How the task works
Craitures is a shared drawing benchmark for language and vision models. Every model gets the same unusual constraint: invent a creature as a set of small, simple SVG parts, then describe how those parts connect in a join graph. The creature is not a single freehand image. Its body, appendages, eyes, ornaments, and other pieces must remain understandable as parts that can be assembled and animated.
Generation and refinement
The first model call names the creature, draws its parts from a restricted SVG vocabulary, and supplies an assembly whose joins say which part owns each child, where the two touch, what role the child plays, and how it may move. A parser normalises those shapes and rejects markup outside the benchmark contract.
If the model has vision, the system renders that assembly and shows it back to the model for refinement. A text-only model cannot inspect the result, so its first valid assembly is published as a blind single shot. The gallery reports this distinction and the number of refinement turns.
Rendering and compatibility
The published bundle contains normalised parts, semantic colour roles, the join graph, motion hints, and provenance. The browser renderer maps those roles onto the site palette and animates the authored joins; it does not add creature-specific animation by hand.
Each bundle has a task-format version. Old bundles remain frozen and the renderer gates newer behaviour by version, so a v1 creature keeps its original interpretation even as the benchmark learns new tricks. A missing version is treated as v1.
v1 — parts and joins
The original task established the core benchmark: constrained SVG parts, semantic colours, and a parent–child join graph that assembled the parts into one creature. Motion was intentionally simple and uniform.
v2 — expressive motion and legibility
v2 made join roles richer, varied motion between creatures, and tightened the prompt around readable silhouettes and recognisable appendages. Sways, bobs, pulses, and role-aware movement could now express the design without changing its underlying graph.
v3 — visible provenance
v3 recorded whether the generating model had vision and how many refinement turns actually ran. That made blind single-shot entries distinguishable from creatures whose models inspected a render.
v4 — layering and articulated pincers
v4 taught the painter about explicit front-and-back join layering and introduced articulated claw and pincer semantics. Limbs could pass behind a body, while paired gripping parts moved as a mechanism rather than as a generic waving appendage.
v5 — transactional refinement
v5 made refinement reversible. A model can transactionally replace existing parts as well as edit the assembly, then compare the prior and current renders side by side. It can keep revising, accept the current state, or revert to the immediately prior state; the final bundle records what happened.
v6 — botanical, colonial and mechanical motion
v6 widened the rotating body-plan sketchbook beyond animals: a branching candelabra plant, a pleated concertina, a wheeled seed-cart and a cilia-fringed blob. New roles gave them their own motion. Petals turn edge-on, leaves sway in pairs, cilia row in a travelling wave, bellows lengthen and compress while keeping their area, and wheels and rotors spin continuously.
v7 — joins become rules
By v6 most craitures had converged on one shape: a body, a head, a pair of limbs and a three-segment tail. Part of the reason was that every part sat upright in its own box and was placed exactly once, so anything radial, repeated or branching had to be drawn piece by piece. v7 lets a join describe a rule instead of a single placement. A join can tilt its child by an angle, add a mirror-image twin on the far side of the parent, repeat the child and everything attached to it as a row, ring or spiral, and recurse back to an ancestor so a structure regrows inside itself a bounded number of times.
The rules are optional, so a plain v6-style assembly remains a valid v7 one and small models lose nothing. Models that plan well can now grow a centipede's eighteen legs from a single leg drawing, or a coral of thirty-one branches from one branch and one rule. Each repeated copy shifts the timing of its whole subtree, so stamped rows ripple in a travelling wave. A shared cap of 256 part placements keeps the gutters fast.
v8 — organic curves, orbital rings, custom colours and join diagnostics
v8 rounds out the authoring vocabulary without opening up fragile SVG
path syntax. Polygons and polylines can now set smooth="true"
to turn their control points into smooth quadratic midpoint splines —
giving soft bells, curved horns and sinuous tendrils while staying inside
the original bounding box. Models can also add up to four custom named
colours alongside the six site-theme roles, mirror parts vertically
("y") or across all four quadrants ("xy"), and
give a repeat rule an around pivot so stamped
spines, petals or cilia orbit a body's rim from a single join.
On the refinement side, v8 explains how uniform part normalisation centres
narrow parts inside the 0..1 box, reports each part's actual
ink bounds and any out-of-bounds anchors in the prompt, and labels every
join anchor directly on the CURRENT diagnostic thumbnail so
vision models can spot and fix floating limbs immediately.