Mock Commercial · Training Capstone
A thirty-second commercial for a soft drink that does not exist, produced from brief to master as the running example of an AI video course written, presented and produced entirely by Ali Bali.
Where it comes from
Vidéo par IA, les essentiels is a course on AI video production: sixteen video capsules, close to three and a half hours, and a printed companion of methods, formulas and ready-to-use prompts. It was written, presented and produced from the first script to the last export by Ali Bali, founder of 3TOON, for Alliance IA, the training programme run by Vincent Lévi. Its premise fits in one line, the first of its eight formulas: you are not learning a piece of software, you are learning cinema.
A course about making films needs a film. So the sixteen capsules follow a single project from brief to master: a thirty-second spot for SODAI, a soft drink invented for the occasion. Every lesson, from the starting image to the sound mix, is demonstrated on it, and the final capsule projects the finished spot along with the real ratio of what was generated to what was kept.
The brief: a western crossed with cyberpunk, a Hollywood shoot out of a keyboard. A long time from now, in a Far West far, far away, a cowboy rides a saddled hover-drone across the desert to a saloon lit with neon, where a robot waitress serves him a SODAI. He drinks, raises the can to the room, and the packshot lands with the slogan: « Avec SodAI, on est prêt à n’importe quel futur ! » With SodAI, you are ready for any future.
The whole spot was made in front of the trainees, on the Higgsfield platform, and every step became a lesson, including the ones that went sideways: the content filter that rejected the robot four times, the French line that came out three different ways, the first saloon ambience that sounded like the wrong crowd. Nothing was cut from the process, because the process is the course.
Vous n’apprenez pas un logiciel : vous apprenez le cinéma.
You are not learning a piece of software. You are learning cinema.
The first of the course’s eight formulas
The two leads
The hero
A rugged cowboy in his early thirties, worn wide-brimmed hat, weathered leather duster, and a strange metal bracelet at the wrist: the signature anchor the machine never forgets. His description was locked word for word and pasted, unchanged, into every prompt he appears in.
The robot waitress
A saloon waitress of 1883 with a beautiful face and the mechanics showing at the neck and the joints, in the spirit of the original Westworld. The platform’s own filter rejected her four times before a cropped face got her through. She has the spot’s only line of dialogue.
Et ce sera quoi pour le beau jeune homme ?
Generated in French, with lip sync
Before the first shot
Six portraits, an election, then the sheet: front, profile, back and a few expressions, in a 4K square. It is attached as a reference to every generation, and it is why the same two faces show up in every shot.
The spot, frame by frame
00:04
A cowboy, a saddled drone, a trail of dust, and the saloon on the horizon.
00:09
Neon on weathered wood. The drone settles in front of the doors.
00:19
Eyes closed, can in hand. This is where the product meets the character, and where the spot earns its packshot.
00:12
The Gilded Servant, tray in hand, to the hero at the counter: « Et ce sera quoi pour le beau jeune homme ? »
00:16
The can slides along the polished counter and stops facing camera. The product is sacred: zero variation tolerated.
00:28
Logo, can and slogan, set in type at the edit. The slogan is never generated.
The course on one page
The whole course fits on one page of its companion guide. Each step comes with a formula to remember, in French, as the trainees learned it.
The brief
Vous n’apprenez pas un logiciel : vous apprenez le cinéma.
You are not learning software. You are learning cinema.
The starting image
Pour un travail sérieux, on part d’une image.
Serious work starts from an image.
Consistency
Pas de cohérence, pas de personnage.
No consistency, no character.
The prompt
Décrivez le mouvement, pas l’image.
Describe the motion, not the image.
The camera
Le sujet joue la scène. La caméra la raconte.
The subject plays the scene. The camera tells it.
Casting the models
Le meilleur modèle n’existe pas. Le bon casting, si.
There is no best model. There is the right cast.
Sorting
Trier, c’est déjà monter.
Sorting is already editing.
Sound
On pardonne à l’image. Jamais au son.
The eye forgives. The ear never does.
The craft
The images came first, in a set order: the logo, then the can generated with the logo as reference, the two character sheets, and the hover-drone composed from a research image and the elected cowboy. Each validated asset became the reference of the next.
Then the course’s bet: generate the whole thirty seconds in a single pass. A text AI turned the ten-shot storyboard into one long prompt, seven references were uploaded in order, and Seedance 2.5 returned the spot in 21:9, three takes, about a hundred credits each. The edit keeps the best of each take, upscales the master to 4K, sets the opening title and the slogan in type, and adds a score generated separately, a saloon ambience, and a synthesised cowboy voice for the tagline, complete with the sacrificial “OK” at the end that the cut removes. Twenty-seven years of animation craft behind every frame.