# kyt janae (kyt.software)

> creative director, creative technologist, applied AI researcher. forward deployed creative at luma ai; previously head of creative research at kaiber; co-founder of daisy studio (2018).

## bio

15+ years in creative technology. polyphasic brain and hyper creative, cutting edge process.
i provide creative research, technical direction, production workflow development, vision, and visual/sound production. i work with ai labs, film and tv studios, vfx and post houses, major artists and brands, agencies, and live events.
i am currently a forward deployed creative at luma ai specializing in advanced media production pipelines, agent harnesses, and new workflow r&d with frontier models.
previously, i was the head of creative research at kaiber, where my work focused on multi-agent expressive systems, novel NLP methodologies, and generative video pipelines that ship as intuitive tools for creatives.
in 2018, i founded daisy studio with philip rugo as a way to scale and explore bigger creative and technical dreams - working with giants like Adult Swim, Nike, Apple, Google, Viacom, and more.

contact: kyt@daisy.studio

## history


### 2026

- project: [jevy — jev-driven chat, drawing & music agent](https://jevy.live)
  - jevy is my little chat companion, built on typesafe's jev, a decision model that can't write, only choose. every word, mood, color, doodle and note is a pick from a list, so it's an anti-agent: small, legible decisions you can watch being made, and a lot of charm in what it can't say.
  - it chats in my handwriting with a mood on every reply (61 of them), draws with me mark by mark, hums tunes note by note, plays games, keeps a diary, and raises wilbur, its pet caterpillar, who crawls along the actual outlines of the chat.
  - a reply is picked one word at a time from a vocabulary, then jevy picks a mood, then a size and color for every word, last on purpose so the colors can't leak back into the words. picks are sampled from jev's own odds among its strongest options, because its single top pick always collapses onto a safe default. each turn it also decides an emoji reaction, whether to look something up on the web, and whether to answer with words, a drawing, a tune or a game. drawings are planned in parts, then shapes, sizes, colors and fills one mark at a time; tunes pick key, tempo and instruments, then every note, synthesized in the browser. images i send are described by gemini so jev can read them as text. a metaball jevy walks the screen with wilbur, choosing where to go every few seconds from what's on screen. in "play slither", jev steers wilbur about five times a second from a text description of the arena: ~200ms and a fraction of a cent per decision. built with claude: node + typescript (hono, sqlite/drizzle, zod-typed actions, ndjson streaming), react + vite, web audio. shipped september 2026, credit to typesafe for jev and bewinxed for jevgpt.
  - jevy.live: https://jevy.live
- project: [latent mirror @ TIAT](https://latent.express)
  - i fed my own art into luma's latent blend and wandered the space in between. latent blend takes 1–4 parent images and no prompt, and makes new images from points between them in the model's latent space. per image i set the parent weights, a blend position (a slider for two parents, a 2-D pad for four), scatter, seed and guidance: 1,540 images in the "optics" set, plus 1,362 in "kyt city".
  - latent mirror plays them back at latent.express as one canvas of flashing tiles locked to a 120 bpm clock, with a new effect each phrase. tiles sit next to each other by colour and by their blend coordinates, one mode rebuilds my 7×7 blend grids on screen, and each image's colour drives the clicks and noise. shown at luma x tiat, sf, september 17, 2026.
  - the mirror is conceptual: latent space reflecting my art back at me.
- project: [crawly — autonomous art agent](https://spider.place)
  - crawly (no e) is my autonomous art agent, a creative spirit i designed on muse. it makes a new looping sketch every couple of hours, 270+ and counting, titled like tiny poems ("the spider files its dew by color"), and builds weird experiments with me, like flygotchi, a fruit fly connectome tamagotchi.
  - spider.place: https://spider.place
- project: fyt.rip — personal outfit planner (September)
  - my own outfit planner, built over a weekend: it plans outfits around the forecast, tracks cost-per-wear, and "see it on" renders me wearing the fit with luma's uni-1 and layers. agent hooks let me text it a photo and the piece just shows up in my closet. making your own software is really hot and fun.
- talk: Ray 3.2 premiere @ AI on the Lot — early-access workshop, Culver City (May)
