How the Generator works.
How Toko turns layered parts into a collection of unique, reproducible tokens — and why it never touches rarity or score.
The Generator is the machine in the corner of the workshop. You give it layered artwork and a set of rules, and it assembles hundreds of unique tokens. Its job is the art — what each token looks like. It deliberately leaves rarity, scoring and supply to the collection.
The building blocks: layers and parts
A generator is a stack of ordered layers (background, body, clothing, accessories…), and each layer holds parts — the individual images that can fill it. To build a token, the Generator picks one part from each layer and stacks them up.
always or only optionally with a probability — so not every token needs an accessory. Parts within a layer must have unique names.Steering the mix: weighting and caps
Each part carries a weight that controls how often it's chosen. Higher weight, more frequent. You can use simple presets or custom numbers:
Weighting only changes how often a part appears in the artwork. It is not how rare the finished token is — that's the collection's separate rarity policy. Watch the 25-second explainer →
Keeping it sensible: compatibility rules
Some combinations don't make sense — a snorkel with a spacesuit, maybe. Compatibility rules let a part require or exclude another part, or a whole layer. The Generator validates your rules up front, flagging contradictions, self-references, missing targets, unreachable selections, and duplicate part names, so a batch can't get stuck.
The assembly loop
For every token, the Generator runs the same loop, rejecting anything invalid until it has the number of valid, unique tokens you asked for:
Deterministic and replayable
The Generator's output is a pure function of its inputs. The same configuration, seed and source data always reproduce the exact same tokens — and if any input changes, the output changes.
Three ways to run a batch
- Limited number — generate a set count of unique tokens.
- Single token — generate one, handy for testing a setup.
- From a CSV — download a template, fill in the exact parts you want, upload and validate.
Curate, pin, export
After a run you preview the results, pin the keepers, and export the pinned set into a collection. That export is the only path from the Generator into a collection — there's no separate import wizard. Exported tokens arrive as Draft with their composition and provenance attached, and a collection links to a single generator source, so its history stays clean.
The Generator makes the art and nothing else. It doesn't assign rarity, attribute desirability, scores, or collection-wide scarcity — those are the collection's job. Keeping the two apart is what lets you re-roll art without disturbing the rules buyers rely on.
In short
- Layers + parts: the Generator picks one part per layer and stacks them into a unique token.
- Weighting sets how often a part appears (presets 1–8, default Balanced 3); caps limit a part per batch.
- Compatibility rules keep nonsensical combinations out; duplicates are rejected.
- Deterministic: same config + seed + inputs → identical tokens, with a locked snapshot for audit.
- It makes art, not rules — export pins into a collection as Draft; rarity and score are set there.