Pixelated
Kisses↘

Overview

Pixelated Kisses is a projection-mapping and motion experiment inspired by Joji’s song of the same name.

I used my own artwork as the visual input for RunwayML, then generated movement and transformations from those images before editing the results into a projected sequence. I was interested in seeing how work I had already created could be reinterpreted through a generative system without completely losing the visual language it began with.

The project was not about asking AI to produce an aesthetic for me from nothing. It was about treating my existing artwork as source material, then deciding what transformations were useful, what still felt connected to my work and what needed to be rejected or edited.

Context

This project began as an experiment with bringing generative video into a projection-mapping workflow.

I used artwork I had already made as the starting point because I wanted the visual identity of the piece to come from my own practice rather than from a generic text prompt or an imitation of someone else’s style. RunwayML became one stage within a larger process that also included visual direction, selection, compositing, editing and projection.

Pixelated Kisses felt appropriate because the song already carries a tension between intimacy and digital distance. The title suggests affection being filtered through a screen, fragmented into pixels or changed by technology. Using my own artwork and passing it through an AI system created a similar tension between something personal and the way a machine interprets, distorts or reconstructs it.

Process

I began with my own illustrations and visual experiments rather than generating the project entirely from text prompts. These images gave the sequence its original palette, shapes, textures and emotional direction.

I imported the artwork into RunwayML and experimented with image-to-video generation, using prompts mainly to describe movement, atmosphere and transformation. I tested how the system extended or distorted elements that already existed in the source images.

The results varied heavily. Some generations preserved enough of the original artwork to feel like an extension of it, while others replaced its character with generic imagery or changed the composition too aggressively. I selected only the moments that still felt connected to my visual language.

This selection process was important because I did not want the generated output to determine the project by itself. I treated it as unstable raw material. I brought the clips into After Effects, where I controlled the pacing, layering, color, scale and relationship to the music.

Once the sequence was edited, I adapted it for projection mapping. This required thinking beyond the rectangular frame and considering how contrast, movement and scale would behave on a physical surface.

Position on AI

I am not interested in using AI as a replacement for artists, designers or the process of developing a visual language.

I am especially critical of the destructive, unsustainable and exploitative ways generative AI is often developed and used. These systems can depend on enormous amounts of energy and water, datasets built without meaningful consent and labor practices that remain hidden behind the appearance of automation. They are also frequently used to imitate artists, reduce creative work to a prompt or avoid paying people for their skills.

Because of that, I approached AI cautiously in this project. I used my own artwork as the source material and treated the generated results as one experimental stage within a process I continued to direct and edit.

This does not remove the wider ethical and environmental problems surrounding the technology. Instead, the project gave me a way to examine what a more limited and intentional use of it might look like, while remaining aware that experimentation with these tools is not neutral.

Reflection

This project taught me that generative tools do not remove the need for authorship or decision-making. The process still depended on what I chose to input, which outputs I rejected, how I edited the material and how the final visuals were placed into space.

Using my own artwork helped the piece retain a connection to my practice, but it also showed me how quickly a generative system can flatten a specific visual language into something more generic. The strongest results were not necessarily the most technically impressive. They were the ones where I could still recognize the source image while seeing it behave in a way I could not have animated as quickly on my own.

At the same time, the project made me more critical of when and why I use AI. I do not want it to become a default step in my work simply because it is available. Its environmental cost, unclear data practices and effect on creative labor make that kind of casual use difficult to justify.

If I continued the project, I would keep the use of generative tools limited and purposeful. I would also document the source artwork and transformation process more clearly so that the final piece does not obscure where the imagery came from or imply that the system created the visual direction independently.