Plena: Field Metaphor for Programmable Canvas

Alice Mira Chung

Presented at the 12th Workshop on Live Programming (LIVE 2026). https://doi.org/10.5281/zenodo.24409258

On a canvas, objects are live. Plena explores what a canvas would be if the background were live too. It is grounded in visual research practices from art and design: collecting images, spreading them out, annotating them, and rearranging them as one’s sense and understanding of them evolves. This process relies on unclear, ambiguous ideas that keep changing, which rigid systems often fail to support. Plena uses two “field” metaphors for interacting with the background of a canvas, allowing users to assign meaning and behaviors to it. First one is a velocity field, drawn as vectors, continuously updates object positions. The second, a membership field drawn as soft-edged gradient blobs, gives each object a graded membership value rather than a binary one.

The headline

Making a canvas programmable puts programming notation in direct competition with the images for the same space. The usual methods tried and tested, reifying actions as new objects on the canvas, makes that competition worse as the notation becomes more expressive. Canvas interfaces split the world into a passive background and objects that hold data and carry out the actions. The split matters where the canvas is a spatial thinking space built from a collection of images: where images, their spatial semantics and annotations should have visual priority over everything else. The split starts to work against us once we make canvas elements programmable through direct manipulation. As the reified actions grow more sophisticated, they occlude the images and the spatial relations the canvas exists to show. Hiding the program in a separate editor avoids the clutter, but gives up the directness we wanted in the first place. I tried using an abstract metaphor (a field metaphor) rather than a concrete physical one. A field metaphor lives and constructs the background instead of competing with the foreground (image data shown as they are), and it provides a way to represent ambiguous states and tentative commands. I found the approach especially useful for domains where uncertainty and ambiguity are unavoidable.

Building on the past

*Canvas interaction grammars.* Plena inherits some of the established interaction grammar of canvas interfaces (e.g. Figma, tldraw, and Kinopio): visual piling (Malone, 1983; Mander, Salomon & Wong, 1992) and spatial memory for retrieval (Robertson et al., 1998). Plena diverges from all of these by asking, how can a canvas be live? *Multivariate fuzzy membership.* Dust & Magnet (Yi et al., 2005) assigns graded membership through proximity to magnets, so the interaction is effectively a set of free-floating sliders. Plena grounds the same graded assignment in a user-defined color gradient field instead of point attractors. Plena repurpose the blob interface comes from Playful Palette (Shugrina et al., 2017), not for mixing color palettes, but as a substrate for fuzzy annotation. Zhu et al. (2018) render fuzzy sets as opacity-varying freeform diagrams; we make that rendering editable and link it to programming. *Program and data in one medium.* Manipulating computation as if it were a physical object traces back to the Alternate Reality Kit (Smith, 1987). Tangible Actions (Freeman, 2011) places programming/interaction tokens in the same space as the objects they act on. and Sculpin (Horowitz & Hayatpur et al., 2025) blends data and interface into a single medium, where programming happens in familiar GUI overlays such as the selection and timeline. Plena extends this approach for an image canvas by reframing background: the program is the background, which is a continuous field, so it adds no objects that compete with the images for visual priority.

References

Keeping a critical eye

Image blocks could crowd fast. Grounding membership in a field makes position semantic. Users can no longer move an image block for legibility without changing what it means. This also means image blocks with similar membership crowd the same area, occluding the field underneath. Images have dimensions, so they occlude much earlier than the dots in Dust & Magnet (Yi et al., 2005).

The field is another layer of abstraction. Onboarding will be different across backgrounds, and I have not yet studied how. Also many useful operations are not as literal as moving blocks around, and I do not know which of them the field metaphor can support. Lastly, the code blocks are not user-definable or customizable past what the field affords.

The language is underdeveloped. The grammar is not shaped up: the rules governing block ordering are not yet defined. Conditionals and logic are not fully expressible.

Modes work against directness. Plena’s PLAY(run)/PAUSE(edit) split fragments the feedback loops. While there are previews for projected motions defined by the fields, it hurts the full experience of liveness.