We present AttentionBender, a tool that manipulates cross-attention in Video Diffusion Transformers to help artists probe the internal mechanics of black-box video generation. While generative outputs are increasingly realistic, prompt-only control limits artists’ ability to build intuition for the model’s material process or to work beyond its default tendencies. Using an autobiographical research-through-design approach, we build on Network Bending to design AttentionBender, which applies 2D transforms (rotation, scaling, translation, etc.) to cross-attention maps to modulate generation. We assess AttentionBender by visualizing 4,500+ video generations across prompts, operations, and layer targets. Our results suggest that cross-attention is highly entangled: targeted manipulations often resist clean, localized control, producing distributed distortions and glitch aesthetics in place of linear edits. AttentionBender contributes a tool that functions both as an Explainable AI-style probe of transformer attention mechanisms, and as a creative technique for producing novel aesthetics beyond the model’s learned representational space.

Increasingly, AI video models are marketed as "world simulators," suggesting their ability to model infinite realities. Despite such claims, these models systematically exclude significant aspects of embodied human experience, particularly sexuality. World Simulator is a video installation exploring the poetic friction between these universal claims and the models’ inherent blindness. To do so, the work feeds explicit gay erotica into an AI video-to-video pipeline. Lacking the training data to recognize these images, the system hallucinates surreal alternatives, transforming intimate acts into banal scenes of kitchen appliances, strange architectures, and abstract flesh. By visualizing the limits of synthetic knowledge, the work challenges the hubris of the "world simulator" label, asking how a system, trained primarily on large filtered video datasets, can claim to simulate the world while remaining structurally blind to the body. Beyond this critique, we question the value of simulation itself, asking what forms of sensual representation might offer more expansive, life-affirming possibilities.

Musicians are using neural audio models to synthesize new sounds, but the sound space accessible through such models is often constrained to essentially interpolated approximations of existing training data, limiting the true sonic novelty musicians can elicit from generative models. To address this problem, we present Latent Novelification (LN), a machine learning method which learns to transform vectors in any generative model’s latent space into new, contrasting vectors according to an objective function which we define. The LN objective enforces that these contrasting vectors explicitly diverge from existing training data while remaining semantically meaningful to the generative model, facilitating generative novelty beyond simple interpolation. LN supports divergent creativity by helping musicians generate new and novel sounds, integrating into any neural audio model interface to controllably and interactively expand accessible sound space. This work contributes a new algorithmic method—implemented in open-source software—which musicians working with neural audio synthesis can use to synthesize new sounds to incorporate into their art.

Me vs. You is a multi-channel video installation that explores the complexities of queer intimacy by co-opting an AI machine vision system. In this work, footage from a wrestling match is transformed through a generative video pipeline into a fluid interaction that oscillates between aggression and tenderness. The initial wrestling footage is fed through an AI depth map network—designed to separate bodies—before being reconstructed using a diffusion video process. Rather than rendering discrete fighters, the system produces unstable, shifting forms that sensually collide and merge, destabilizing a clear reading of the interaction. The work exploits the machine vision system’s inability to delineate entangled bodies, challenging computational frameworks of classification and control; instead, it repurposes AI as a tool for poetic ambiguity. Situating Me vs. You within experimental filmmaking and AI surveillance debates, this paper examines how emerging technologies can disrupt narrow modes of machine perception and proposes more expansive ways of seeing.

Systems to interactively generate audio from human movement are used by artists including dancers to support their performances and practice. However, current real-time movement-to-sound systems require specialized hardware or expertise, or map only very simple movement-to-audio relationships. We present a new technique and system implementation for interactive sonification of human movement through unsupervised machine learning. Our system maps between latent spaces, linking a pose estimator to a neural audio generator to enable sonification of human bodies. This may lower barriers to entry for artists to generate sound from their embodied movement through complex mappings. Our system requires no specialized hardware or niche AI expertise, minimal data to learn a user's custom movements, and trains extremely fast. It represents a new method for mapping custom data to a latent space through unsupervised learning, and advances state-of-the-art interactive movement sonification through its increased accessibility and ease of use relative to its complexity.

