A collaboration between ID Atelier and OMNI Labs: a concept campaign reimagining a Hermès-grade equestrian editorial — "Le Défilé des Chevaux Persans". The original scarf design — the horse motifs and the full carré composition — is the work of İdil Kıbaroğlu; OMNI Labs carried that artwork into an in-house synthetic model, scarf styling, product, and a cinematic hero.

Every frame was generated, directed, retouched, and finished through the studio pipeline: identity locked across shots, wardrobe and props art-directed, and a moving-image hero produced from the same design language. No external shoot, no licensing — a complete brand-grade campaign from artwork to film.

A concept campaign built around Aesop's Resurrection Aromatique Hand Wash — the everyday ritual treated as craft. Warm, tactile, restrained: brass dishes, low light, the gesture of the hand reaching for the pump.

Stills and motion were directed and produced entirely in-house through the studio's AI production pipeline, then finished frame-by-frame to brand standard. The hero is a 10-second cinematic close-up — product, light, and ritual, no shoot required.

No client, no brief: a self-initiated proof that OMNI Labs can stand up a complete brand world from a single idea. OMNI DEATH is a Liquid-Death-grade provocation for still & sparkling water — "murdered from a mountain, canned in silence" — with a gold-on-black skull system, a death-to-plastic manifesto, and a cult mythology around a 500ml ration called The Big Void.

Every layer was produced in-house: the can label system, three product SKUs rendered as clean packaging photography, two original synthetic faces cast and locked across editorial and motion, and a campaign film — all from artwork to finished frame, no external shoot.

No client, no brief: a self-initiated proof that OMNI Labs can stand up a complete fashion brand — and ship its storefront — from a single idea. ÉTAT is the anti-streetwear position: a premium tee "recognized by the cut, not the logo." The whole world is built backwards from the logo — no chest print to sell, no seasonal hook to chase, only the garment, resolved until it stands on its construction alone. Carbon, Bone, Dust. No seasons, no drops. Made once, kept.

Every layer was produced in-house: the brand doctrine and mark system, a 21-shot garment line across three colorways, a synthetic flagship face cast and identity-locked across editorial and motion, a thirteen-slide launch carousel, an 83-second designed brand film, and a complete e-commerce site — coded, deployed, and live. Artwork to storefront, no external shoot.

Built in vanilla TypeScript + Three.js, no framework bloat: four pages, GPU shaders running live in the browser, every effect scroll-gated so only what's on-screen renders. The brand world from the concept case, turned into something you move through instead of watch.

The hero bends a branded texture around a gravitational mass; a second page drops you onto a real-time 3D ocean tuned to a dead, cold void palette; product spins on a drag-to-rotate turntable; and the cult pages run live video loops inside member dossiers. Premium dark-metal direction held consistent across every surface.

THE DESCENT is an endless tunnel-faller in vanilla TypeScript + Three.js. The fantasy is the brand's thesis turned into a mechanic: plastic is the enemy, so the walls you dodge are glowing plastic, and the further you fall the faster they come. Score is depth. Hit a wall and you drown.

The craft is in the game-feel. A real combo system rewards skimming wall edges — near-miss for gold sparks, a slow-mo kiss, and a multiplier that scales every metre but collapses if you play it safe. The tunnel itself is alive: procedurally-drawn skull and slogan engravings scroll past you, and the whole world shifts colour every depth zone. Every hazard and particle is object-pooled so the RAF loop never allocates — a locked 60fps with bloom. Audio is fully synthesized WebAudio (zero asset files) to stay inside the site's strict CSP. The pure-logic core ships with a Vitest suite that stays green.

Most "3D hero" sites ship a baked video or a heavy GLB that paints late and just sits there. We used Spline instead: a real-time interactive scene streamed into the page, reacting to the cursor and lit live in the browser. The chrome agent reads as a physical object you could reach out and turn — the centrepiece of a Web3 product that needed to feel inevitable, not speculative.

