"This research proposes a participatory, decolonial and feminist approach to AI through the concept of “AI as commons”, offering a practical approach for cultural practitioners to decolonise existing AI systems. Conducted in Seoul, South Korea, the study combines critical theory with practical intervention, introducing “critical data injection” as a method for communities to influence representation in generative AI. At a workshop at Yonsei University, participants created datasets of everyday Seoul images which challenged typical AI-generated urban imagery, then trained Low-rank Adaptation (LoRA) models to generate alternative visual narratives. The accompanying LoRA Training Toolkit makes AI model training accessible to non-expert users. While acknowledging limitations, such as dependence on biased base models and infrastructure constraints, the research demonstrates how local communities can create representational alternatives within existing AI systems. The study argues for publicly-funded, modular AI infrastructure which enables democratic participation in AI development rather than relying solely on corporate or state-controlled systems." (Abstract)