Stability AI

Combines a strict isometric viewpoint, clean PBR-style materials, and embedded text in one tightly framed scene. It tests whether a model holds a consistent 45-degree perspective, renders the labels accurately, and keeps the composition centered and uncluttered.

A logo and branding test. It checks accurate text (the accented Caffè and Est. 1720), a clean vector-emblem look rather than a photographic one, and a cohesive vintage style. Small text and crisp flat shapes are common weak spots for image models.

This competition tests how well AI image models truly understand language versus how much they rely on visual habits from their training data. The prompt is deliberately simple on the surface but devilishly hard in practice. Most models default to the familiar trope of an astronaut riding a horse. By forcing the reversal, we measure three critical capabilities that separate good models from great ones:

• Strict instruction following (including negations)
• Accurate subject-object relationships and spatial hierarchy
• Resistance to strong dataset biases

This challenge forces models to use one consistent handwritten style across an entire dense menu instead of defaulting to clean printed text for the smaller details, a very common failure that reveals how well they actually understand and maintain stylistic coherence.

A character-portrait test focused on fine materials. It checks how well a model renders engraved metal under warm torchlight, layered textures (leather, cloth, braided hair), and lifelike eyes and skin without drifting into a plastic, over-smoothed look.

This challenge seems to be difficult for models because it mixes reality with fiction. Most models struggle to keep the taxi realistic or loose instructions like placing the passenger not in the backseat.

A layout and typography test. The model has to produce a clean, professional menu with legible text, a clear three-section grid, and food imagery that looks intentional, which separates real design sense from a default template look.

A structured infographic test. The model must place six labeled steps in order, hold a consistent flat-vector icon style, and render several short labels (step names, crew names, Tranquility) accurately. It rewards layout discipline and punishes garbled text.