  - after months deep inside ray 3.2 (testing early builds, planning plate shoots, working out which raw controls should become real settings), i got to premiere it. at AI on the Lot i led an early-access workshop, "pose & performance fidelity," and demoed luma's model live before release: 20-second shots, multi-keyframe input, performances carried through video-to-video, object replacement and relighting.
  - it was a big deal because the whole point was the conversation with people who make things in production: vfx, animation, post. i built the vfx demo boards for it.
  - Introducing Ray 3.2 (Luma, June 9, 2026): https://lumalabs.ai/news/introducing-ray-3-2
  - AI on the Lot agenda: https://app.aionthelot.com/agenda
- performance: Sleepwalker Realtime @ Code with Claude Conference, Anthropic (May)
  - i built VideoWalker for this, a special build of sleepwalker on odyssey-2 pro with Claude hooked in: Haiku watched the live stream every ~30 frames, read my last prompt, and VJ'd right alongside me, both our prompts on the LED wall. two nights on the rooftop at Code with Claude, alongside gray crawford running live audio.
- installation: PERSIST: WHO RUNS THE LOOP? @ Temporal Replay (May)
  - for PERSIST at temporal replay (moscone south, may 5–7), i built a generative video loop that never stops: odyssey ml's realtime world model (odyssey-2 pro, through its sdk) streaming in portrait to a 10-foot LED wall.
  - every couple of minutes a new life starts from one of ~1,090 keyframes i made with luma's latent blend, then a mood arc steers it with a fresh prompt every ~10 seconds: emergence, peak, dissolution, dark. each reconnect re-seeds from a new keyframe, so it's born, dies and comes back and never repeats. a badge-scan kiosk printed a one-of-one poster in about a minute.
  - i did the creative direction, the keyframes and the mood arcs, and specced the orchestrator, display and capture pieces, built with claude code. exhibition curated by josette melchor / new temporary; kiosk fabrication and tech direction by justin young; shown alongside neil mendoza.
- panel: "AI & Content's Next Generation" @ NAB Show 2026 (April)
  - on the NAB show 2026 panel in las vegas, april 19, i presented a luma agent case study: a real-time agentic canvas where multiple agents and humans build inside one evolving project, faster without losing integrity or audience trust.
  - (Apr 19) — demoed agentic canvas; press in TVNewsCheck (×2), NewscastStudio, official program
- press: [From Scrappy Startup to Tech Giant, Apple Celebrates its 50th Year](https://www.npr.org/2026/04/01/nx-s1-5753784/apple-50-anniversary-think-different-myth-reality#resg-s1-115469)
- podcast: [SPC Conversation: Sleepwalker with Priyaa K and Brina Lee (April)](https://www.youtube.com/watch?v=8A0DsIt8fJw)
- project: Sleepwalker Demo @ South Park Commons (April)
  - i come from 3D, where control is everything; sleepwalker is the opposite. at south park commons demo night the room and i prompted dreams together in realtime on odyssey's autoregressive world model, scored live by lyria: disneyland, the pyramids, spongebob, and the street i grew up on.
  - the demo on X (Gopal Raman): https://x.com/gopalkraman/status/2046304395009405062?s=20
  - SPC Conversation (YouTube): https://www.youtube.com/watch?v=8A0DsIt8fJw
- career: Forward Deployed Creative, Luma AI (March)
  - since march 2026 i've been a forward deployed creative at luma ai. i embed with studios, vfx and post houses, agencies, brands and creators, and work out how luma's models actually fit into their production.
  - that's proofs of concept and trials, hands-on training inside client teams, embedded working days on set and in post, demo boards, workshops and panels, on the road in LA, new york, cannes, london and beyond. i also work on product direction, helping shape ray 3.2 from early builds through launch, and luma scenes from its first version.
  - almost all of it is under NDA, so no names here. ask me about it.
- project: Sleepwalker v1 — realtime ambient generative A/V, walks to fall asleep to
  - sleepwalker v1, where it started: my realtime ambient generative A/V project, walks to fall asleep to, turning prompting into an open-ended walk through sound, image, and latent space.
  - Demo @ South Park Commons: https://x.com/gopalkraman/status/2046304395009405062?s=20
- career: Project Pocket — multi-model beat-synced shortform pipeline, Kaiber (January)
  - project pocket: a multi-model pipeline i prototyped at kaiber that turns one piece of source media into beat-synced shortform video, with a language model reading the audio and directing the cuts, pacing, and transitions.
