Making of · Aún sin Nombre

How we made a 90-second manifesto film, and what each mistake taught us

«El valor colectivo de la cultura» is a short film for Aún sin Nombre, a Dominican cultural collective building a public-policy and investment proposal for the cultural sector. This page follows it from the first idea to the published film, with the iterations left in: what looked wrong, why it happened, what we changed, and the rule we wrote so the next film starts better.

El valor colectivo de la cultura — Aún sin Nombre

«El valor colectivo de la cultura»

On the collective’s site, in Spanish, with dubbed versions in English, French, Brazilian Portuguese and Catalan, and subtitled versions with the Spanish voice.

Watch the film

The method in one loop

The film was made with the MediaMax workflow: a written canon of production rules, prompts written to that canon, a human judging every output by eye and ear, and every lesson written back into the canon.

  1. 1

    Canon first

    Before a single prompt, the rules: how a shot moves, what a prompt must never name, how a voice is directed.

  2. 2

    Prompts written to it

    Every prompt is a file, written once and pasted exactly, never retyped from memory.

  3. 3

    A human judges

    Nothing goes in the film until a person has watched it and listened to it.

  4. 4

    The lesson goes back

    Each fix becomes a written rule, so the next film does not repeat the mistake.

From idea to published film

1. Conceptualize

The script was written for a series, in three directions. Carlos chose the first, «La caravana»: Dominican culture already moves together — güira, drums, painters, actors, a carnival caravan — and could do more with a policy that takes it seriously. The film carries a single figure, the share of GDP the country spent on culture in 2024, and it was checked against the Central Bank’s national survey before it went in. A film that argues for policy can afford one number, and that one is sourced on screen.

2. Cast the voices

An audition of several takes on Google AI Studio’s speech page, run in Carlos’s own browser. He chose two voices by ear, one female and one male, both with a Dominican accent, so the film speaks in two voices instead of one.

3. Score it

Music takes came from Lyria, Google’s music model, each one approved before it was generated. Two prompts were refused by the model’s filter for the words «hit» and «heartbeat»; rewritten as «accent» and «pulse», they went through. In the mix the music ducks under the voice and comes back between lines.

4. Build the first cut

The film is assembled in HyperFrames, where a video is written as a web page and rendered to MP4. The first cut used still images made in Google Flow, captions in the collective’s own typeface, the two voices and the music: 16:9, 90 seconds.

5. Five languages

Subtitles first, then dubs into English, French, Brazilian Portuguese and Catalan. Each language speaks at its own pace, so the build re-times every picture and caption to that language’s voice instead of forcing one timeline on all of them.

6. From stills to moving pictures

About two dozen silent shots were made in Google Flow from the stills, using each still as the first frame. A short three-shot test decided which model to use where: Veo for most shots, Omni for close-ups of hands. The voice and music stay ours; a clip’s own sound is kept only as foley, on the shots Carlos picked by ear.

The iterations

This is the part most making-of pages leave out. Each change below started with someone watching or listening and saying «that’s not right». We kept the old shots where we could, so you can see the difference for yourself.

01

Stills that wanted to move

Before
The still-image film
After
The same ten seconds, moving

Turn the sound on for this one.

What we saw
The first full cut was made of still images with camera moves over them. A film about culture moving together, built from pictures that stand still.
Why it happened
The stills had already locked the story, the timing and the look, in every language. What they could not do was move.
What we changed
Each still became the first frame of a short generated clip, so the motion inherits the look we had already approved. The clips are silent; voice and music stay as they were.
The rule it became
Approve the look on stills first, then animate from them. A still is a contract for the shot.
02

The same zoom, over and over

Before
Three pictures, one quick push-in each
After
The same seconds, one slow move per picture
What we saw
In the still-image cut, every picture used the same quick push-in. By the fourth image the eye stopped seeing the pictures and started seeing the trick.
Why it happened
Every image got the same push-in, of the same amount whatever the shot’s length, so the short montage shots zoomed fast.
What we changed
One move per image, chosen from what is in it (push or pull toward a focal point, pan, tilt, or hold), never the same move on two shots in a row, all set by a slow fixed speed. Very short shots simply hold.
The rule it became
Ken Burns as rates, not amounts: a camera move is defined by its speed, so a long shot travels further instead of crawling and a short one doesn’t lurch.
03