The craft is in the choreography. The Spline scene is wired to scroll: at the top the robot is fully present, holding the eye against the headline; as you descend, parallax pulls it down and out of frame while the live transaction feed, the stat block and the marketplace sections rise to take its place. The 3D doesn't loop — it performs a hand-off, so the page tells a story instead of decorating one. Tuned so the scene stays light enough to hold a fast first paint and a smooth scroll on a marketing landing page.

The whole site is hand-built in vanilla HTML, CSS and JavaScript — no framework, no build step, served by a tiny zero-dependency Node server. Swiss grotesk typography, an aluminium-and-white base, and a flavour-accent colour per mood give it a clean editorial feel with restrained scroll motion. Three pages — home, flavours, and a B2B "Pro" portal — share one nav, one design system and a full Turkish/English toggle.

Underneath it runs a custom image CMS. Every photo on the site is a drag-and-drop image-slot web component, and a purpose-built admin panel lets the client swap, reframe or remove any of 31 image and video slots — writing overrides straight to disk, ready to commit and publish. The brand can keep the site fresh without ever touching code.

This is a concept spec piece — an imaginary luxury campaign produced to show what an OMNI Labs synthetic film looks like end to end. No shoot, no crew, no location. One original synthetic character held identical across eight shots, one product held identical alongside her, and a designed motion layer that turns raw clips into an actual ad film instead of an AI montage.

The point isn't the model — it's the direction. The image-to-video was generated on an inexpensive, near-zero-cost video model: deliberately cheap, because the craft is in everything wrapped around it. Identity locking across frames, a verdict mechanic built for the feed, kinetic typography cut to the beat, and a full original soundtrack — a first-person narration in the archivist's own voice, scored under a custom audio mix. Cheap generation, expensive taste.

The source is the orbit footage above — a helix flight that rises while it circles, so the camera sees every façade from multiple elevations. That coverage is the whole trick: a flat circle leaves the vertical surfaces ambiguous, but a helix gives the solver the parallax it needs to resolve window lancets, cornices and brickwork. We pull sharp frames from the flight, then run them through a classical Structure-from-Motion pass.

COLMAP solved all 407 of 407 cameras — a clean 100% registration, every frame placed in 3D space with the sparse point cloud to match. That posed dataset feeds Brush, a WGPU Gaussian-splat trainer, which optimises millions of splats into a continuous radiance field you can fly through. The result is a walkable world reconstructed from a real place, rendered in a Three.js viewer — the third local splat rail in the OMNI Labs creative-tech stack.

The spine is a scroll-scrub frame sequence — hundreds of pre-rendered stills mapped one-to-one to scroll position, so the page scrubs a film frame by frame as you move. Three separate sequences carry the story: the hero junction turn, the engine teardown, and the night-run finale. Every frame is a fixed still, but the scroll makes it motion — the same technique behind the highest-end product launch pages, rebuilt from scratch as a standalone site with no build step.

The centrepiece is the engine disassemble: the car's V8 lifts apart into a fully exploded teardown, every intake, header and pulley separating in mid-air. That sequence was authored with an AI image-to-video model driven by start-and-end keyframe interpolation — one frame shows the assembled engine, another the fully exploded state, and the model resolves the disassembly in between. The result is scrubbed on scroll like the rest, so the viewer teardown-and-rebuilds the engine with their own hand on the wheel.

This is a concept spec piece — an imaginary broadcast collaboration produced to show what an OMNI Labs macro nature film looks like end to end. No lens, no garden, no crew. Every wingbeat, every pollen grain, every iridescent vein in that translucent wing was generated, not filmed — then directed like a real natural-history segment: species cards, a consistent on-screen information system, and a closing brand card that plays it straight, the way a real broadcast credit would.

The craft is in the restraint. The image-to-video was produced on an inexpensive AI video pipeline and finished with a 2× local upscale — deliberately cheap, because the direction is where the budget actually goes. A kinetic broadcast-UI layer holds identity and framing steady across the cut; a cinematic narrator reads the moth's Latin cue and a fictional "Leica 35mm APO-Summicron" credit line in the same breath a real wildlife documentary would; and underneath it all, a layered mix — a scored bed, a garden's worth of ambience, and the voiceover ducked cleanly above both — does the quiet work of making a generated reel feel like it was actually recorded somewhere.