At a glance
Luvu is a mobile fitness app built around Luvu, an animated marshmallow trainer who coaches users through home workouts with realtime voice motivation. Co-founded by Alexis Sursock and Creston Brooks, the app uses computer vision to track exercises through the phone's camera and pairs it with an entertaining, personalized trainer voice that counts reps, calls out names, and keeps users coming back. The product runs in English, Latin American Spanish, and Brazilian Portuguese. Voice is central to the experience: when a user is thirty glute bridges in and thinking about stopping, Luvu is what makes the difference.
“Inworld reduced our TTS costs by about 10x, while remaining neutral or better on all the metrics we care about, which is pretty incredible.”
Creston Brooks, Co-founder & CTO, Luvu
The problem
The cost of voice at the center of your product
For Luvu, TTS is not a feature. It is the core of the product. Luvu's coaching voice runs continuously through every workout, which made TTS one of the company's largest infrastructure costs.
Luvu had been running on their previous TTS provider, partly through a startup grant that was running out. As the app grew, the unit economics stopped working. Sustaining a free tier at scale required a significantly cheaper provider. But switching voices on a character users had already formed a relationship with carried real retention risk: any drop in workout completion, time spent exercising, or chat engagement would signal that the new voice wasn't landing.
The technical requirements narrowed the field further. Luvu needed a custom voice distinct to Luvu's character in three languages, plus word-level timestamps to drive realtime viseme lip-sync on the character animation. Getting both together, from one provider, was the actual constraint.
The solution
Luvu's voice in three languages, with visemes
Inworld worked with the Luvu team to design Luvu's character voice in English, Latin American Spanish, and Brazilian Portuguese. Each voice shipped with word-level timestamp alignment for phoneme-to-viseme mapping, keeping the character's animated mouth movements in sync with generated speech.
Luvu ran a careful, phased rollout: new users on Portuguese and Spanish first, then English, each language in a controlled A/B experiment against the previous provider's baseline. Over several months, Inworld iterated on the voices, improving naturalness, expressiveness, and multilingual consistency, until every language met Luvu's engagement benchmarks.
Use case 01
Realtime workout coaching
During workouts, Luvu counts reps, calls users by name, and delivers mid-set encouragement. The voice runs through a mix of pre-cached phrases and dynamically generated callouts, with character alignment timestamps driving the animated lip movements. Luvu serves the workout voice at 1.15x speed to match the energy of a live training session, across all three supported languages.
Use case 02
In-app chat with Luvu
Between and after workouts, users chat with Luvu directly. These conversational responses are generated dynamically and served realtime. Chat message volume is one of Luvu's core engagement metrics. After fully switching to Inworld, it came in flat against the previous provider's baseline, alongside workout completion, time spent exercising, and free-to-paid conversion.
“We have many users with 100+ day streaks who say that this is the first fitness habit they've kept in their lives. This is only possible due to the realtime workout magic that is delivered through Inworld TTS!”
Alexis Sursock, Co-founder & CEO, Luvu
Results
Luvu tracked five core engagement metrics across the A/B tests: time spent exercising, workouts started, workouts completed, chat messages sent, and free-to-paid conversion. After fully migrating all new users across English, Spanish, and Portuguese, every one came in neutral or slightly better compared to the previous provider's baseline. Workouts completed, free-to-paid conversion, and chat messages sent all trended slightly higher. The cost was 10x lower.
10x
Reduction in TTS cost
5/5
Core engagement metrics maintained or improved
What’s next
Scaling with continuous experimentation
As Luvu scales, the team plans to bring Inworld's Router into the stack for autonomous live experimentation, testing models, prompts, coaching copy, and UI variations without manual A/B setup for each change.
The partnership moves from a one-time voice migration to a continuous optimization loop, the next experiment in an ongoing data-driven collaboration.