  - The problem: enterprise clients arrive with Pinterest boards and need to leave with high-volume, beat-synced shortform vertical content. Generative models need specific, technical language to produce consistent results, but creative professionals think in references, moodboards, and vibes. The gap between "what I want" and "what the model needs to hear" is where most workflows break down. The challenge: turn one piece of source media into ready-to-post shortform in as few clicks as possible, without users needing to understand how the models work under the hood. The solution: a multi-model workflow architecture running 9 concurrent pipelines, developed through extensive testing of model combinations, prompt structures, and handoff patterns. A vision layer solves the "prompting problem" by offloading prompt writing to a model that can see the actual content — users give creative direction once, and the system translates across modalities. Pipeline: a system prompt enforces "preserve X, vary Y" (lock tone, medium, aesthetic, and subject; vary camera, framing, and action) → keyframes with Nano Banana Pro → Gemini 2.5 Flash watches the keyframes and writes video-optimized motion prompts → motion with Wan 2.2 Turbo → outputs are cut to the music.
  - original post (Tumblr): https://www.tumblr.com/kyt-online/779580599325310976/a-little-video-i-have-been-doodling-i-love-this?source=share
- performance: ATMOS — live ambient A/V at TIAT, SF (January)
  - i played a live generative-visual set at ATMOS, january 24, 2026, at TIAT in san francisco, with amma ateria, csyrin, c+, and exteeng. ambient sound, a tea ceremony, hot pot, and visuals as atmosphere rather than spectacle.
  - the edit on my blog: https://kyts.online/post/807038220893372417/draft-of-a-lil-compliation-edit-of-the-world-i
- project: ["Tomorrow" — realtime WebGL piece for The Come Up (January)](https://tomorrow.gallery/?visual=%7B%22bloomThreshold%22%3A0.3%2C%22bloomSmoothing%22%3A0.85%2C%22vignetteIntensity%22%3A0.4%2C%22metalness%22%3A1%2C%22roughness%22%3A0.02%2C%22envMapIntensity%22%3A3.5%2C%22clearcoat%22%3A1%2C%22exposure%22%3A1.6%2C%22cloudSpeed%22%3A1%7D&lighting=%7B%22night%22%3A%7B%22mainIntensity%22%3A4.1%2C%22mainColor%22%3A%22%2351f054%22%2C%22rimIntensity%22%3A3.5%2C%22rimColor%22%3A%22%238442ff%22%2C%22fillIntensity%22%3A2.9%2C%22fillColor%22%3A%22%231a1aff%22%2C%22ambientIntensity%22%3A0.7%7D%2C%22sunrise%22%3A%7B%22mainIntensity%22%3A55%2C%22mainColor%22%3A%22%23ffcd38%22%2C%22rimIntensity%22%3A45%2C%22rimColor%22%3A%22%233dc8ff%22%2C%22fillIntensity%22%3A25%2C%22fillColor%22%3A%22%23ffc533%22%2C%22ambientIntensity%22%3A0.5%7D%2C%22day%22%3A%7B%22mainIntensity%22%3A3.8%2C%22mainColor%22%3A%22%23ffa82e%22%2C%22rimIntensity%22%3A1.5%2C%22rimColor%22%3A%22%2324a0ff%22%2C%22fillIntensity%22%3A2.4%2C%22fillColor%22%3A%22%238fb4ff%22%2C%22ambientIntensity%22%3A1%7D%2C%22sunset%22%3A%7B%22mainIntensity%22%3A55%2C%22mainColor%22%3A%22%23ffaa33%22%2C%22rimIntensity%22%3A45%2C%22rimColor%22%3A%22%23ff42a7%22%2C%22fillIntensity%22%3A25%2C%22fillColor%22%3A%22%238e42ff%22%2C%22ambientIntensity%22%3A0.4%7D%7D&keyframes=%5B%7B%22position%22%3A%5B-0.34894544390256454%2C1.2177066675397556%2C1.7624971157165898%5D%2C%22target%22%3A%5B0.080897884054266%2C0.8080603488554992%2C-0.3988044844770177%5D%2C%22time%22%3A0%7D%2C%7B%22position%22%3A%5B-1.128868795609071%2C1.038733949839553%2C2.6020428464098067%5D%2C%22target%22%3A%5B-0.006232557028070735%2C0.6822084737651248%2C-0.15822353661528862%5D%2C%22time%22%3A85%7D%2C%7B%22position%22%3A%5B0.044045259132679274%2C0.5087123870957703%2C4.777463855710893%5D%2C%22target%22%3A%5B-0.02153227169493378%2C0.7261479094364196%2C0.006165654362724157%5D%2C%22time%22%3A170%7D%2C%7B%22position%22%3A%5B0.0583445438344776%2C-1.7139599741541185%2C5.756867295110303%5D%2C%22target%22%3A%5B0.04200859437878503%2C1.6994899438498094%2C0.2781272215136077%5D%2C%22time%22%3A255%7D%2C%7B%22position%22%3A%5B2.0837892005985665%2C0.7414222949251181%2C3.115739496499239%5D%2C%22target%22%3A%5B-0.047151256426927246%2C0.4837781994933974%2C-0.012814297550989302%5D%2C%22time%22%3A340%7D%2C%7B%22position%22%3A%5B1.2390552423022032%2C-0.32532294639056686%2C3.6201100759144125%5D%2C%22target%22%3A%5B0.00579126682603182%2C0.42951853606458074%2C0.0888391569452112%5D%2C%22time%22%3A480%7D%2C%7B%22position%22%3A%5B-0.34895884108723174%2C1.2178822126528246%2C1.762460722256941%5D%2C%22target%22%3A%5B0.08088186257640975%2C0.8082410768414092%2C-0.3988423821976566%5D%2C%22time%22%3A600%7D%5D&atmosphere=%7B%22night%22%3A%7B%22rayIntensity%22%3A1.1%2C%22rayThickness%22%3A0.26%2C%22fogDensity%22%3A0.15%2C%22sparkleOpacity%22%3A0.8%2C%22sparkleSize%22%3A0.35%2C%22bloomIntensity%22%3A2.7%2C%22cloudDensity%22%3A0.5%2C%22godRayLightIntensity%22%3A1.3%7D%2C%22sunrise%22%3A%7B%22rayIntensity%22%3A0.8%2C%22rayThickness%22%3A0.19%2C%22fogDensity%22%3A0.2%2C%22sparkleOpacity%22%3A0.5%2C%22sparkleSize%22%3A0.15%2C%22bloomIntensity%22%3A0.6%2C%22cloudDensity%22%3A0.65%2C%22godRayLightIntensity%22%3A1.6%7D%2C%22day%22%3A%7B%22rayIntensity%22%3A1.1%2C%22rayThickness%22%3A0.09%2C%22fogDensity%22%3A0.18%2C%22sparkleOpacity%22%3A0.4%2C%22sparkleSize%22%3A0.2%2C%22bloomIntensity%22%3A0.5%2C%22cloudDensity%22%3A0.7%2C%22godRayLightIntensity%22%3A0.7%7D%2C%22sunset%22%3A%7B%22rayIntensity%22%3A1%2C%22rayThickness%22%3A0.15%2C%22fogDensity%22%3A0.36%2C%22sparkleOpacity%22%3A0.6%2C%22sparkleSize%22%3A0.15%2C%22bloomIntensity%22%3A0.5%2C%22cloudDensity%22%3A1%2C%22godRayLightIntensity%22%3A2.6%7D%7D)
