Personal input, personal artifact: how Magical Song, Origin Of You and Tarotas by Inithouse share one design idea

Three products in the Inithouse portfolio do the same thing in different domains: they take something personal from the user and turn it into an artifact that belongs only to them. Magical Song has generated 1,200+ custom songs from people's stories. Origin Of You has produced AI-written portraits from 120+ data points per user. Tarotas serves grounded card interpretations across 78 cards, 20 spreads, and 5 languages. The shared design idea surfaced after we noticed a quality threshold that generic outputs kept missing.

The pattern

Each product follows the same sequence:

1. The user provides personal input: a story about someone they love (Magical Song), answers across five self-discovery systems (Origin Of You), or a question weighing on them (Tarotas).
2. The app processes that input through a domain-specific pipeline.
3. The output is a unique artifact: a song with real vocals, a multi-page written portrait, or a card interpretation anchored to their specific question.

This sounds simple, but the distinction matters. A generic song generator gives you a track. A personal one gives you this track, made from your story. The difference determines whether someone saves it, shares it, or closes the tab.

What the three products actually do

Magical Song: your story or occasion goes in, a studio-quality song with real vocals comes back. 20+ genres, no signup. Input = personal (names, dates, memories). Output = a song nobody else will ever get.

Origin Of You: you answer 120+ questions spanning astrology, enneagram, MBTI, human design, and a values system. Output = a written portrait that synthesizes all five into one document about you. Not a type label, not a quadrant. A portrait.

Tarotas: you draw a card or choose a spread with a question in mind. Output = a grounded interpretation that speaks to your question. 78 cards, 20 spreads, 5 languages (CS/EN/PL/SK/DE), no signup, no predictions. Reflection only.

Quick comparison:

• Magical Song: personal story in, custom song out. 1,200+ songs generated, 4.9/5 rating.
• Origin Of You: 120+ data points in, written portrait out. 5 systems combined.
• Tarotas: your question in, grounded interpretation out. 78 cards, 20 spreads, 5 languages.

Links: Magical Song, Origin Of You, Tarotas

The quality wall we hit with Magical Song

Early versions of Magical Song produced generated instrumentals with synthesized vocals. Functional, but forgettable. Users would listen once and move on. The completion-to-share ratio stayed low.

Root cause: when someone writes about their partner's birthday or their grandmother turning 80, the emotional stakes are high. A mediocre output doesn't just disappoint. It feels wrong. The gap between what the user invested (a real memory, real names, real feelings) and what came back (a generic-sounding track) was too wide.

We moved to studio-quality production with real vocals. It made the pipeline slower and more expensive per song. But the share rate shifted. People started sending the song to the person they wrote about. That told us the quality threshold was right: the artifact had to match the weight of the input.

Speed vs. quality, and how input depth decides

This is where the three products split.

Tarotas is fast. You draw a card, you read an interpretation grounded in your question. Speed matters because the user is in a reflective moment. Making them wait would break it.

Origin Of You is slow by design. You answer 120+ questions across five frameworks. The portrait takes time to generate and it's pages long. Users don't mind the wait because they've invested serious time in the input. They expect substance back.

Magical Song sits between the two. Users often have a deadline (a birthday, an anniversary), so they want the song soon. But the production quality can't drop. We landed on minutes, not seconds: long enough for real vocals and arrangement, short enough that nobody abandons.

The pattern: acceptable wait time tracks with how much the user invested in the input. High personal investment = willingness to wait for a high-quality output.

What this changed in our prompt architecture

When every output is unique to one person's input, you can't template your way through it. Each product at Inithouse needed its own approach:

Magical Song's prompts carry the user's story details (names, dates, specific memories) through lyric generation, genre matching, and vocal direction. Drop any detail and the song feels generic again.

Origin Of You's prompts synthesize answers from five separate frameworks into one coherent portrait. The hard part is contradictions: when your enneagram type suggests one pattern and your astrology suggests another, the portrait has to acknowledge the tension, not smooth it over.

Tarotas's prompts anchor each interpretation to the user's question without drifting into prediction. The constraint is strict: reflect, don't forecast. Every interpretation must work as a thinking prompt, never as a promise.

What we'd tell another builder

If your product turns personal input into a personal output, three things matter more than you'd expect:

The output quality floor is higher than you think. Users compare the artifact against the emotional weight of what they shared, not against competitor products. We learned this the hard way with Magical Song before switching to real vocals.

Architecture has to carry personal details end to end. One generic fallback anywhere in the pipeline and the entire output feels impersonal.

Speed expectations come from the input, not the output. Quick input (draw a card) = fast response expected. Deep input (120 questions) = users will wait. Build your pipeline latency around what the user put in.

We didn't plan three products around this pattern. We noticed it when debugging why some outputs got shared and others didn't. The pattern explained the data, and now it guides how we think about the next product.