SongRise: AI-generated affirmation songs sung in your own voice.
Generic affirmations don't stick. The brain hears them as someone else's words. SongRise records 60 seconds of you singing, clones your voice, and generates affirmation songs in that voice. The first song lands in roughly 60 more seconds.
Live at songrise.app.
Screenshot from songrise.app, captured 2026-05-06.
The problem
The self-improvement category has been selling affirmations for thirty years. Apps and journals and audiobooks. Most of it doesn't stick. The brain treats words written by a stranger or read by a stranger as low-priority information.
The known psychological lever is self-referential encoding. Information processed about yourself, by yourself, sticks better. Singing activates both hemispheres of the brain at the same time and adds emotional encoding on top. SongRise's product hypothesis: marry the two and the affirmation lands.
The ICP is clear and underserved: women 25 to 45 who already buy self-improvement products, who already journal or meditate, who want a daily ritual that does not feel like another generic app.
What Downshift built
Nine surfaces wired around the moment a user hears themselves sing for the first time. The "holy shit that's me singing" moment is the conversion event.
60-second voice capture
Browser-based singing recorder using the MediaRecorder API. No app install. The full first-song flow happens in one tab.
Voice clone via Mureka
60 seconds of singing produces a vocal_id from Mureka's voice-cloning API. That id keys every future song the user generates.
Song generation in ~60s
AI composition + voice synthesis through Mureka. The first song lands roughly a minute after the upload finishes. Two song variations per generation.
10 affirmation themes
Curated themes (confidence, gratitude, self-love, abundance, etc.) so users do not face a blank-prompt problem on day one.
Personalized lyrics
Lyrics generated from the user's chosen theme and goals, then sung in their cloned voice. Self-referential at every level of the artifact.
Weekly delivery
Subscribers receive a fresh song every week without lifting a finger. The product becomes a calendar habit.
Streamable + downloadable MP3
Account-bound song library. Stream from the web app or download MP3 to keep on a phone for the gym, commute, or shower.
Free first song
Record voice, generate one song, no credit card. The first song is the demo. Conversion happens because hearing the artifact sells itself.
Stripe subscription tiers
$2.99 single song, $7.99 monthly (unlimited + weekly), $49.99 annually (~$4.17/mo effective). All standard, cancel any time.
Screenshot from songrise.app, captured 2026-05-06.
The architecture
Queue-first because every step is a 60-second-class media job. Browser-only capture so the day-1 flow does not require an app install.
Stack
- Bun monorepo with five apps
- Fastify 5 API (port 6900)
- React + Vite + Tailwind + shadcn web (port 6910)
- Astro SSG marketing site
- BetterAuth + Google OAuth
- Stripe billing
AI surface
- Mureka API for voice cloning (vocal_id from 60s capture)
- Mureka API for song generation (composition + synthesis)
- Two variations per generation by default
- Self-referential encoding as the product hypothesis
Data + queues
- PostgreSQL (Drizzle ORM)
- BullMQ + Redis (DB 18) for the generation pipeline
- Cloudflare R2 for media (shared bucket, songrise/ prefix)
- PostHog + Sentry for analytics + errors
Apps + roadmap
- api: Fastify backend
- web: React SPA day-1 flow
- marketing: Astro pSEO site
- admin: ops dashboard
- mobile: Expo phase 2 (after web cohort)
Real outcomes
Numbers below are from songrise.app's live marketing copy and from the production codebase. No fabricated user counts.
Singing recorded in the browser to clone the user's voice.
From upload to first finished song in the user's library.
Affirmation themes available at launch.
Song variations returned per generation.
Single-song one-time price (2 variations).
Monthly subscription. Unlimited songs, weekly delivery, all 10 themes.
Annual subscription. Effective ~$4.17/mo.
Deeper affirmation retention claimed for self-sung audio versus reading or listening (page-stated; user studies in progress).
Sources: pricing tiers ($2.99 / $7.99 / $49.99), 60s capture, ~60s generation, 10 themes, 2 variations, and the 2x retention claim are quoted from songrise.app. Stack and pipeline counts come from ~/Github/SongRise/CLAUDE.md. Per-user MAU and revenue are not on the public marketing page and are shared on request.
Screenshot from songrise.app, captured 2026-05-06.
What's in flight
Phase 2 is the Expo mobile app, scheduled to land after the web cohort proves retention against weekly delivery. The shared monorepo means the same auth, the same library, the same Stripe customer. The next theme expansion goes from 10 to 20+ once retention data isolates which themes actually convert subscribers into renewals.
Want this for your venture?
Downshift partners with non-technical founders to ship production-grade AI products in 3 to 6 weeks. Fixed scope, no equity.
See more shipped Downshift ventures on the wins page, or read the RepairSnap, FairStack, and Rivalize case studies.