Magical Song by Inithouse has produced over 1,200 custom songs since launch. That number is the starting point for everything else we measure about running an AI custom song generator with real vocals.
What the data shows
1,200+ songs created. Not demo clips or MIDI sketches. Full songs with real vocals, mixed and mastered. Each one starts with a user typing their story, picking from 20+ genres, and hitting generate.
4.9 out of 5 average rating. We prompt for a rating right after first playback, so the score captures the immediate reaction before any follow-up or second listen.
20+ genre options. Pop, rock, jazz, country, R&B, electronic, hip-hop, classical, reggae, and others. We added genres based on actual request patterns from users, not a preset list we thought sounded complete.
Finished songs in minutes. The pipeline from story submission to a playable, shareable link runs in single-digit minutes. That covers vocal rendering, mixing, and link generation.
What building it taught us
Real vocals change sharing behavior. Most AI music tools produce fully synthetic voices. We went with real vocal production. The cost per song is higher and the pipeline is heavier. But the songs get shared. Users send them to family, play them at events, post them to group chats. Synthetic-sounding output almost never leaves the creator's browser tab.
Story input is the product, not the genre picker. We initially placed genre selection at the top of the creation form. Session recordings showed people spending far more time crafting their story text than choosing between pop and jazz. We moved genre below the story field and saw completion improve.
The dominant use case is gifts. Birthdays, anniversaries, weddings, retirements. The most common pattern we see is someone creating a song for someone else. We designed the shareable link around this: the recipient opens a direct link, hears the song, no account required.
What we track weekly
Three metrics tell us whether the product is working:
Completion rate. How many users who start the creation flow actually finish and pay. Friction in the form shows up here within days of a change.
Share rate. How many finished songs get opened via the shareable link by someone other than the creator. This is the closest proxy we have for output quality.
Return rate. Repeat purchases say more about product-market fit than any single satisfaction score. A user who orders a second song for a different occasion validated the product harder than a 5-star rating.
We are not publishing exact funnel percentages yet. The product is early stage and the flow changes week to week as we test pricing and onboarding variations.
Why we publish this
Inithouse builds products and measures what happens. We share operational numbers like these because the AI music space runs on vague claims and demo reels. Specific, verifiable data points are more useful to anyone evaluating the category or building in it.
Try it yourself: magicalsong.com
Top comments (0)