutting-edge technologies. This juxtaposition of diverse cultural references, processed
through AI systems, raises considerations about cultural representation that, while not the focus of
this article, remain important aspects of AI-mediated creative work. As Prass et al. [
"O Desafo de Carbonia" exemplifes the convergence between adaptive narratives and complex
simulation systems, creating a learning environment that responds dynamically to user
interactions. The system architecture, based on autonomous agents with emergent behaviors,
allows each game session to develop unique trajectories, making the educational experience
simultaneously structured and unpredictable. This approach directly enacts the adaptive narrative
logic described by Rodrigues et al. [
A particularly innovative aspect of the project was the implementation of a feedback system
that translates abstract concepts of sustainability and circular economy into tangible game
mechanics. The reinforcement learning algorithms implemented in virtual agents not only simulate
realistic ecological behaviors but also adapt their responses to players' decisions, creating a virtual
ecosystem that evolves over time. This capacity for dynamic adaptation exemplifes Prass et al. [
The pedagogical dimension of "O Desafo de Carbonia" deserves special attention, as it demonstrates how AI systems can be employed not only as tools for artistic creation but as mediators of complex educational processes. The simulation allows abstract concepts such as environmental externalities, tragedy of the commons, and sustainable development to be experienced in a concrete and interactive way, facilitating the understanding of systemic dynamics that are difcult to apprehend through traditional pedagogical methods.
The integration between narrative, game mechanics, and scientifc models in "O Desafo de
Carbonia" exemplifes what Goodfellow et al. [
In order to represent the socio-political and cultural aspects of Carbonia, we created a fctional universe composed of eight nations, each with its own characteristics. The visual representation of these nations was developed using ComfyUI with Flux.1-Schnell1, allowing for an aesthetic that aligns with the thematic diversity of the project. This combination of procedural content generation and cultural representation refects the creative potential of generative AI in commercial applications.
The "PindorAIm" installation presented at the 2025 Virada Cultural constitutes perhaps the most complete expression of the fusion between technology and cultural tradition. Through the generative mapping of visual patterns on the facade of the Pateo do Colégio, in São Paulo, the work established a profound dialogue between collective memory and technical innovation. The use of ComfyUI – a node-based interface for managing AI models such as Stable Difusion, allowed for the rapid development of creative ideas by detecting the physical characteristics of the architectural design. This system was able to recreate textures and topologies using image guiding lines as a reference. By leveraging this process, we were able to generate dozens of possible scenarios to transform the reality of the facade through mapped projection.
"PindorAIm" represents an exploration of generative videomapping that transcends traditional
limitations by dynamically adapting visual compositions to the architectural context. The work
establishes a dialogue between the historical heritage of the Pateo do Colégio, in São Paulo, and
cutting-edge technologies, creating an immersive experience that simultaneously activates
collective memories and individual perceptions. As Santaella [
A particularly innovative aspect of the project was the implementation of a system capable of
interpreting the architectural features of the building's facade, dynamically generating visual
patterns that respect the structure's historical and physical characteristics. This architectural
responsiveness exemplifes what Colton and Wiggins [
The historical and cultural dimension of the installation deserves special attention, as it demonstrates how generative technologies can be employed not only to create innovative aesthetic experiences but also to reactivate and recontextualize collective memories. By mapping generative projections onto the historical architecture of the Pateo do Colégio, the work establishes a temporal continuum that invites participants to refect on urban and cultural transformations.