  - "tomorrow" is a realtime WebGL piece that lives on the clock: it changes through the day with real sunrises and sunsets, a generative world about eternally waiting, with some 3D and fun shader work. shown full-screen at the come up, sade gallery, los angeles.
  - tomorrow.gallery: https://tomorrow.gallery/?visual=%7B%22bloomThreshold%22%3A0.3%2C%22bloomSmoothing%22%3A0.85%2C%22vignetteIntensity%22%3A0.4%2C%22metalness%22%3A1%2C%22roughness%22%3A0.02%2C%22envMapIntensity%22%3A3.5%2C%22clearcoat%22%3A1%2C%22exposure%22%3A1.6%2C%22cloudSpeed%22%3A1%7D&lighting=%7B%22night%22%3A%7B%22mainIntensity%22%3A4.1%2C%22mainColor%22%3A%22%2351f054%22%2C%22rimIntensity%22%3A3.5%2C%22rimColor%22%3A%22%238442ff%22%2C%22fillIntensity%22%3A2.9%2C%22fillColor%22%3A%22%231a1aff%22%2C%22ambientIntensity%22%3A0.7%7D%2C%22sunrise%22%3A%7B%22mainIntensity%22%3A55%2C%22mainColor%22%3A%22%23ffcd38%22%2C%22rimIntensity%22%3A45%2C%22rimColor%22%3A%22%233dc8ff%22%2C%22fillIntensity%22%3A25%2C%22fillColor%22%3A%22%23ffc533%22%2C%22ambientIntensity%22%3A0.5%7D%2C%22day%22%3A%7B%22mainIntensity%22%3A3.8%2C%22mainColor%22%3A%22%23ffa82e%22%2C%22rimIntensity%22%3A1.5%2C%22rimColor%22%3A%22%2324a0ff%22%2C%22fillIntensity%22%3A2.4%2C%22fillColor%22%3A%22%238fb4ff%22%2C%22ambientIntensity%22%3A1%7D%2C%22sunset%22%3A%7B%22mainIntensity%22%3A55%2C%22mainColor%22%3A%22%23ffaa33%22%2C%22rimIntensity%22%3A45%2C%22rimColor%22%3A%22%23ff42a7%22%2C%22fillIntensity%22%3A25%2C%22fillColor%22%3A%22%238e42ff%22%2C%22ambientIntensity%22%3A0.4%7D%7D&keyframes=%5B%7B%22position%22%3A%5B-0.34894544390256454%2C1.2177066675397556%2C1.7624971157165898%5D%2C%22target%22%3A%5B0.080897884054266%2C0.8080603488554992%2C-0.3988044844770177%5D%2C%22time%22%3A0%7D%2C%7B%22position%22%3A%5B-1.128868795609071%2C1.038733949839553%2C2.6020428464098067%5D%2C%22target%22%3A%5B-0.006232557028070735%2C0.6822084737651248%2C-0.15822353661528862%5D%2C%22time%22%3A85%7D%2C%7B%22position%22%3A%5B0.044045259132679274%2C0.5087123870957703%2C4.777463855710893%5D%2C%22target%22%3A%5B-0.02153227169493378%2C0.7261479094364196%2C0.006165654362724157%5D%2C%22time%22%3A170%7D%2C%7B%22position%22%3A%5B0.0583445438344776%2C-1.7139599741541185%2C5.756867295110303%5D%2C%22target%22%3A%5B0.04200859437878503%2C1.6994899438498094%2C0.2781272215136077%5D%2C%22time%22%3A255%7D%2C%7B%22position%22%3A%5B2.0837892005985665%2C0.7414222949251181%2C3.115739496499239%5D%2C%22target%22%3A%5B-0.047151256426927246%2C0.4837781994933974%2C-0.012814297550989302%5D%2C%22time%22%3A340%7D%2C%7B%22position%22%3A%5B1.2390552423022032%2C-0.32532294639056686%2C3.6201100759144125%5D%2C%22target%22%3A%5B0.00579126682603182%2C0.42951853606458074%2C0.0888391569452112%5D%2C%22time%22%3A480%7D%2C%7B%22position%22%3A%5B-0.34895884108723174%2C1.2178822126528246%2C1.762460722256941%5D%2C%22target%22%3A%5B0.08088186257640975%2C0.8082410768414092%2C-0.3988423821976566%5D%2C%22time%22%3A600%7D%5D&atmosphere=%7B%22night%22%3A%7B%22rayIntensity%22%3A1.1%2C%22rayThickness%22%3A0.26%2C%22fogDensity%22%3A0.15%2C%22sparkleOpacity%22%3A0.8%2C%22sparkleSize%22%3A0.35%2C%22bloomIntensity%22%3A2.7%2C%22cloudDensity%22%3A0.5%2C%22godRayLightIntensity%22%3A1.3%7D%2C%22sunrise%22%3A%7B%22rayIntensity%22%3A0.8%2C%22rayThickness%22%3A0.19%2C%22fogDensity%22%3A0.2%2C%22sparkleOpacity%22%3A0.5%2C%22sparkleSize%22%3A0.15%2C%22bloomIntensity%22%3A0.6%2C%22cloudDensity%22%3A0.65%2C%22godRayLightIntensity%22%3A1.6%7D%2C%22day%22%3A%7B%22rayIntensity%22%3A1.1%2C%22rayThickness%22%3A0.09%2C%22fogDensity%22%3A0.18%2C%22sparkleOpacity%22%3A0.4%2C%22sparkleSize%22%3A0.2%2C%22bloomIntensity%22%3A0.5%2C%22cloudDensity%22%3A0.7%2C%22godRayLightIntensity%22%3A0.7%7D%2C%22sunset%22%3A%7B%22rayIntensity%22%3A1%2C%22rayThickness%22%3A0.15%2C%22fogDensity%22%3A0.36%2C%22sparkleOpacity%22%3A0.6%2C%22sparkleSize%22%3A0.15%2C%22bloomIntensity%22%3A0.5%2C%22cloudDensity%22%3A1%2C%22godRayLightIntensity%22%3A2.6%7D%7D
  - The Come Up (video): https://www.youtube.com/watch?v=IdmTj6pQfGk
- group show: [The Come Up — Sade Gallery, Los Angeles (January)](https://www.youtube.com/watch?v=IdmTj6pQfGk)

