Hybrid filmmaking, part 2: what was never filmed —AI aging, ash and score

What comes after the shoot

In part 1 we covered the hybrid filmmaking behind Pedazos del alma: Sara Camacho performed in a living room against a grey backdrop, and we carried that performance into an alpine valley with video-to-video AI. This part covers what was never filmed: what was generated, what was transformed beyond the location, and what you hear.

Five shots that do not exist in the footage

Four landscapes give the film room to breathe between Sara's shots: the summit, the creek valley, the grass at the cliff edge and the braided river from the air. They were generated with the same world and the same cold palette as everything else, so they cut against her shots without a jump.

The fifth is a corgi in the forest, seen from behind, its face never shown. It is a small figure walking away, like the people the voice talks about.

Old age as an interruption

The voice asks life to let her grow old. We wanted the audience to see her old for an instant, with no makeup and no digital morph. We aged Sara in Seedream on a real frame of her close-up, with her hand on her face. Starting from her face, not from a description, is what keeps the likeness.

That image went into the video-to-video pass as a second reference. In the film the old face appears twice, for a little over half a second each, inside the extreme close-up against the rock. It does not read as a transition: it reads as a stumble in the image, and then she is young again.

The ending: from a walk to ash

The ending comes from the same walk-away take that opens the film. The walk is Sara's; the AI moves it into a dark forest and turns her into ash that rises toward the light. It is a single generation cut into four shots: the walk, the hands, the back, and the wide shot where what is left of her disappears.

How to ask for a disintegration that does not look burned

• 01From many points at once. When the effect travels in order, from one end to the other, it reads as an erase. Here it opens at several points on the body at the same time and the breaches merge.
• 02Material, not transparency. The body cracks into flakes half a millimetre to five millimetres wide that keep their color and turn to grey ash as they rise.
• 03No reference image of the effect. A reference frame caught mid-disintegration made Sara look burned and crusted. We removed it and described the effect in text only.
• 04The damage adds up. In one test, the wide shot showed her whole again. The prompt keeps a ledger of what is missing at every cut, and what is gone stays gone.
• 05The hair goes last. If the hair comes apart before the head, she looks bald. The hair stays full and goes together with the head.
• 06She does not rise. She stays where she stands, and what is left of her comes apart from the feet up.

THE DAMAGE LEDGER — one event, never reset. Every shot begins in exactly the state the previous shot ended in; what is gone stays gone in every later shot and from every angle. After 2.0 s she is NEVER whole again.

HER HAIR AND HEAD — her long dark wavy hair stays FULL, LONG and on her head until the very end. It is never still, lifting and swaying in the rising air; only its very ends shed a thin trail of dark filaments. It is NEVER thinned, NEVER shortened, NEVER peeled away. Her scalp and the shape of her skull are NEVER visible; she is NEVER bald, not for a single frame. Her head and all of her hair come apart TOGETHER, in one instant.

The score: a silence and a heartbeat

The score was generated in Magnific with Lyria 3 Pro, inspired by the sound of ethereal Nordic folk and with no lyrics. It took three versions:

• 01First. Instrumental, with electronic risers. It sounded like a different film.
• 02Second. Orchestral and acoustic, with a wordless female voice and a heartbeat. The final climax sounded heroic, like a crusade, and the film is an inner dialogue.
• 03Third. Intimate, one minute long: piano, strings and the wordless voice; a silence of about four seconds, then the heartbeats alone, and the voice returns with the violins, no drums.

In the final cut the music drops out at 38.5 seconds and the heartbeats begin right under the line "con el corazón usado", with a well-used heart. We did not force it in the edit: the score starts with the film and the silence lands there.

Film-style subtitles in two languages

Each language has 19 subtitles, one per phrase, timed to the voice. No line runs past 42 characters or 17 characters per second. They are set in italics, the film convention for voice-over.

The English is written for the viewer, not translated word for word: "con el corazón usado" became with a well-used heart, "descorazonados" became the heartbroken, and "el alma desmoronada" became a crumbled soul.

The edit: 23 real shots, 5 AI shots

The final cut joins four 15-second generations into 59 seconds and 28 shots: 23 come from Sara's performance and five are AI only. That ratio is the idea behind the whole project: AI builds the world, and a person carries the film.

Hybrid filmmaking: AI builds the world, a person carries the film

That is the rule of hybrid filmmaking at ZerezoFilms AI Studio, the AI production arm of ZerezoFilms in Mexico City: what matters gets filmed, and what is impossible gets generated. In Pedazos del alma the generated parts are four landscapes, a corgi, half a second of old age, the ash and the score; everything else is Sara Camacho in front of a camera. The same hybrid filmmaking method is behind commercials like the Zerezo Bites spot.