The walkers who looked like zombies

Before
The accordion player, first take
After
The accordion player, redone
Before
The painter, first take
After
The painter, redone
What we saw
Two people walk toward the camera: an older man with an accordion on a rural road, a woman with a canvas in a town street. Both walked stiffly and stared into nothing. Carlos’s verdict on the first: «it feels like a zombie».
Why it happened
The prompt said where the body goes («walks steadily toward the camera») but not how they walk, why, or what they look at. The model filled the silence with the emptiest possible walk.
What we changed
Describe the gait (relaxed shoulders, arms swinging, weight rolling from foot to foot), give each person a reason to move that shows in the body («on his way to play at the fiesta, a little late»), and something to look at: the cane in the wind, a friend down the street. No emotion adjectives. Veo got both right on the first take.
The rule it became
«People in motion» is now a section of the MediaMax prompt canon: every moving person gets a gait, a reason to move, and a gaze target. This film’s before and after are recorded there as evidence.
The prompt, before / The prompt, after

Prompts are shown in English, the language the video models take instructions in. Only the part that changed is shown; the shared style rules are left out.

The prompt, before

The older man walks steadily toward the camera along the rural road at sunset, the accordion case on his back, the sugar cane moving in the wind. The camera moves slowly backward ahead of him.

The prompt, after

The older man walks toward the camera along the rural road at sunset with an easy, unhurried country stride: relaxed shoulders, his arms swinging loosely, his weight rolling naturally from one foot to the other, the accordion case shifting a little on his back with each step. He is on his way to play at the fiesta, a little late, and keeps a good pace: a light smile with closed lips as he walks. His eyes are alive and move naturally: he glances to the side at the sugar cane moving in the wind, then looks ahead down the road toward the village where he is expected, and gives a small nod to someone he passes. The camera moves slowly backward ahead of him.
04

A güira that wasn’t being played

Before
One slow stroke a second
After
Fast and rhythmic

Turn the sound on for this one.

What we saw
The film opens on a güira, the metal scraper at the heart of merengue. The first clip showed one slow stroke a second. Anyone who has heard merengue knows that is not how it is played.
Why it happened
The prompt asked for «a slow, steady rhythm» plus «a very slow push in»: we asked for calm and got calm. It read as cinematic in the text and wrong on screen.
What we changed
Redone as a musician keeping the beat of a band: quick, short scrapes, several a second, the wrist loose and bouncing, the camera still. Generated with Omni, the model the test had picked for hands.
The rule it became
Describe the instrument as it is really played, with its tempo. A cultural detail is part of the brief, not decoration, and the people it belongs to will notice.
The prompt, before / The prompt, after

Prompts are shown in English, the language the video models take instructions in. Only the part that changed is shown; the shared style rules are left out.

The prompt, before

The hand scrapes the metal güira with the fork-like scraper in a slow, steady rhythm, the metal catching the single bulb’s light on every stroke. Camera locked off, with a very slow push in toward the scraper.

The prompt, after

The hand plays the metal güira in a fast, steady merengue rhythm: quick, short, rhythmic scrapes of the fork-like scraper up and down the ridged metal, several strokes every second, the wrist loose and bouncing, the scraper flicking back for the next stroke without pause, like a musician keeping the beat of a band. The metal flashes with the single bulb’s light on every stroke. The camera is still, held close on the hand and the güira, with almost no movement.
05

A güira you could see but not hear

Before
The opening: silence until the voice
After
The opening: the güira alone, then under the voice

Turn the sound on for this one.