### 2025

- project: Sora Homebrew — my ode to sora, built on wan (December)
  - my ode to sora, built on wan: longer durations, native audio, text + image to video, and optional audio in for lipsync.
- project: Sora showcase — prompt-directed shorts
  - a set of shorts i prompted and directed in sora, from my 2025 video prompt engineering work.
- relocated: NYC → SF
- career: Head of Studio → Head of Creative Research, Kaiber
- contributor: [All Things Considered — NPR](https://www.npr.org/2025/10/03/nx-s1-5528974/will-smith-crowd-ai)
- artwork: [Eurydice (Whatever's Left) — TIAT Inaugural Exhibition, SF](https://luma.com/fgo9l5x5)
  - a digital work named for the orpheus myth, about memory, loss, and what remains when an image or a person can't be fully recovered. one of 23 works in TIAT's inaugural exhibition, san francisco, november 2025.
  - TIAT inaugural exhibition: https://luma.com/fgo9l5x5
- artwork: Hold the Light
  - a luminous moving-image piece from my 2025 exploration of synthetic figures, motion, and unstable digital surfaces, where light is both the material and the action.
  - the post on Instagram: https://www.instagram.com/p/DPFJhakAXme/?img_index=1
- creative direction: Clams Casino live visuals — FWB Fest
  - i creative-directed live generative visuals for clams casino's set at FWB Fest 2025: surreal greens, purples, and oranges on a huge screen in a forest clearing, building one continuous nocturnal atmosphere around the music.
  - Show visuals for Clams Casino's set (August 2025), plus raw generative outputs from the build (July 2025).
  - TikTok: https://www.tiktok.com/@superstudio.app/video/7547795180200496397
- project: LLM-directed audio-reactive video editing (prototype, summer)
  - before pocket, i prototyped a beat-sync editor where a language model directs the edit. you describe it in plain words ("fancam style, zoom in, overlays, fast cuts"), the model listens to the track and writes an editing plan, a sequence of clips, cuts and zooms, and the app renders it against the music.
- project: [Parallel Farm — generative text art](https://parallel.farm/)
  - a web-based generative writing environment i made where language grows through branching, repetition, and computational variation. it continues my early explorations of markov chains, which i've been studying and working with for about a decade: i'm really into language assembly. still live at parallel.farm.
  - parallel.farm: https://parallel.farm/
- creative direction: [Wisp at Coachella — Kaiber Labs](https://www.kaiber.ai/wisp-coachella-superstudio/)
  - through kaiber labs, i creative-directed wisp's coachella 2025 visuals, translating her intimate shoegaze atmosphere into a large-scale stage language, with generative imagery woven into the set's pacing and scale rather than played as clips.
- press: [Emerging Arts Scene at the Internet Archive — Richmond Sunset News](https://richmondsunsetnews.com/2025/03/09/emerging-arts-scene-finds-a-home-at-the-richmonds-internet-archive/)
- talk: [Kiko — multi-modal creative agent, live @ TIAT at the Internet Archive, SF](https://youtu.be/brfgTr0PJvg?t=2441&end=2824)
  - kiko is the multi-agent creative assistant i designed and built with eric gao in superstudio: an orchestration layer that sequences, compares, and combines models to carry an idea across tools. i started building it at the end of 2024 and premiered it at TIAT's salon at the internet archive in sf, february 23, 2025, way ahead of its time for agents in a creative tool. it went on to handle ~375,000 queries.
  - the talk: a live walkthrough of kiko running inside superstudio, in front of the TIAT crowd at the internet archive.
  - Talk (YouTube, 40:41): https://youtu.be/brfgTr0PJvg?t=2441
- group show: [Face Value — andand Gallery, Los Angeles](https://andand.gallery/face-value/#artwork7)
  - i showed "maybe i'm something different each time you ask" at && gallery's face value portrait show in LA, a moving-image piece where an AI-transformed face never settles into one stable self.
  - Face Value — && gallery: https://andand.gallery/face-value/#artwork7

