The 16:9 problem
Most AI image models render 1:1 or 16:9. The web requires 1.91:1 for OG cards, 9:16 for vertical, 4:5 for IG, 3:1 for banner. Notes on living between aspect ratios.
Every shipping piece in the catalog has to live across at least five aspect ratios.
The native render — whatever the model produced. The gallery preview — the cover thumbnail. The OG card — 1.91:1, for X and Discord and Slack. The Instagram crop — 4:5 portrait. The Twitter banner — 3:1. The mobile vertical — 9:16, for Stories and Reels. The piece detail page — full-bleed but capped by viewport.
The catalog cannot afford to make each piece for each format. The piece is made once. Every other format is a crop of the same piece. The art is held responsible for being legible after a crop the artist may not see.
What the constraint forces
Compositions that survive cropping.
A composition that puts the subject dead-center is forgiving of crop. A composition with the subject pushed to a corner is fragile — most crops will lose part of the subject. A composition that depends on the bottom-edge text strip is going to lose that strip on the 4:5 IG crop.
The catalog's discipline, learned over years of cross-platform posting, is to put the subject inside the inner 4:3 box of any composition. Whatever is essential to the piece sits in roughly the middle 75% of the frame. The outer 25% is allowed to be background, atmosphere, or negative space. When the crop happens — IG, OG, banner — the outer 25% is what gets sacrificed.
What the constraint takes away
Edge-to-edge compositions. A piece where the figure stretches from the absolute left edge to the absolute right edge cannot crop to 4:5 without amputating. A piece where the figure's head touches the top edge cannot crop to 1.91:1 without decapitating.
The catalog has shipped a small number of edge-to-edge pieces and learned that they only work as gallery detail views. Their OG card is broken. Their IG share is broken. They are good pieces that live only on the page. That trade-off is sometimes worth it. Usually it is not.
The native render bias
Most AI models are biased toward square. Soul_2, nano_banana_2, the diffusion stack at large — all default to 1:1 or close to it. 16:9 is available but not native. 9:16 vertical is more available than it used to be but still produces weaker compositions than square.
That bias is at war with the web's bias. The web is overwhelmingly horizontal — desktop browsers, OG cards, banners. The mobile web is overwhelmingly vertical — Instagram, TikTok, Stories. The desktop and mobile worlds want opposite aspect ratios, and the AI model wants neither.
The compromise the catalog has settled on: render at the closest native aspect ratio that gives the best subject quality. Then crop intentionally — generate the canonical 4:5 IG crop and the 1.91:1 OG crop from the same source, knowing each will look different from the gallery detail view.
The dialect, applied
Cover compositions push the subject into the inner 4:3 box and let the outer space hold atmosphere, type, or chrome. Portrait compositions center the head in the upper third of the frame so the OG card crop (which loses the bottom half of a portrait) keeps the face.
Backgrounds are designed to be tileable in both directions — a violet gradient that bleeds in from one side reads the same way whether the crop loses 200 pixels from the top or 200 from the bottom. Wordmarks and type, where they appear, sit inside the inner 16:9 box of any composition. They never touch the outer margin.
The single rule: compose for the smallest aspect ratio you'll ship to. Everything else is a more generous crop of the same composition.
What the 16:9 problem teaches
That the aspect ratio is part of the piece. A composition designed without knowing where it will land will only land in one place. A composition designed with crop discipline ships everywhere and looks intentional everywhere.
The constraint is the brief. Most artists fight it. The catalog has decided to use it.