This workshop explores the integration of Artificial Intelligence (AI) into filmmaking, focusing on AI-driven video content analysis (VCA), AI-assisted content creation, and ethical considerations.As AI continues to reshape creative workflows, it opens new possibilities for filmmaking while raising important questions about human-AI collaboration.The workshop aims to bridge perspectives between creative practitioners, industry professionals, and AI researchers, fostering interdisciplinary dialogue on AI's evolving role in creative practice.We will discuss how AI-powered VCA advances film grammar analysis and audience cognition research, informing creative decision-making, and how generative AI supports production processes while expanding artistic possibilities.Additionally, we will examine how AI-driven analysis can inform ethical practices, alongside addressing the risks associated with generative AI.Through a combination of theoretical discussion and practical demonstrations, participants will gain hands-on experience with AI filmmaking tools while critically engaging with the future directions of AI-augmented creativity.

This workshop focuses on emerging approaches for using Artificial Intelligence (AI) systems to support and augment personal creativity. Recent developments in generative Machine Learning demonstrate the ability of AI systems to perform tasks which are often associated with creativity - generating imagery, composing music, writing prose, etc. This workshop will examine opportunities for incorporating this kind of functionality into the creative practice of designers, artists and craftspeople, in practical and experimental ways. It will focus on how AI might enhance, rather than supplant, individual human creativity, through collaboration, serendipity, and creative reflection. We seek to engage a broad range of creative practitioners and researchers, bringing together those already using AI in their practice with those who are new to the technology, to understand emerging approaches and define future opportunities.

Through a process of robust co-design, we created a bespoke accessible survey platform to explore the role of co-researchers with learning disabilities (LDs) in research design and analysis. A team of co-researchers used this system to create an online survey to challenge public understanding of LDs [3]. Here, we describe and evaluate the process of remotely co-analyzing the survey data across 30 meetings in a research team consisting of academics and non-academics with diverse abilities amid new COVID-19 lockdown challenges. Based on survey data with >1,500 responses, we first co-analyzed demographics using graphs and art & design approaches. Next, co-researchers co-analyzed the output of machine learning-based structural topic modelling (STM) applied to open-ended text responses. We derived an efficient five-steps STM co-analysis process for creative, inclusive, and critical engagement of data by co-researchers. Co-researchers observed that by trying to understand and impact public opinion, their own perspectives also changed.

In this paper, we propose VR Rehearse & Perform - a Virtual Reality application for enhancing the rehearsal efforts of performers by providing them access to accurate recreations - both visual and acoustical - of iconic concert venues.

Public attitudes towards learning disabilities (LDs) are generally reported as positive, inclusive and empathetic. However, these findings do not reflect the lived experiences of people with LDs. To shed light on this disparity, a team of co-researchers with LDs created the first online survey to challenge public understanding of LDs, asking questions in ways that are important to them and represent how they see themselves. Here, we describe and evaluate the process of creating an accessible survey platform and an online survey in a research team consisting of academic and non-academic professionals with and without LDs or autism. Through this inclusive research process, the co-designed survey met the expectations of the co-researchers and was well-received by the initial survey respondents. We reflect on the co-researchers' perspectives following the study completion, and consider the difficulties and advantages we encountered deploying such approaches and their potential implications on future survey data analysis.