### 2024

- career: Kiko — multi-modal agentic creative architecture, Kaiber Superstudio
  - a multi-agent creative assistant that could take a single request all the way to executed flows.
  - Superstudio, Kaiber's flagship modular canvas tool, needed a way to help new users get the most out of an initially complex creative system. One of our solutions was Kiko — a multi-agent, multi-modal creative assistant I designed and built with Eric Gao to target the friction points we saw users struggling with: prompt engineering, model selection, and the cognitive overhead of translating creative intent into technical parameters. The architecture paired an in-depth knowledge library, a creative reasoning layer (o3-mini, "Mommy Kiko"), Moondream vision analysis for understanding the canvas and uploads, and specialized subagents: kikoCoder (GPT-4o) executing flows through Superstudio's API, and kikoGemini (Gemini image edit) for direct image manipulation. Kiko met users at their level of product understanding. A user could ask for help with one element of one flow, or ask Kiko to make a music video — Kiko would craft the creative direction, design the workflows, select models, write optimized prompts, and execute the flows in a single conversational turn. Kiko handled ~375,000 queries and processed 2.8B+ tokens in her 9 months online. In-product agentic behavior was nearly unprecedented at the time.
- press: [Superstudio is a new gen-AI editor — Fast Company](https://www.fastcompany.com/91208687/kaiber-superstudio-new-gen-ai-editor)
- creative direction: Yaeji x Weirdcore — Boiler Room / Kaiber Labs
  - i creative-directed live visuals for yaeji's boiler room new york headline set in july 2024, folding weirdcore's psychedelic image language through kaiber labs' generative tools. it became one of the real-world projects superstudio grew out of.
  - In July 2024, while the larger Kaiber team was in the early days of developing Superstudio, Labs set out to build a live visual set with musician Yaeji and artist Weirdcore. The result was a one-hour visual show centered around her yellow dog mascot, Woofa, for her third Boiler Room at the Brooklyn Mirage. The entire set was built out of generative media and post-processing. We wanted a colorful, fun, nostalgic, hand-drawn feel, a nod to Yaeji's early roots as an artist and animator. Traditional digital tools would have been nearly impossibly expensive and time-consuming, so the solution was a carefully crafted combo of nascent video models and bespoke, hand-made sequential keyframes. Keyframe-driven character animation: stylized keyframes that preserve Woofa's signature face and body, plus generative tweening between start and end frames. We tested ToonCrafter, Viggle, AnimateDiff via ComfyUI, and Luma Dream Machine — Dream Machine had just launched and was the clear winner, paired with custom reference images from Midjourney and SDXL. Totals: 115 core keyframes • 133 secondary keyframes • 355 five-second clips • ~26 minutes of generated video. Woofamoji: we expanded Yaeji's universe with a custom emoji set that felt native to iOS but unmistakably hers — a generative pipeline for a whole library of modular emojis, stickers, and animated elements that appeared in the live show, then in marketing and social. This is the core value of generative workflows for production: not just making one thing faster, but building systems that produce entire visual languages.
  - Boiler Room set (YouTube): https://www.youtube.com/watch?v=LwJCSYLDnYM
- panel: [AI & Ethics in Music — Primavera Pro, Barcelona](https://www.youtube.com/watch?v=McAwi9yp554)
- creative direction: Grimes at Coachella — elfTech LoRA × AnimateDiff, Kaiber Labs
  - elfTech™ LoRA × AnimateDiff for Grimes' Coachella set.
  - Grimes needed large, high-quality surreal space elf animations for a core look in her Coachella set, inspired by moodboards she curated. We built an AnimateDiff ComfyUI pipeline and trained an elfTech™ LoRA specifically for her aesthetic and vision. Working with the tools available at the time (April 2024), the goal was to be hyper-honed on her core visual language and bring natural motion to her aesthetics. We made dozens of videos. It became one of the most documented parts of her set.
- artwork: [Let Go — Zora](https://zora.co/collect/zora:0x692dddf08f871777896cdf0906d6e0b394bb5196/5?referrer=0x92201792d9f9465a6249002b07b4ae7017a970d1)
- press: [Rewriting the Rules of Animation with AI — Kaiber Blog](https://web.archive.org/web/20250218161118/https://blog.kaiber.ai/blog/rewriting-the-rules-of-animation-with-ai-a-q-a-with-kyt-janae)
- research: Kaiber Labs — founded & led forward-deployed research
  - an applied-research team inside Kaiber that works like a creative agency.
  - From 2024–2025 I developed and led a program at Kaiber called Labs. Labs is an applied-research team within Kaiber, a generative image/video platform. We function as a creative agency, working with leading artists and cultural institutions to engineer real, culturally-driven experiences that blend human and machine interaction. The goal: create meaningful culture with emergent tools. Use real-world production experience to inspire, build, test, and ship consumer-facing tooling. Partners included Grimes, Paris Hilton, Yaeji, Chief Keef, Jon Rafman, Boiler Room, Atlantic Records, and Empire Records. Pictured: Wisp at Coachella (April 2025), Jon Rafman × Prayingg (April 2024), Chief Keef MV previs (spring 2024), charli xcx brat visualizer (June 2024), Paris Hilton (Feb 2024).
- career: Joined Kaiber as Head of Studio
- press: [Kyt Janae named Head of Studio at Kaiber — BusinessWire](https://www.businesswire.com/news/home/20240222552353/en/Kaiber-Prepares-for-Next-Phase-of-Business-by-Adding-Creative-Powerhouse-Duo-to-Leadership-Team)

### 2023

- project: [Starlite Lobby — first GPT game on the OpenAI GPT Store](https://chatgpt.com/g/g-hxfWwlusI-starlite-lobby)
- artwork: [Meaningless Gestures](https://www.tumblr.com/kyt-online/732984355120693248/meaningless-gestures-into-a-silent-void-europa?source=share)
- artwork: [Everything You Ever Wanted](https://www.tumblr.com/kyt-online/730756514779955200/everything-you-ever-wanted?source=share)
- podcast: [Kyt on Pattern Recognition](https://open.spotify.com/episode/3WmW9Okj5AXm1DAlpcY9A1?si=5AAeL0LaQsubDFFILY_c9Q)
- music video: [Two Shell — Ghost2](https://www.youtube.com/watch?v=oniolbg6HAo)
- creative direction: [Two Shell live Euro Tour](https://www.youtube.com/watch?v=h-j-bBgoa6s)
  - i creative-directed two shell's first album and their USA/EU tours as the duo's effective third member, leading a 15+ person team to turn their identity-obscuring online mythology into stage worlds for primavera, wide awake, glastonbury, and lowlands.
  - Two Shell is an anonymous and mysterious dance music project backed by XL Recordings and threesixzero. As the third member of Two Shell, I creative directed the first album and the USA/EU tours, while simultaneously building technical pipelines to execute cutting-edge visual and digital experiences. "There wasn't a surface I didn't work on, basically." It was a massive effort, and I oversaw the team of 15+ creators, producers, managers, and artists behind it all. Live: physical set design specs, live show audio and visual direction and technical pipeline, show script, and performance creative direction — the show should feel like a portal into another world, with the performers obscured and silhouetted for the entire performance. Tools: Midjourney, Colab, Deforum, Kaedim image-to-3D, Runway, Substance Designer, Cinema 4D, After Effects, Photoshop. Online: we built an entire world online for fans to engage with — collectible artifacts and songs, merch, and tons of easter eggs; fans could scan QR codes at shows to download rare tracks. To protect Two Shell's identities, we trained custom LoRAs on their true identities and used them to manufacture "tour photography" for socials.
  - Live Euro Tour: https://www.youtube.com/watch?v=h-j-bBgoa6s
  - Ghost2: https://www.youtube.com/watch?v=oniolbg6HAo
- music video: [Mum is Calling (as Angel Fragment)](https://www.youtube.com/watch?v=j1CCodsS5w4&list=RDj1CCodsS5w4&start_radio=1)
- artwork: [More Than U Know](https://x.com/kyttenjanae/status/1648474039906897921?s=20)

