LALAL.AI 2025: Our Biggest Year Yet

As 2025 comes to an end, it’s clear this has been the biggest year so far for LALAL.AI and our community of creators. More musicians, producers, podcasters, video editors, and streamers have been using the service to clean up audio, clone voices, separate stems, and handle everyday audio tasks faster.

Over the past twelve months, LALAL.AI has grown both in numbers and in what it can do. Audio splitting has become more accurate, noise removal has improved, and support for different stems has expanded, helping people spend less time on routine edits and more time on actual creative work.

We invite you to take a look at the key stats, the most popular formats and stems, the main improvements, and the new products that defined this year.

By the Numbers

2025 pushed LALAL.AI to new usage highs, with more processed files and more users than in any previous year, showing clear growth compared to 2024 across all key metrics.

Creators most often turned to LALAL.AI for stem separation and audio clean-up, with vocal-focused splits and core audio formats accounting for the majority of all processed content. The numbers below highlight what our community used most and the year-over-year increase in activity.

Most Popular Stems

• Vocals took the lead among all separation options at 32,291,908 splits.
• Voice ranked second with 11,547,921 splits in total.
• Drums came third with 5,452,872 splits, while Bass is fourth at 2,455,230.
• Electric Guitar took fifth place at 2,054,361 splits.

Most Used Formats

• MP3 remained the top choice for both uploads and downloads at 19,128,688.
• MP4 kept a steady share at 3,076,263, reflecting how often users prepare stems directly from video files.
• M4A saw 2,121,989, FLAC 775,132, and OGG 131,155, as more users opted for higher-quality sources when splitting stems.

Activity Highlights

• Total splits and processed file duration increased noticeably compared to 2024, with 63,811,258 splits and 14,830,175 hours (or 53,388,632,000 seconds), respectively.
• The number of registered users reached 6,793,709, up from 3.8 million last year. That’s 1.79 times more users than in 2024!

Key Updates

This year brought seven major releases and improvements to LALAL.AI, listed below in chronological order from earliest to latest.

1. Desktop App Improvements

In January, LALAL.AI desktop applications received several updates in one go: a modernized UI with dark/light modes, customizable settings (Enhanced Processing, De-Echo, neural network selection), stem-splitting history, output format choice (MP3, WAV, FLAC, etc.), and lead/backing vocal separation into 4 stems. These let users fine-tune results, track past splits (up to 6 months), and process tracks more precisely.

2. Perseus Support Expansion

In February, the Perseus neural network expanded to support 3 additional stems: Acoustic Guitar, Electric Guitar, and Piano. Previously handled by Orion and Phoenix, these now use Perseus’ transformer models for better precision on subtle harmonics and tonal qualities.

3. Voice Cloner Launch

In March, LALAL.AI Voice Cloner launched as a new standalone product. Users upload their voice samples (up to 1 hour total), which the AI analyzes for pitch, intonation, rhythm, and individual traits to build a digital voice clone. Creators now use it for audiobooks, video voiceovers, ads, and vocal experiments, saving time and money by avoiding the need to book a recording studio.

4. SoundCloud Playlist Debut

In May, LALAL.AI started an official SoundCloud playlist to highlight user-created tracks. The playlist features the best submissions made with LALAL.AI tools, giving creators a chance to gain exposure.

5. Mobile Multi-Stem Feature

In July, mobile apps introduced multi-stem separation from a single upload. Instead of uploading the same track multiple times for different stems, users can select vocals, drums, bass, or other stems from the same upload. This cut down on steps and made mobile editing quicker for on-the-go work.

6. Voice Cloner Custom Samples

In September, LALAL.AI Voice Cloner added support for uploading custom samples during previews. In addition to the built-in samples, users can now test their potential voice clones on their own audio before committing to creating a full, paid clone. It reduced trial-and-error and let people hear the exact results with their material right away.

7. Andromeda Network Release

In December, the Andromeda neural network debuted as LALAL.AI’s most advanced model to date. Trained on four times more data than Perseus, it uses transformer technology for up to 40% faster processing and 10% better SDR (Signal-to-Distortion Ratio) on benchmarks. Now the default in Stem Splitter (on Vocal & Instrumental + Voice & Noise stems), Voice Cleaner, Lead & Back Vocal Splitter, and Echo & Reverb Remover, it reduces vocal bleed, reverb artifacts, and high-frequency loss while handling softer/louder tracks consistently.

Looking Ahead

2025 was quite a productive year for LALAL.AI—record usage, new tools, and steady improvements across every platform. Our team delivered several key releases that raised precision, speed, and variety in audio processing, and grew in size to ensure the timely implementation of user‑requested features and the realization of our own ideas that bring audio source separation to a new level.

We’ve always followed a simple rule: under-promise and over-deliver. What we can say with confidence is that 2026 will bring at least two major milestones. First, the release of the LALAL.AI VST plugin, which will allow users to take advantage of our technology directly inside their DAWs, locally and without an internet connection. Second, Andromeda will expand to support more stems, delivering higher-quality separation not only for vocals and speech but also for individual instruments.

Here’s to another year of progress, new ideas, and great results together!

Follow LALAL.AI on Instagram, Facebook, Twitter, TikTok, Reddit, and YouTube to keep up with all our updates and special offers.