Creative computing is an approach to computing education which emphasises the creation of interactive audiovisual software and an art-school influenced pedagogy. Given this emphasis on Dewey's "learning by doing", we set out to investigate the processes students use to develop their programs. We refer to these processes as the students' 'coding behaviour', and we expect that understanding it will provide us with valuable information about how students learn in our creative computing classes. As existing metrics were not sufficient, we introduce a new set of quantitative metrics to describe coding behaviours. The metrics consider factors such as students' vocabulary use and development, how fast and how much they alter the functionality of code over time and how they iterate on their code through text insert and delete operations. Many of our lessons involve providing students with demonstrator code which they use as a base for the development of their programs, so we use demo code as an entry point to our dataset. We look at programs students have written through developing the demo code in a dataset of over 16,000 programs. We clustered the demo code using the set of descriptive metrics. This lead to a set of clusters containing programs which are associated with distinct coding behaviours. Four was the ideal number of clusters for cluster density and separation. We found that the clusters had distinct behaviour patterns, that they were associated with different instructors and that they contained demo programs with different lengths.

For the last decade, there has been growing interest in the STEAM approach (essentially combining methods and practices in arts, humanities and social sciences into STEM teaching and research) to develop better research and education, and enable us to produce students who can work most effectively in the current and developing market-place. However, despite this interest, there seems to be little quantitative evidence of the true power of STEAM learning, especially describing how it compares and performs with respect to more established approaches. To address this, we present a comparative, quantitative study of two distinct approaches to teaching programming, one based on STEAM (with an open-ended inquiry-based approach), the other based on a more traditional, non-STEAM approach (where constrained problems are set and solved). Our key results evidence how students exhibit different styles of programming in different types of lessons and, crucially, that students who tend to exhibit more of the style of programming observed in our STEAM lessons also tend to achieve higher grades. We present our claims through a range of visualisations and statistical validations which clearly show the significance of the results, despite the small scale of the study. We believe that this work provides clear evidence for the advantages of STEAM over non-STEAM, and provides a strong theoretical and technological framework for future, larger studies.

Digital approaches to physical rehabilitation are becoming increasingly common and embedding these new technologies within a musical framework may be particularly motivating. The current feasibility study aimed to test if digital musical instruments (DMIs) could aid in the self-management of stroke rehabilitation in the home, focusing on seated forward reach movements of the upper limb. Participants (n=3), all at least 11 months post stroke, participated in 15 researcher-led music making sessions over a 5 week intervention period. The sessions involved them 'drumming' to the beat of self-chosen tunes using bespoke digital drum pads that were synced wirelessly to an iPad App and triggered percussion sounds as feedback. They were encouraged to continue these exercises when the researcher was not present. The results showed significant levels of self-management and significant increases in functional measures with some evidence for transfer into tasks of daily living.

We have applied interactive machine learning (IML) to the creation and customisation of gesturally controlled musical interfaces in six workshops with people with learning and physical disabilities. Our observations and discussions with participants demonstrate the utility of IML as a tool for participatory design of accessible interfaces. This work has also led to a better understanding of challenges in end-user training of learning models, of how people develop personalised interaction strategies with different types of pre-trained interfaces, and of how properties of control spaces and input devices influence people's customisation strategies and engagement with instruments. This work has also uncovered similarities between the musical goals and practices of disabled people and those of expert musicians.

We investigate an approach to the artistic stylization of photographic images and videos that uses an understanding of the role of abstract representations in art and perception. We first learn a database of representations from a corpus of images or image sequences. Using this database, our approach synthesizes a target image or video by matching geometric representations in the target to the closest matches in the database based on their shape and color similarity. We show how changing a few parameters of the synthesis process can result in stylizations that represent aesthetics associated with Impressionist, Cubist, and Abstract Expressionist paintings. As the stylization process is fast enough to work in real-time, our approach can also be used to learn and synthesize the same camera image, even aggregating the database with each new video frame in real-time, a process we call "Memory Mosaicing". Finally, we report the user feedback of 21 participants using an augmented reality version of "Memory Mosaicing" in an installation called Augmented Reality Hallucinations, where the target scene and database came from a camera mounted on augmented reality goggles. This information was collected during an exhibition of 15,000 participants at the Digital Design Weekend at the Victoria and Albert Museum (co-located during the London Design Festival).