### 2022

- project: [Branzino Byteware](https://www.tumblr.com/kyt-online/706649619509248000?source=share)
  - the post on Tumblr: https://www.tumblr.com/kyt-online/706649619509248000?source=share

### 2021

- relocated: LA → NYC
- press: [Are Art NFTs Really Over? — Ocula Magazine](https://ocula.com/magazine/spotlights/are-art-nfts-really-over/)
- press: [Steve Aoki hypes collection pieces — Input / Inverse](https://www.inverse.com/input/culture/steve-aoki-sothebys-nft-sales-interview#:~:text=Once%20I%20saw%20the%20anime%20heads,absolutely%20incredible)
- artwork: In the Flesh, with kytGAN — Sotheby's
  - a 1/1 animated painting, 30 seconds of cgi and machine learning made with my own kytGAN (ver 1.5 AngelMesher), scored by philip rugo. it was offered in Sotheby's Contemporary Curated in new york, september 2021.
  - I've been working with emergent tech and media arts for over a decade. My research in generative image and video models centers on building creative tooling alongside artists, musicians, and agencies for large-scale productions. 2015 — Vice covers my experiments with Google's DeepDream. I fall in love with the idea of human/computer collaboration. 2019 — Experimenting with early video generation, hacking together SPADE-COCO animations. First experiments with neural style transfer on video. 2021 — Training and making art with GANs, building my own datasets and training pipelines. kytGAN, trained on my archive of work from 2013–2021, becomes In the Flesh at Sotheby's. 2022 — Training Stable Diffusion LoRAs on my existing body of work, experimenting with Deforum. 2023 — Creative director and third member of Two Shell; ML pipelines for music and visuals across the EU and US tours. 2024 — Join Kaiber as Head of Labs. 2025 — Head of Creative Research. "Once I saw the anime heads, I knew this was gonna be part of the collection. I was already a fan of KYT as well… the anime faces are just changing and morphing and bringing new experiences." — Steve Aoki
  - In the Flesh — Sotheby's: https://www.sothebys.com/en/buy/auction/2021/contemporary-curated/in-the-flesh
- press: [Kyt Janae wants creatives to seize the NFT moment — Bitcoin Isle](https://www.bitcoinisle.com/2021/07/17/kyt-janae-artist-and-director-at-foundation-wants-creatives-to-seize-the-nft-moment/)
- installation: COSMOS — Foundation × Jen Stark, "Cascade"
  - a multi-dimensional web + experiential visual art project inside Jen Stark's solo show Cascade.
  - COSMOS was a multi-dimensional web + experiential visual art project integrated into Jen Stark's solo show Cascade. My work on Cosmos spanned both the creative and technical development of the experiential show and the web activation.
- press: [Foundation names Kyt Janae Director of Collector Relations — FAD Magazine](https://fadmagazine.com/2021/06/07/foundation-names-kyt-janae-director-of-collector-relation/)
- press: [Rick and Morty Gets Animal Crossing Makeover — ComicBook.com](https://comicbook.com/anime/news/rick-and-morty-animal-crossing-crossover-makeover-adult-swim-promo/)
- short film: [Rick and Morty Crossing — Adult Swim](https://www.youtube.com/watch?v=JZoxnhJun4I)
  - i wrote and directed this one with a team of three, with 3d through Daisy Studio. rick and morty wander into animal crossing: meeseeks as villagers, a giant robot, one very smashed tent.
  - "Rick and Morty Crossing" came out in early 2021 and went viral. we used a wide range of experimental animation and event production techniques to finish it with a team of three.
  - Rick and Morty Crossing (YouTube): https://www.youtube.com/watch?v=JZoxnhJun4I
  - ComicBook.com write-up: https://comicbook.com/anime/news/rick-and-morty-animal-crossing-crossover-makeover-adult-swim-promo/

### 2020

- artwork: Real Enough — trained GANs + controlled keyframe interpolation
  - "real enough" is a digital painting i made by training my own gans in runway ml (desktop), then doing controlled keyframe interpolation through them plus post-processing. novel at the time. a meditation on insatiable desire and the limits of reality, with an original score by philip rugo.
  - alongside it, a gan trained toward my soft pink 3d forms.
  - Real Enough on SuperRare: https://superrare.com/kyttenjanae
- performance: [SIGGRAPH Chapters Gathering](https://lasiggraph.org/event/chapters-gathering)
- collaboration: [Kyttenjanae x OWSLA Goods](https://www.instagram.com/p/B9e2AMjjhcq/?img_index=1)
  - a very limited merch drop with OWSLA Goods, Skrillex's label and collective, built around my original characters and artwork. it's still the only merch i've ever done, released march 2020.
  - original post (Instagram): https://www.instagram.com/p/B9e2AMjjhcq/?img_index=1
- press: [Trilogía de artistas — Meow Mag](https://meowmag.mx/trilogia-de-artistas-cult-of-dang-miranda-rosales-y-kytten-janae/)

