Chigüiro Media is the studio that came out of two years of making films at night with generative models. It makes 30-, 45- and 60-second videos for any business: storyboard first, finished film 48 hours later, at a fixed price. This page covers the brand I built for it, how a film gets made, the client work (Ocubio, EyeCatch, KUASA) and the personal projects where the method was learned.
From a hobby to a studio
For two years I made films at night with generative models: a children's series, a music video, four short films for an electoral reform, history explainers for a Malaysian civic group. Somewhere along the way the hobby turned into a method, and the method turned out to be the part people wanted to buy.
In October 2026 it got a name. Chigüiro Media makes short videos for any business, a café, a clinic, a construction firm: the client approves a storyboard scene by scene, and the finished film arrives 48 hours later, at a fixed price, in any style and any language. No shoot, no crew, no software for the client to learn. It lives at chiguiromedia.com, in six languages.
The name and the mark
A chigüiro is the capybara of the Colombian and Venezuelan plains: calm, sociable, unhurried, and impossible to mistake for anything else. The name keeps the Colombian root and is never translated: not Capybara Media, not Carpincho Media. And always with the Ü, because the two dots are part of the logo.
The mark has four parts. A chigüiro facing right, forward along the timeline, with an angular red eye. A laser grid cut into its body, three by four diagonal incisions, halfway between a woodcut and a 3D scan. A trail of sixteen blocks behind it: the client's inputs assembling into motion. And a heavy condensed wordmark, where the dots of the Ü become two red squares.
Around it: an Obsidian, Volt and Infrared palette, Barlow Condensed with Plus Jakarta Sans and JetBrains Mono, a clock that counts from 00:00:00 to 48:00:00, and a 35-page guideline book with the logo in every format, icons, templates and the voice rules in six languages. I designed and built all of it, the website included.
How one gets made
Nothing is generated in the order you watch it. The characters come first, drawn once as turnarounds — front, side, back, same scale — and those drawings are the only thing the video model is ever shown. Then a still frame for every shot, approved before a second of video is spent, because a frame is cheap to redo and a clip is not. Only then the clips, which run about ten seconds each, so a three-minute episode arrives as twenty-five separate pieces.
Song comes before picture. The music is written and recorded first, and the shots that sing are generated with the real lyrics, framed so the mouth stays visible — the shapes land close before anything is corrected. The video models won't take an audio file, so the sync itself is a second pass, video against vocal, run only on the few shots that need it. The stop-motion stutter goes on at the very end, where it also covers whatever drift is left.
Then it gets cut like anything else: the model's audio muted, the real mix underneath, and the pacing, the text and the silences decided by hand.
Keeping it the same film
Every clip is generated on its own and remembers nothing. Continuity is something you build, not something you get. Each character carries one colour and one object so the silhouette reads even when the model wanders, and each voice comes from a line of description pasted into every prompt without changing a comma.
Where two shots are really one moment continuing, the second starts from the last frame of the first, which carries position, light and colour across the cut — but only in runs of two or three. Chain a whole episode and by the fifth generation the character has quietly melted into someone else.
One model per job
No model is good at everything, and using the expensive one everywhere is how you run out of credits halfway through. Stills and character sheets come from Nano Banana. Motion and sound from Veo and Gemini Omni Flash, inside Google Flow. Songs from Suno, voices from ElevenLabs and Gemini. Lip-sync from sync.so, Kling or OmniHuman depending on the shot. Then Premiere or CapCut — the oldest tool on the list, and still the one that decides whether any of it works.
For client films the photoreal work now runs on MiniMax H3 with one take per shot, and every character's line is generated inside a single take, so the voice stays the same wherever that person is heard.
Three films to launch it
The studio's own showreel had to prove the promise: any business, any style, 48 hours. So it is three films, each with a different trick. Your Move hands one sentence between seven people in two languages. Lucía Turns Seven walks one story through seven styles. Stamp It is a song across twelve businesses and twelve styles.
Client work
Ocubio: a stand film for an eye-care fair
A 45-second film projected as the hero video of Ocubio's stand at a US eye-care fair. Real patient testimonials, re-enacted word for word by AI-generated actors, with the legal line on screen. Everything was planned for a noisy hall: it works with the sound off, every five-second window carries the idea on its own, it loops without a seam, and the grade lifts the shadows so the projector doesn't wash the navy out. The client approved script, cast, packaging and voices before a single clip was generated.
EyeCatch: The Money Booth
An eye-catcher for EyeCatch's screens at NACS 2026, the US convenience-store show: a loud 1990s TV game show re-shot as a premium commercial. A host, a giant red button, a store owner in a glass booth full of flying cash. The style is locked in one paragraph pasted into every prompt, which is what keeps thirty seconds of photoreal footage looking like one shoot.
KUASA
I have worked with this Malaysian civic organisation since 2022 — brand, media-literacy programmes, a weekly news show. AI production arrived there for the reason it usually does: a small team with far more to say than it can afford to film. Two strands run on it now, both in Malay — a series on Malaysian political history, and explainers on local-council elections, the same subject as the Selangor grassroots survey I helped design. Archival, documentary, talking heads: a third thing again.
Where it started
Before there was a studio there were personal projects, made for my own reasons. They are where the method was learned, and they are still the best proof of it.
El Valle
Ten episodes for children of three to six, about living together: sharing, saying no, helping, learning by watching. Cut paper, six voices, and a three-note motif that opens and closes every one of them. It goes out as Chigüiro Director on Instagram.
There are two silences in it that nothing is allowed to fill — ten seconds on a still stone in the middle, and four absolute seconds when the story asks the child a question. If anything makes a sound in those four seconds, the child doesn't answer. That is not a technical decision, and it is the kind of decision the tools have no opinion about.
Ordena Tu Voto
A campaign for a change to how Colombia picks a president: a ranked ballot in place of the run-off. Four short films make the same case four different ways — the lesser evil, the bill, the restaurant, the signature that binds — because one film only ever reaches the people it was already written for.
Vertical, photoreal, in Colombian Spanish: the opposite register to the paper animation, out of exactly the same process.
What it is, and what it isn't
This doesn't replace a crew, and it isn't cheaper than one when you already have a crew. What it changes is who gets to make a film at all — an organisation with no production budget, a children's series that exists because one person could finish it, an argument that needs to exist in four versions before anyone will back the first.
It started as a hobby. Now it has a name, a price list and clients, and I know exactly how it works.