A chance-operations score generator for human-AI co-creative practice — specifically for iterative dialogue between autographic gesture and AI image generation (Stable Diffusion image-to-image workflows). Each cast of two eight-sided dice produces a composite score across three sheets: theoretical-affective vocabulary (Language), operational image directives (Image), and temporal-durational instructions (Process). The score is not a prescription. It is a field that the artist enters and completes through practice.

The instrument draws from three vocabulary sheets on each cast. Each sheet column has a coloured dot beside its header — this dot indicates the sheet is active and can be toggled on or off by clicking the header. Toggling a sheet off excludes it from the cast entirely: no items are selected from it, it does not contribute to the score card, and its mark field dims. This allows you to work with any combination of one, two, or all three sheets — for example, using only Language and Process without Image directives, or casting from a single sheet for a more focused score.

The dot behaves differently depending on when you toggle. Before a cast: toggling a sheet off dims the entire column — the dot darkens, the mark field fades, all items go dark. The sheet reads as simply inactive. After a cast (settled): toggling a sheet off dims the column, but the activated items were already illuminated by the cast. They dim back toward dormant, and the dot dims — but there is a residual warmth from the settled state that makes the dimming look different from a pre-cast toggle. This is correct: the two states are genuinely different. In both cases, the next cast will exclude a toggled-off sheet regardless of when it was toggled.

The Score column (IV) has no dot because it has no toggle — it is the output of the cast, not a pool to draw from.

I. Language — Theoretical and affective fragments drawn directly from the research framework of neural collage practice: Deleuzian vocabulary (intensive difference), embodied cognition (Varela), and terms coined within the practice itself (refound object, alreadymade, the almost-known). Three items are activated per cast. They function as conceptual-affective orientation for the session — not as tasks.

II. Image — Operational directives derived inductively from actual studio practice: real procedural moves in the iterative exchange between hand and machine. Three items are activated per cast. The compositional diagram (placement principle, rotation, scale) is an additional spatial score generated from the same throw. It is a suggestion. Human obstinacy prevails.

III. Process — Temporal and durational scores drawing on Cage's process logic and autoethnographic observation of how the human-AI session unfolds in time. Two items are activated per cast. They govern the session as a whole: sequencing, duration, the decision of when to stop.

Each of the three sheet columns is headed by a generated field of marks — lines, arcs, curves, and rectangles. These are not decorative. They are produced by the same deterministic seed as the cast: the same sequence of numbers that selects the vocabulary items also determines the position, weight, and opacity of every mark in the field. The visual composition and the textual score are one event seen in two registers simultaneously.

The marks are causally related to the cast but not symbolically related to the I Ching. There is no iconographic correspondence between a hexagram and a particular mark type or configuration — the relationship is structural, not interpretive. Think of it as the cast expressed spatially rather than linguistically: the same throw, read differently.

When a sheet is inactive (toggled off by clicking its header), its mark field fades. When active, it illuminates at full opacity. The mark field is the sheet's visual identity — it changes with every cast, making each score visually distinct as well as textually distinct. After a cast, dormant (unselected) vocabulary items dim back and the mark field settles — the score asserts itself against the pool from which it was drawn.

Two eight-sided dice are cast — digitally or entered from physical dice. Each die face maps to one of the eight I Ching trigrams (☰ Heaven through ☷ Earth). Their combination yields one of 64 hexagrams — the complete set of the I Ching. The hexagram number deterministically seeds all score selections: the same throw always produces the same score, making results citable and reproducible as research data. This follows John Cage's use of the I Ching from the early 1950s — not as randomness, but as a fixed yet impersonal mapping that displaces personal preference without abandoning structure.

Structurally referential, not directly derived. Cage's original consists of ten transparent sheets — carrying lines and points — overlaid in any combination, with the performer reading intersections as instructions. This instrument preserves three elements from that logic: multiple simultaneous sheets (Language, Image, Process are overlaid, not sequential); chance operations as compositional method; and the score as a generative field rather than a fixed instruction set. It departs from Cage's graphic overlay method and the open indeterminacy of performer interpretation — the vocabulary here is fixed, the selections deterministic, the directives textual rather than graphical.

Following Umberto Eco's Opera Aperta (1962), this score is conceived as an open work: a structured field of possibility that the artist completes through the act of practice. The score does not determine the work. It opens a space in which the work becomes possible. Each of the 64 hexagrams is a different opening of that space.

A curated set of source images that advances sequentially through a pre-session shuffle, so image selection is partially determined by sequence rather than taste. To use: collect your images into a folder, then either drag all files onto the drop zone or click browse files and select them (use Shift+click or Cmd+A to select multiple files). The bank shuffles automatically on load. Each cast reveals the next image in the sequence alongside the score directives — shown in full so the complete image is visible.

The bank controls:

hide / count — toggles the image counter (e.g. "14 remaining") on or off. Use hide during presentation if you prefer not to foreground how many images remain.

reshuffle — re-randomises the current image set and resets the pointer back to the beginning. The same images remain loaded but in a new random order. Use this to start a new sequence without reloading. Note: the reshuffle is recorded in the performance log with a timestamp.

clear bank — unloads all images and returns to the drop zone, so you can load a completely different set of images. Use this to switch to a different image folder mid-session. The previous sequence is gone but the performance record is unaffected.

Note: images are not stored across browser close — if you close and reopen the file you will need to reload your images. The performance log and session record are preserved.

The SD Prompt Directive shown at the bottom of the score card is a condensed composite drawn from the cast selections — a subset of the subset. One item is selected at random from each sheet's activated pool (one from the Language selections, one from the Image selections, one from the Process selections), joined by em-dashes with a closing period. Unlike the score itself — which is fully determined by the throw and always reproducible — the prompt selection is genuinely open: the same hexagram cast twice will produce the same score directives but potentially a different prompt each time. Up to 18 different prompts are possible from any single cast (3 Language × 3 Image × 2 Process selections). The prompt is not the score — it is a variable secondary layer, a way of speaking the score to the machine. The score is the oracle's word; the prompt is how you choose to address the algorithm with it. It can be ignored, modified, or used as written. The word "algorithm" may be read as referring to any generative transformation process — not exclusively Stable Diffusion.

The score card displays selected items in the order they were drawn by the seeded RNG — the oracle's sequence, which is the sequence of the cast as a temporal event. The vocabulary pools in the three sheet columns display items in their fixed structural order (top to bottom as written). These will differ. The oracle's sequence and the structural sequence are two different readings of the same cast. A toggle between the two views is planned for a future version.

Each sheet is planned for expansion to exactly 64 items — one per hexagram — enabling direct indexical mapping (hexagram number → specific item) and structural coherence with the I Ching's complete 64-hexagram system. At present, items are selected by seeded shuffle from each sheet's vocabulary. The physical dice entry preserves the talisman quality of the throw as the primary event; the tool reads the oracle.