01 / Context
The rapid rise of generative artificial intelligence has reignited longstanding debates about the nature of creativity and artistic agency. Text-to-image and text-to-video systems are now capable of producing complex audiovisual artefacts from simple prompts, raising fundamental questions about whether creativity can be automated and how authorship is reconfigured when aesthetic decisions are delegated to algorithmic processes. These developments invite a re-examination of historical avant-garde movements that similarly sought to decentre the artist and disrupt conventional understandings of creativity. Among these, Dada occupies a particularly significant position - through its embrace of randomness, procedural methods, and anti-authorial practices, Dada echoes and potentially anticipates some of the conceptual issues raised by contemporary generative AI systems.
02 / Experiment
This research explores the convergence of Dada and generative AI as socio-cultural and creative phenomena through an experimental adaptation of Julius Evola's 1920 Dadaist poem "La fibre s'enflamme et les pyramides" into AI-generated moving images. It compares two video adaptations produced under distinct infrastructural conditions: a commercial, proprietary adaptation using KlingAI; and an open-source adaptation using WAN 2.1 T2V deployed locally in ComfyUI with a Low Rank Adaptation (LoRA) fine-tuned on surrealist and neuralist imagery.
Research
Questions
Question 01
Can GenAI make Dada? In other words, can probabilistic GenAI models that rely on large technofeudal corporate infrastructures ever reproduce anti-systemic artistic practices? If so, how?
Question 02
What does the adaptation of Dada through different generative AI infrastructures reveal about contemporary machine creativity? Proprietary and open-source models are not neutral tools but distinct interpretive regimes. How do these regimes differ in how much of their logic remains open to the user, and what kind of "chance" they produce?
Julius Evola / 1920
The
Poem
la fibre s'enflamme et les pyramides(très vite) aeaeaeaeaea eda s'éclairent les digues verticaleslédah égales torpilleus aux fointaines ne touchez passous l'orage extrarose mourir mourir les ancresles soeurs grises et les philosophes sur l'ultratlantiqueles coupoles ocegooraaa crépusculederrière le pastel les perforatrices les perforatriceshhhaa il a signé le quadruplebregan a eaeaeaeaeaeaeaeaeaeaeaea
Methodology
Aesthetics
vs. Principles
This research does not ask whether GenAI can produce Dada aesthetics as a surface style - the models certainly can, and they reproduce them effortlessly from their training distributions. Rather, it asks whether GenAI can reproduce the discipline of Dada: its anti-authorial ethos, its rejection of the artist's intentionality, its systematic decentring of creative agency, and its commitment to procedures that exceed individual will.
The Results
Proprietary models
Sora 2 Fast
Whole Poem as Prompt
Line-by-Line Poem as Prompt
Kling 2.5
Whole Poem as Prompt
Line-by-Line Poem as Prompt
Veo 3.0
Whole Poem as Prompt
Line-by-Line Poem as Prompt
Open-Source model
WAN 2.1 T2V
Whole Poem as Prompt / Without LoRA
Line-by-Line Poem as Prompt / Without LoRA
Whole Poem as Prompt / With LoRA
Line-by-Line Poem as Prompt / With LoRA