What we saw
With the fast güira in place, Carlos listened and heard no güira at all.
Why it happened
The music has its own güira, but it only starts two seconds in and sits far under the voice. The picture promised a sound the mix never delivered.
What we changed
The fast clip’s own scraping now opens the film, loud and alone, then steps back into the background as the first line starts.
The rule it became
If the picture shows a sound, the audience must hear it. Check the mix against the picture, second by second.
06

A brand name printed on the frame

Before The edge markings, top to bottom on the right
The edge markings, top to bottom on the right
What we saw
One clip, a tailor cutting fabric, suddenly turned into a strip of film with the stock’s name printed down its edge.
Why it happened
The shared style rules named a real film stock to describe the grain. The model took it literally and drew the film itself, edge markings and all.
What we changed
That clip was cropped and cut short, and every later prompt describes the look («fine film grain, warm amber highlights against deep blue-green shadows») without naming a stock.
The rule it became
Prompts never name a film stock or any other brand. Describe the look, not the product.
07

A scratch at the end of every line

What we saw
The English takes had a click and a scratch after each line. Early takes also read their own stage directions aloud: «warm, Dominican accent…».
Why it happened
The voice model returns a complete audio file, and our script wrapped it in a second file header, so the inner header played as noise. The stage directions were written inside the text to be spoken, so they were spoken.
What we changed
Keep only the audio data; put the direction in its own style field, never in the text.
The rule it became
Listen to every take. A checker in the MediaMax toolset now looks for both faults before a take is accepted.
08

A black preview

What we saw
Shared links showed a black rectangle as their preview.
Why it happened
The film faded in from black, and many apps take the first frame as the preview.
What we changed
The film now opens on the güira itself and carries an embedded cover: the end card, with a QR code to the collective’s site.
The rule it became
The first frame is a poster. Never open a film meant for sharing on black.
09

Sounds that were there but not heard

What we saw
The caravan’s chanting, its drums and feet, the mural’s spray cans: all in the mix, all buried.
Why it happened
Each clip’s own sound was brought to one common level and laid under the voice and music. Subtracting two renders, to hear only those tracks, showed them sitting far below everything else.
What we changed
Raised, shot by shot, to the level Carlos asked for by ear. The plaza clip had no sound of its own, so its children-playing ambience is a public-domain recording from Wikimedia Commons, checked for intelligible words before use.
The rule it became
Foley earns its place only if it is heard. And a borrowed sound needs a licence you can show.
10

«Tá» or «Está»

What we saw
The Portuguese voice says «Tá nas mãos…», casual Brazilian speech, where the script had «Está».
Why it happened
Speech models speak the way people speak. Sometimes that is a defect; sometimes it is exactly right.
What we changed
We prepared a retake. Carlos listened, judged it natural, and kept it. The retake was never run.
The rule it became
A human decides what «correct» means. Not every deviation from the script is an error.

How every change was reviewed

Every change followed the same loop. Render one language. Measure it: loudness second by second, and the difference between the old and new render, so only what changed can be heard. Look at the frames. Then Carlos watches and listens. Only after his yes are all the versions rendered again.

Publishing

The masters are large, so the site plays light web copies, in the visitor’s language, at the top of the collective’s home page, with a share button and a WhatsApp link under the player. Every version is filed «in review» in a MediaMax Studio production, and every image, clip and music track sits in the MediaMax media library next to the prompt that made it. The old shots stay there too, which is why this page can show them.

Who did what

Carlos Miranda Levy

Direction. Every creative choice by eye and ear, and the approval for anything that costs money.

Claude Code: the production session

Wrote the script, the prompts, the build, the mix and the collective’s web page for the film.

Claude Code: the canon session

Keeps the MediaMax canon, and turned each lesson from this film into a rule for the next one.

Claude Cowork

Drove Google Flow and AI Studio’s speech page in Carlos’s browser, on his own plan, following prompt files exactly and reporting credits and checks after each run.

The tools

Gemini text-to-speech for the voices, Lyria for the music, Google Flow (Veo and Omni) for the shots, HyperFrames for the build, a local speech recognizer for word timings and checks, and MediaMax Studio to file every piece.

Take this to your own videos

  • Lock story, timing and look on stills before you spend on motion.
  • For people in motion, write how they move, why, and what they look at.
  • Define camera moves by speed, and vary them from shot to shot.
  • Describe a look; never name a brand, a film stock or a real person in a prompt.
  • If the picture shows a sound, make sure it is heard.
  • Keep stage directions out of the text a voice will read.
  • Keep your old takes. They are how you learn, and how you teach.
  • Let a person judge every output, and write each lesson down where the next project will find it.

The production rules cited on this page belong to the CEMI MediaMax canon. cemi.media

Want to practise these moves yourself? The Shot Lab has challenges built on the same canon.

Open the Shot Lab