The integration between digital generative systems and architectural heritage in "PindorAIm"
exemplifes what Prass et al. [
The comparative analysis of the three cases presented reveals signifcant patterns in the
integration between generative systems and human artistic expression. In all projects, it is
observed that the aesthetic and conceptual coherence of the works arises not from algorithmic
autonomy alone, but from iterative calibration guided by human curatorship and theoretical intent,
as defned in Jordanous [
A particularly relevant aspect that emerges from the analysis is the importance of algorithmic curation as a contemporary artistic practice. In "SOMOS," the careful selection of parameters and deliberate navigation through the latent space of the difusion model constituted fundamental elements of the creative process. Similarly, in "O Desafo de Carbonia," the calibration of multiagent systems to balance emergent complexity and narrative coherence demanded an iterative process of adjustments and evaluations. In "PindorAIm," the integration between responsive systems and pre-existing architectural elements required constant mediation between technological possibilities and cultural context.
This centrality of algorithmic curation directly dialogues with the concept of "hybrid
authorship" proposed by Boden [
The analyzed cases also reveal signifcant ethical challenges that emerge from the use of generative systems in artistic contexts. A particularly sensitive aspect concerns algorithmic raises concerns regarding algorithmic cultural appropriation, particularly when models trained on datasets with Euro-American and Asian dominance are used to generate content inspired by Indigenous and African cosmovisions, demanding ethical vigilance. In "SOMOS," for example, the juxtaposition between Brazilian indigenous cosmovisions and AI technologies trained on global datasets raises important questions about representation and cultural authenticity.
Another relevant ethical challenge concerns transparency about the generative processes
employed in the works. As Boden [
The systematic evaluation of creative systems, as proposed by Jordanous [
The analysis of the three empirical cases presented in this study reveals that the integration between generative artifcial intelligence systems and audiovisual artistic practices is producing a paradigm of generative co-creation that challenges traditional divisions of author, tool, and medium, fostering polyphonic artistic ecologies. As demonstrated in "SOMOS," "O Desafo de Carbonia," and " PindorAIm" generative systems do not function merely as tools serving a preexisting artistic vision, but as active agents in a dialogical process that reconfgures both the expressive possibilities and the conceptual foundations of audiovisual creation.
A particularly signifcant aspect that emerges from this investigation is the centrality of
algorithmic curation as a contemporary artistic practice. In all analyzed cases, it was observed that
the aesthetic and conceptual efectiveness of the works did not derive from the absolute autonomy
of the algorithms, but from a generative dialectic between intentional composition and emergent
behavior, where authorship becomes a negotiated space. This dialectic between artistic
intentionality and algorithmic processing, as Santaella [
The concept of "hybrid authorship," as proposed by Boden [
The ethical challenges identifed in this study, particularly concerning algorithmic cultural
appropriation and transparency about generative processes, demand the elaboration of specifc
protocols for responsible creation. McCormack et al. [
The implications of this transformation extend beyond the strict feld of artistic production,
suggesting the need for new theoretical models that account for the complexity of these
humanmachine interactions. As Colton and Wiggins [
Ultimately, the fndings of this study indicate that the future of audiovisual creation will not be defned by the replacement of human creativity with autonomous systems, but by the emergence of complex creative ecologies where diferent forms of agency – human, algorithmic, cultural, material – interact in dynamic and unpredictable ways. In this scenario, the challenge for artists, researchers, and cultural institutions is not to resist the integration of generative systems into creative practices, but to develop critical and refexive approaches that allow exploring their expressive potential in ethical and culturally signifcant ways.
The authors would like to thank Noisetupi Audiovisual Productions for supporting the development of the projects discussed in this study. Special thanks are extended to the collaborators and technical teams who contributed to the realization ofSOMOS: A criação de
feedback from colleagues at various academic and artistic institutions, which helped refne the conceptual and technical aspects of this work.
through the Secretaria de Cultura de Santo André. This support was fundamental in realizing the interdisciplinary and creative aspects of the project.
The author employed ChatGPT-4 for grammar refnement and textual fuency improvements. No content was generated autonomously by LLMs. Conceptual development and critical framing were entirely authored by the researcher. Additionally, the following AI tools were used for generating visual content:
O desafio de Carbonia and PindorAIm: ComfyUI with Flux.1-Schnell by Black Forest Labs.