Skip to content
decosa

Turn a family interview into a short film

Record Grandma telling her stories, in her language, and add her old photos. Get her chapters, her best lines in her own voice, English subtitles under her own words, a trailer, a film and a book with QR codes that play her quotes.

About 50 s from upload to chapters for a 3½-minute interview (measured; longer interviews take longer)About 2 cents a film at list prices, no video model18 of 18 quotes re-heard at their timestamps in the sampleDetails, API and self-host

Make the family film

Try it in one click

Watch Rosa's stories become a film

A 3½-minute interview in Italian and eight old photos go in. About a minute later: her chapters, her best lines (each one traced to the second she said it), English subtitles under her own words, and a trailer in her voice.

Rosa is fictional and her voice is a designed synthetic voice; the photos are public-domain pictures standing in for a family album.

Your family

Before you record

Studio writes an interview guide in her language, chapter by chapter, with gentle follow-ups. Then she says yes in her own words, and you record.

The rules
  • Her consent comes first, recorded in her own words and checked for meaning; it is kept in the consent ledger, and her own link takes it back and deletes her recording, the transcript and every film.
  • No model ever draws, animates or changes a photo. Photos are only shown, panned slowly and captioned.
  • Children are paused for now. Before a photo is kept, its caption is read and the picture is looked at for anyone under 18, drawn or photographic, and for words that say so; a photo like that, or one the check can't rate, is refused and nothing from the upload is kept. A family mode, where a parent verifies and consents for their children, is planned.
  • The document reader reads only the back of a photo, and only after a face detector finds no face in it.
  • Her real voice only: no voice clone.
  • Private by default: one private link per person you name, each one you can turn off; never listed anywhere. Anyone else on camera must agree too.
  • Every film carries a content credential.
What it's for

For the grandchild who has two hours of Grandma on a phone that nobody will ever edit. Today that recording sits unedited, or it goes to a text-first memoir service, or to a hired family filmmaker. Studio does the slow part (finding the chapters and the lines worth keeping, and timing and translating them) and leaves the choices to the family: rename a chapter, leave a quote out, move a photo, cut again.

Models and where it runs

Speech recognition Qwen3-ASR-1.7B; who is speaking: ECAPA-TDNN on CPU, matched to her consent recording; chapters and quotes: Qwen3.8-27B; English: Hy-MT2-7B or Qwen3.8-27B with the meaning check; the photo check for anyone under 18: Qwen3.8-27B; photo backs: PaddleOCR-VL-1.6; maps: Natural Earth (public domain); score: ACE-Step 1.5, cleared; the edit: ffmpeg on CPU. All Apache-2.0 or MIT. Hosted on Decosa's own GPUs, or self-host so the recordings never leave the family.