A comic-book apparel brand for developers - 889 original designs, a search that takes a feeling instead of a keyword, and a pipeline that makes new artwork a build step rather than a budget line.
The Aimito.
Say it out loud: daj mi to.
In English it reads "the AI-mito." In Croatian it's "gimme that" - the exact sound an impulse buy makes. The AI angle and the purchase intent, hiding in the same four syllables. The buy button on every product just says Gimme that.
The site is English-only. The Croatian is saved as a reveal on the About page, not sprinkled through the UI.
Developers are one of the most identity-driven, meme-fluent and well-paid groups on the internet. They already buy clothing to signal who they are - conference tees, company swag, "it's always DNS" mugs. The trigger is humour plus belonging, and it repeats.
Nobody owns the curated lane: original art, a brand world, and a way to find the thing you meant.
"AI slop." Vibe coding. The confidently wrong model. An entire comedic seam opened up in the last two years and nobody had mined it properly - which is why two of the ten squads are AI Slop and AI Voices.
Every design is original artwork, which turns the marketplaces' structural weakness - takedowns and trademark risk - into a moat instead of a threat.
Anyone can put a joke on a shirt. The defensible part is the loop underneath: the marginal cost of a new design is compute, and so is the marginal cost of a new advert. Competitors buy content as a line item. Here it's a build step.
A local generation stack produces artwork in batches - no per-image API bill, no external rights questions, and the freedom to throw most of it away.
A quality and diversity pass prunes hard. Generation is cheap, so the catalogue is defined by what gets cut - the same lesson the game taught us about taste being the scarce input.
A promo pipeline turns any design into a vertical hero clip starring a recurring caped character, with the deterministic branding pieces composited identically every time.
Short-form video and the places the tribe actually lives. A bootstrapped brand can't buy attention, so the plan is built around content that costs compute rather than budget.
Vibe search turns a described feeling into a product page. Print-on-demand fulfilment means no inventory, no upfront capital and worldwide shipping from day one.
Which jokes convert decides what gets generated next. The loop closes, and the catalogue starts learning the audience instead of guessing at it.
The interesting problems weren't "make an image." They were everything that has to be true before a generated picture becomes a garment somebody is happy to receive.
Type "3am on-call despair" or "npm chaos" and get the shirt you meant. The query and the artwork live in one shared embedding space, so text finds images directly - no keyword tagging, no manual taxonomy across 889 designs. Product pages use the same vectors in reverse for visually similar neighbours.
A multimodal model running on local hardware reviews designs at publish time - reading back the text, catching botched background removal, and flagging anything not print-worthy. No cloud call, no per-image cost.
Diffusion models are bad at legible text and precise layout, and a joke shirt is mostly legible text and precise layout. So the recurring formats are built by deterministic generators instead: certification badges with curved ring text, syntax-highlighted code panels, OS error dialogs, captcha grids, nutrition-facts labels, flowcharts, tonal ASCII art. The model supplies illustration; code supplies typography.
The exact printable area of each garment was measured by printing a calibration pattern through the fulfiller's own mockup generator and reading the result back - not estimated.
A pre-launch sweep of all 889 designs found 65 with no print-resolution file - every one of which would have handed the printer a dead link on a real order. All 65 were regenerated and upscaled before anyone could buy them.
Every product photo is the real garment with the real print, not a glossy fake. It's the brand's trust promise: what you see is what ships.
Pale artwork disappears on a cream tee. Rules decide the garment colours each design is offered on, and a colour-adaptation step retunes two-tone text so it survives the switch.
The storefront is live, the catalogue is full, and the purchase path - checkout, tax, fulfilment, confirmation email - is wired end to end against real providers.
What's left is the part no pipeline can automate: distribution. Getting the first hundred people to find it, wear it, and tell someone. That's the honest state of it, and it's the phase the whole engine was built to make survivable on a bootstrapped budget.
The Aimito is the commercial version of a thesis we keep testing: when generation gets cheap, the advantage moves to whoever built the machinery around it - the curation pass, the deterministic composition, the automated quality check, the loop that feeds sales data back into what gets made next. A brand that manufactures its own catalogue and its own marketing is just the clearest way to demonstrate that. The same engine works for anyone whose growth is throttled by how much content they can afford to produce.
If your growth is limited by how much content you can afford to make, that's a pipeline problem, not a budget problem. Go look at the store first - then let's talk about yours.