Her Parents Came to Bless Her: How We Made an Eight-Minute AI Tribute Film for a 60th Marriage

A Sashtiapthapoorthi gathers everyone who came to the first wedding and everyone born since. For this family, two people could not come. The bride lost both her parents years ago, and the brief, when it arrived, was a single wish: that they should be there to bless her.

So the film is built around their return. The temple doors are closed. Her father calls her name before we see him; her mother's voice follows. The doors open, and they come down the steps together, greet her and her husband, and talk to her — about the small girl who used to walk holding her father's finger, about the family she holds together, about watching over her still. Then the couple bow, her parents' hands come down on their heads, and when the couple rise the hands are gone. The film ends in clouds.

Around that centre sit the parts a family film usually has: a kavithai over their photographs at the opening, her father's song over the places of her childhood, a farewell song, and end cards with the date and the names. Eight minutes and two seconds in all, in Tamil, made for the screen in the hall — and delivered ten days after the first photographs reached us.

A film like this can go wrong in a second and in front of everyone: a face that is almost right, a line the family would never have said, a mouth that does not match its words. So the rules came first, and every decision afterwards was checked against them.

Her parents' faces come only from their own photographs. The old prints were restored, dressed for the day and approved by the family, and those two approved portraits are the single source every shot of them was generated from.

No living person's face is generated. The couple were in the hall, watching. At the blessing the camera stays behind them: two backs bending, and her parents' faces above them. The little girl her father walks through her hometown is generated and seen from behind, never taken from a family photograph.

Every word is the family's. The dialogue was written with them, line by line, and her parents' speeches were recorded by real people, not a text-to-speech voice. The only generated voices in the film are the two short calls heard from behind the closed doors. And nothing went into the film that the family had not confirmed first.

What was cleared for the hall stays in the hall. The opening poem is another writer's work, and the songs are film songs; both are fine for a family function and a family WhatsApp group, and neither is ours to publish. That is why the excerpt on our website carries only the film's own dialogue, over music we generated ourselves.

Each speaking shot was made the same way. The line's own recording drives the performance, so the face moves with the rhythm of what is actually being said — the pauses, the stressed words, the smile between two sentences. Then a separate lip-sync pass re-shapes the mouth to that exact recording, so the words you hear are the words you see.

The order matters more than any setting. A shot generated silently and lip-synced afterwards has what we came to call the generic talking mouth: the lips move, but the face is saying nothing in particular, and people feel it even when they cannot say why. One of her mother's shots was exactly that, and the family caught it. A second lip-sync pass on the same shot changed the mouth by almost nothing we could measure. What fixed it was a new performance, driven by her line, and then the sync.

Two more lessons were paid for in retakes. Keep the eyes open: a face that speaks with its eyes half-closed reads as absent, even when the lips are perfect. And give every line its own shot. When two faces are in frame and one of them speaks, the model moves both mouths, so a two-person greeting became a cut between two singles — the only way two people can talk to you in a film like this.

Dialogue for a film like this is easy to over-write and hard to cut, because every line means something to someone. We ran it through the checks an editor would. A repetition audit across all of the lines found six places where a phrase came back minutes after it had already landed, and each was cut or rewritten; a greeting repeated inside the same twelve seconds was left alone, because two people greeting two people is simple courtesy.

One line needed more care than any other. An early draft had her mother asking who was looking after her now. Spoken in a hall full of the husband, sons and daughters-in-law who do exactly that, it is a question with an accusation inside it — nobody would say it aloud, but everyone would hear it. The rewrite thanks the family first, and then makes the loss the mother's alone: she is simply not there to ask whether her daughter has eaten. The same ache, and no blame.

Her father's song plays over her childhood, and the pictures are his walks with her through the town she grew up in: carrying her across the fair ground at dawn, walking her by the hand, the places the family named for us, each shot held for a line of the song. The song was cut down from the recording to its sung lines, with the repeats taken out, so it lands where the film needs it and ends on its own last note.

The blessing is the one moment in the film that belongs to the living couple, which is exactly why their faces are never shown. We see them bow from behind, her parents' hands come down on their heads, and her mother's words. When they rise, there is no one in front of them. The parents lift and scatter into gold, and the film opens out into clouds.

A film this long, rebuilt after every round of notes, drifts in ways nobody sees until the morning it is shown. So every build ran automated checks before anyone watched it, the way a software release runs its tests. Each check exists because it once caught something real.

Lip-sync was measured, not eyeballed: the delay between voice and lips on sampled shots, against a limit of 120 milliseconds; the worst shot in the final film is at 60. Every shot that carries its own sound was matched against the finished film to prove it is heard over its own picture and not the shot next to it. No gap of silence longer than two and a half seconds is allowed anywhere. And the songs were checked to run unbroken across the sections they cover, into the end cards.

The film belongs to the family, and most of it stays with them. The 73-second excerpt on our site was cut specifically for the site: the temple, her parents speaking, two walks, and the blessing. It has none of the family's photographs, none of their relatives or grandchildren, no end cards with their names, and none of the poem or the songs. The dialogue is subtitled in English so that a visitor who does not speak Tamil can follow what her parents say.

If you commission a film like this and want to share it publicly afterwards — on YouTube or Instagram rather than in the family group — tell us at the start. We will cut a public edition with music we are allowed to publish, and it costs far less to plan for than to rebuild.

Our rate card is public and the same for families as for businesses: ₹6,000 for every 30 seconds of finished film, with a project minimum of ₹40,000 and one round of revisions included. A three-minute film is the minimum; an eight-minute one is ₹96,000. The quote is fixed and in writing before we start. What that price covers is in its own post.

This film was made in ten days because the date could not move. Plan for two to three weeks if you can: the slowest part is never the film, it is finding the photographs. Send us the date, a dozen photographs and a few lines about the people through the AI Video page, and we will come back with a fixed quote and a shot list.