2025 Student Academy Award Winners Series Part 5: flower_gan

flower_gan, dir. Mati Granica

“The work that you are seeing is perfectly legal—but should it be?”

The unconventionally inventive short film flower_gan stands at the crossroads of creativity and technology, confronting head-on the future of filmmaking with both curiosity and critique. It challenges the promises of AI-driven creation where speed and scale threaten to dissolve the human touch and asks what is lost when art becomes data. In a landscape rushing toward automation, flower_gan reveals the costs beneath the surface: the energy drained, the authorship blurred, the ethics unsettled. It holds a mirror to the filmic future, one shaped by algorithms and endless reproduction, and invites us to question: how can storytelling survive and evolve within these shifting systems?

Director Mati Granica is a visual artist based in London, working with photography, 3D modelling, and artificial intelligence. His practice moves through the friction between rapid technological growth and ecological collapse. With a critical eye and a seductive surface, his work reveals the contradictions beneath digital progress. As a winner of a Student Academy Award in the Alternative/Experimental category, flower_gan is certainly among the most radically innovative, unorthodox films to ever win an Academy Award.

flower_gan is a custom Generative Adversarial Network (GAN) that attempts to create images of flowers. The project traces the making of artificial intelligence, paradoxically using extractive systems to reflect on their own impact. Through AI, flowers are abstracted, blurring, fading, becoming unrecognizable. 27,616 images emerge, echoing the excess of machine learning. Trained on absence, not essence, the flowers distort into impossible forms almost like ghosts of their originals, untethered from what they once were.

Generative AI has the potential to be the biggest innovation in filmmaking since perhaps digital video, or computer-generated imagery. flower_gan lives within the tensions of the present including AI, authorship, and extraction. It speaks to systems built on imbalance: data mined without consent, resources consumed without pause, profit driving the pulse of the machine. The dataset used, oxford_flowers102, is a collection of images scraped from the web, freely licensed, yet not freely given. 8,189 blooms, gathered for science, classification, creation and now repurposed and caught in a cycle of endless generation. flower_gan questions the ethics of this quiet taking, a reflection on ownership, on the ethics of use, in a time when such questions can no longer be ignored.

The use of Generative AI is not without controversy. In the last several years, films as diverse as Late Night with the Devil and The Brutalist have had their use of AI (for the creation of intertitles, and the correction of actor’s accents, respectively) bristle purists in the filmmaking community. But since then there have been films that qualified for Academy Awards made entirely with Generative AI (The Eggregores’ Theory, a horror short film that has won short film festival awards) and at least two films in the most recent Toronto International Film Festival have been open about their use of GenAI: Fuck My Son! has AI-created animated characters, and the short film 09/05/1982 was entirely created with Generative AI.

Ironically, the creation of flower_gan uses the environmental resources that make the flowers grow and then disappear, a paradox that Granica recognizes and pushes against, even has he makes use of these resources. Each image, a flicker of energy spending 2.9 watthours, with 1.3 grams of CO₂ released. What happens when one becomes hundreds, thousands, millions? Generative AI does not pause, it is built to multiply, to scale without end. With each new image, the weight grows heavier as it is trained on vast archives of millions of images scraped, stored, and consumed.

Generative AI was sold on its limitless potential. But it is fragile, and used blindly, with its costs growing unchecked. We build, we consume, we expand without pause. Not progress, but acceleration toward collapse. In the end, Granica leaves us with more questions than flower_gan is able to answer: Who gains from this future? And who is left paying for it? Is this the future you wanted? Must it unfold this way? Could AI be powered with care, trained with consent, designed for something more than profit? Who holds the power? Who pays the energy bill? Where is accountability? flower_gan moves within this contradiction, critiquing systems it cannot escape. It uses the very tools it questions. It asks what complicity looks like, and whether it is possible to break the loop... Or what happens if we don’t.

Review by: Joshua Hunt