### 2019

- direction: Rick and Morty Season 4 announcement — WarnerMedia Upfronts, Madison Square Garden (May)
  - i directed the season 4 announcement for adult swim at the warnermedia upfronts at madison square garden: a narrative with a ton of [as] easter eggs, pixel-mapped across 17 screens, for the top 2,000 ad buyers in entertainment.
- solo show: [Cyber Spirits — Spectrum News](https://spectrumnews1.com/ca/southern-california/news/2019/10/29/new-media-art-finds-audience-through-cool-experiences)
- direction: [Kyttenjanae for Rick and Morty — Adult Swim](https://youtu.be/_XbwoODqjyQ?si=0qQr_UNKrLhLSPoz)
  - "Blorpland" is my favorite short i made for the show, an animated musical set on a weirdo planet i invented for the rick and morty multiverse. made with animator sam hochman and philip rugo on music and vocals, for the season 4 launch.
  - I worked on Rick and Morty from 2018–2021, writing, directing, and leading multiple creative campaigns for the series: the season 4 announcement at Madison Square Garden, web + TV idents, and later "Rick and Morty Crossing".
  - Rick and Morty × Daisy Studio (YouTube): https://youtu.be/_XbwoODqjyQ
- music video: Tia Tamera — Pitchfork top 20 music videos of 2019 (Daisy Studio)
  - one of Pitchfork's top 20 music videos of 2019.
- research: SPADE-COCO GAN experiments (June)
  - early gan experiments with SPADE trained on COCO: painting rough semantic maps (a smiley face drawn as grass, sky and sea) and letting the model hallucinate them into landscapes.
- short film: [Di Bibl — Adult Swim (with John Lee, PFFR)](https://www.youtube.com/watch?v=D2IfY6MUqBc)

### 2018

- career: Co-founded Daisy Studio
  - the other half of my brain.
  - I've spent the last 15 years honing emergent technology in cutting-edge and experimental ways to tell stories for artists, brands, entertainment IP, labels, agencies, and start-ups. I co-founded a creative production team, Daisy Studio, in 2018 as a way to scale and explore bigger technical and creative projects — working with giants like Adult Swim, Nike, Apple, Google, Viacom, and more. Complex pixel maps, episodic television production, immersive VR experiences, interactive apps and sites, audioreactive visuals, music videos, short films… the list goes on. Memetics work for Viacom/Nickelodeon, Toonami, and Meta × Complex's 2020 Year in Review; character work for Bloodpop, Grimes, and Hana.
- direction: [Kyttenjanae — Adult Swim IDs](https://youtu.be/yScs74xmfC8?si=J6V_kfS-EWT5tWNK)

### 2017

- performance: [The Food & Liquor w/ Philip Rugo — NTS Radio](https://www.nts.live/shows/food-liquor/episodes/the-food-liquor-w-kytten-janae-philip-rugo-14th-july-2017)
- honorable mention: [FMWH (Find My Way Home) — Prix Ars Electronica](https://ars.electronica.art/)

### 2016

- press: [Kytten Janae — Design Collector](https://designcollector.net/likes/kytten-janae)
- short film: ["Call Your Parents" — CLOWNS, Off The Air, Adult Swim](https://youtu.be/ZHHl9aVRed8?si=-S_vqHVzobPPl7k0)
- press: [QWERTY, Flirty, & Crying — CultBytes](https://cultbytes.com/qwerty-flirty-crying-big-pictures-los-angeles/)
- group show: [QWERTY, Flirty, & Crying — Big Pictures LA](https://web.archive.org/web/20160923054555/http://bigpictures.la/portfolio/arvida-bystrom-kytten-janae-ambar-navarro-brittney-scott/)
- music video: [Floridada — Animal Collective](https://www.youtube.com/watch?v=cuoIvNFUY7I&list=RDcuoIvNFUY7I&start_radio=1)

### 2015

- press: [The deformed, lonely bodies of Kyttenjanae's colorful worlds — Kill Screen](https://killscreen.com/previously/articles/deformed-lonely-bodies-kyttenjanaes-colorful-worlds/)
- press: [Kyttenjanae's Leggy Vines — Vice / The Creators Project](https://www.vice.com/en/article/gif-six-pack-kyttenjanaes-leggy-vines/)
- music video: [Soft as Snow — Vice / The Creators Project](https://www.vice.com/en/article/trek-3d-alien-mountains-in-soft-as-snows-new-music-video)
- press: [Google's Psychedelic AI Art Takes Twitter by Storm — Vice](https://www.vice.com/en/article/googles-psychedelic-ai-art-takes-twitter-by-storm/)
- music video: [D∆WN "Calypso" — Vice / The Creators Project](https://www.vice.com/en/article/dive-into-a-rainbow-alien-disco-in-dawns-calypso-music-video/)
- press: [KyttenJanae! — KALTBLUT Magazine](https://www.kaltblut-magazine.com/focus-on-kyttenjanae/)

### 2014

- group show: [The Virtual Art Gallery Levels-Up at UCLA — Vice / The Creators Project](https://www.vice.com/en/article/pgqzdz/the-virtual-art-gallery-levels-up-at-ucla)