LALAL.AI
LALAL.AI separates vocals and instruments in music recordings. We worked on the project from the first version through the development of a product range: designing core workflows and interfaces, and creating its visual identity and marketing materials. As the project grew, we developed batch processing, the player and sign-in, optimised subscription models, explored new product directions and brought solutions to other platforms.
Services
- Product research
- Interviews, surveys, testing workflows and demand for new tools.
- UX and interfaces
- File upload and batch processing, the player, sign-in, payments and limits. Subscription model optimisation.
- Brand and visual identity
- Brand platform, logo, colour, graphic language and technical illustrations.
- Marketing design
- Product pages, email marketing, Google and Meta ad campaigns, social media, editorial materials and app store listing assets.
- Design system
- Shared tokens and interface principles for web, desktop, mobile apps and a DAW plugin.
- New product directions
- Research and development of new products and use cases, including Voice Cleaner, Voice Changer and Voice Cloner.
From the first version to a product range
- audio separation operations in 2025. Company report for 2025.
- +63.8m
- registered users by the end of 2025
- +6.8m
In 2020, the founders were testing a hypothesis: AI audio separation could be useful to music producers and sound designers. There was no product yet. We needed to show how the technology worked on real recordings and find out whether people would pay for the results. The figures alongside show the product's scale by the end of 2025, not the effect of an individual interface change.
The first product and brand
The first LALAL.AI user experience
LALAL.AI multitrack player
For the first version, we designed a short path: upload a file, listen to the separated tracks and download the result or receive it by email. Registration was optional, so people could immediately check the quality on their own recordings.
Free previews used server resources before payment. We kept that trade-off: people needed to hear the result before deciding to pay. The same principle later guided more complex workflows.
In January 2021, the company reported more than 3,000 uploads a day and 70,000 tracks processed in the previous month. Source.
Announcement graphics: paid packages.
Announcement graphics for the model Cassiopeia.
Announcement graphics: video formats.
Announcement graphics for key detection.
Editorial graphics: AirStems.
Editorial graphics: Utah Carol.
Brand and marketing design
We developed the brand platform and the foundations of the visual identity: a logo, yellow, technical illustrations and simple shapes. For early ads, we proposed removing the music from familiar videos and leaving the vocals. This made the service's capabilities audible in a concrete example.
The work covered product pages, email marketing, Google and Meta ad campaigns, social media and editorial materials. We also prepared assets for app store listings. For user stories, we established a shared approach to photography and typography. As new features appeared, we developed the brand's graphic language and prepared launch materials.
Developing the core workflows
Batch upload
When people needed to process whole folders, running every file for free would have been too expensive. We designed a flow with one result available without registration: a person uploads a folder, receives the first processed file and creates an account for the rest.
The total duration of the files was known only after upload. The interface showed the remaining minutes, which files could be processed now and where an extra payment was needed. We designed the queue and progress to survive sign-in and payment, so people did not have to start again.
After launch, the number of files and the amount of audio processed per session increased, along with purchases of minutes. We developed the feature with users and collected their feedback. The exact figures are no longer available.
Pricing changes. At first, people bought packages of minutes. As the project grew, we worked with the product team to test and optimise subscription models: a subscription and a subscription with top-ups, including variants for individual markets. Each change required updates to upload, payment and limit displays.
Listening and track controls
Previously, tracks could only be played separately. We worked on a player with simultaneous playback and level controls so people could assess the mix in the preview. The interface needed to stay simple and work across web, desktop and mobile apps.
Interviews and a survey of more than 1,000 users helped refine level and stereo-channel controls, preview length and seeking. Seven of eleven hypotheses were validated; the work went through more than 30 iterations of designs and prototypes.
98% of responses to the feedback form after release were positive. We could not collect product metrics for this release. Designer’s case study.
Sign-in and saved files
Regular users needed access to their results. An email link could open another tab or browser and interrupt work on uploaded files. We replaced it with a six-digit code entered in the same session.
We designed code expiry and resending. Social sign-in options were made equally prominent and ordered using usage data.
| Sign-in method | Change in the number of registrations |
|---|---|
| Apple | +37% |
| +28% | |
| +31% | |
| +6% | |
| All methods | +20% |
These figures describe changes in the number of registrations after release.
The Voice Cleaner interface.
Voice Cleaner
Interviews and Reddit communities revealed demand for removing noise from voice recordings. Upload data also showed that people used the service for recordings without music. We proposed a separate tool, Voice Cleaner, and designed its core workflow. To test demand, we prepared a product page and promotional materials. Usage and revenue confirmed demand.
Design system and new platforms
For web, desktop, iOS, Android and the DAW plugin, we prepared shared tokens and interface principles. These helped carry solutions between platforms and maintain them through updates. Work on individual workflows became a foundation for developing apps on new platforms.
Testing new directions
We researched and developed new products and use cases, including Voice Cleaner, Voice Changer and Voice Cloner. We released minimal versions, studied usage and purchases, and interviewed selected user groups. This helped decide which features to refine and which ideas to test next. The product reached break-even.
Testing hypotheses. To attract users, we added bonuses for following our social accounts, plus affiliate and referral programmes. We also tested new product ideas.
An embeddable widget for uploading and previewing files on other websites and a separate track-mastering service did not pass validation and were not added to the product.
Results
2021 in review: processed tracks and registered users.
2022 in review: users, processed audio and model development.
Development by year
The annual reports use different metric definitions. This table shows scale and product development, rather than a consistent trend in active users.
A simple path to the result
- Period
- 2020–2026
- Platforms
- Web, desktop, iOS, Android, DAW plugin
