I run Cirrela. Most of our traffic comes from an AI assistant citing a comparison page, but the product that actually pays the bills is simpler: a parent types their kid's first name, and Santa says it in a 9–15 second vertical video. €6.99 / $6.99, once, no subscription.
This post is the pipeline, the business model around it, and the details that were harder than they look. Disclosure: it's our product, written with AI help, checked by me.
The pipeline
- A form, not a prompt. The parent picks an occasion (Santa, the tooth fairy, Saint Nicholas, a birthday, New Year, Diwali, Eid, Valentine's…) and a character. They type the child's first name and optionally a short message. No last name, no photo — ever.
- A script table, not generation. For each occasion × language we keep 3 hand-written script variants with a
{name}slot. They're deliberately gender-neutral ("Be kind, go to bed on time" works for a girl and a boy), and ascriptFits()check rejects any script that won't fit the duration once a long name is inserted. - The talking character. The render goes through HeyGen's video model on fal.ai: a fixed character (Santa by the fire, Mrs Claus, an elf, a reindeer…) with a defined voice and motion, lipsynced to the script word for word. ~9 models sit behind the studio side; wishes use this one path because consistency matters more than variety for a keepsake.
- Preview first, pay once. The whole video renders free with our small watermark. One payment removes the mark. There's no "low-res teaser" — the thing you watch free is the thing you keep.
- Refund on failure. If the model fails or the name comes out wrong, the render is refunded automatically and the parent gets a free remake.
The details that were harder than they look
Portuguese is two languages. pt is Brazilian Portuguese ("você"), pt-PT is European ("tu"). A Lisbon grandmother noticed Brazilian phrasing instantly in testing, so the scripts are separate, not machine-translated. Visitors from Portugal land on the pt-PT version automatically.
Names are the whole product. "If a name sounds wrong, write it the way it's pronounced" is in the UI — but the deeper fix is that the name is spoken, not just displayed. The TTS step gets the raw string, so weird spellings (Siobhán, Łukasz) are where quality lives or dies. One free remake covers the misses.
Privacy by design, not by policy page. We never ask for a child's photo — names only. Anything typed into the message field goes through moderation that strips contact info and banned words, because a Santa video is exactly the kind of thing a scammer would love to put a phone number in. Photos that are used (studio side, adults only) go through a vision check that refuses anything with a minor or a real person.
The watermark is the business model. Watching the full video free with our mark is what makes a €6.99 one-off purchase feel safe for an emotional purchase. People pay to keep something their kid will watch every year — that's a different psychology than "unlock the preview".
What it costs to make one
The render is the only real cost (~30 cents ceiling on the video model), so a wish at €6.99 has healthy margin — but volume is everything at that price. 100 wishes/month is a nice side income; it's not a company. That's why the same characters and pipeline also power the studio side (subscriptions for people who want to make their own videos), and why distribution matters more than features right now.
We're small, it's seasonal, and the numbers are honest: about 15 payments in the last two weeks, mostly wishes. If you're building something emotional and one-off, the lesson I'd give is: let people see the whole result before they pay. Trust is the conversion.
