Omni Sale GMBH, registered at Grundstrasse 10 in Risch-Rotkreuz, Switzerland, operates LALAL.AI as an audio and video processing service that began in 2020 with a neural network called Rocknet trained on 20TB of data to pull instrumentals and voice tracks apart. The about page of LALAL.AI records the growth from a two-stem splitter into what the vendor calls "the world's first 8-stem splitter" in 2021 and "the world's first 10-stem splitter" in 2022, and states that registered users rose from 3.8 million in 2024 to 6.79 million in 2025. Today LALAL.AI bundles Vocal Remover, Stem Splitter, Voice Cleaner, Echo & Reverb Remover, Lead/Back Vocal Splitter, Voice Changer and Voice Cloner under one account.
The people LALAL.AI is aimed at are working with finished mixes rather than raw multitracks: cover artists and karaoke makers who need an instrumental, producers lifting drum or bass parts for sampling, podcast editors stripping room tone and background music off dialogue, transcribers and journalists rescuing noisy interview tape, and localization studios that need clean speech and ambience before dubbing. The about page from Omni Sale GMBH names musicians, sound producers, music engineers, video bloggers, streamers, transcribers, translators and journalists as the intended audience for LALAL.AI.
A browser session with LALAL.AI starts by choosing a stem separation type in the upload widget, dropping in a file, and waiting for preview snippets to appear. LALAL.AI accepts MP3, OGG, WAV, FLAC, AIFF, AAC and M4A audio plus AVI, MP4, MKV, MOV and M4V video, and the output format can be changed with the arrow next to the Split in Full button before full processing begins. The settings icon in the corner of the LALAL.AI upload widget exposes Lead/back separation, De-echo, Noise Canceling Level, and a neural network picker, so a track that separates poorly on the newest engine can be regenerated on an older one.
Model choice is a genuine part of the LALAL.AI workflow rather than a footnote. The llms.txt file published by LALAL.AI names v6 Andromeda as the flagship music separation engine, v7 Lynx as a lightweight specialist for Voice & Noise and Voice Cleaner, and Lyra as the on-device model behind the desktop app and the VST/AU plugin, where nothing is uploaded and no minutes are spent. Earlier generations named v5 Perseus, v4 Orion and v3 Phoenix still handle less common stems inside LALAL.AI, and the vendor suggests Orion for noisy vintage recordings.
Packaging in LALAL.AI is metered in minutes rather than exports. The LALAL.AI pricing page lists Starter as always free with 10 Relaxed Queue minutes, Lite at €6.75 per month billed annually (€81 a year), and Pro at €13.5 per month billed annually (€162 a year); the homepage FAQ of LALAL.AI gives the month-to-month stickers as €8.99 and €17.99. Paid LALAL.AI plans add unlimited Relaxed Queue minutes plus a fixed Fast Queue allowance of 90 minutes on Lite and 250 on Pro, and Pro alone unlocks the VST/AU plugin, local desktop processing and API access. One-time Fast-minute packs of 750, 3,000 and 5,000 sit alongside the subscriptions, and businesses can request a custom quote from LALAL.AI through the enterprise form.
The limits of LALAL.AI matter before a card is entered. Minutes in LALAL.AI are deducted as file length multiplied by the number of selected stem types, so a five-minute song split three ways spends fifteen minutes of balance. The free Starter plan of LALAL.AI allows online listening only with no full downloads and caps uploads at 200MB per file, unused Fast Queue minutes do not roll over, and files longer than 20 minutes can only run in the Fast queue. The refund policy of Omni Sale GMBH states that amounts paid for LALAL.AI are generally non-refundable and that processing a file in full counts as agreement with the quality shown in the preview.
Against category peers, LALAL.AI sits between one-click browser splitters and studio restoration suites. Where a plugin such as iZotope RX lives entirely inside a DAW and where Moises leans on practice features like chord detection and tempo change, LALAL.AI concentrates on separation breadth (ten stem types), delivery breadth (web, desktop, iOS, Android, VST3 plugin, REST API), and voice work that spans cleaning, changing and cloning in the same account. The in-house model story is also part of the LALAL.AI pitch: the vendor states that every network is its own technology trained on proprietary datasets rather than a wrapper over third-party models.
Buyers evaluating LALAL.AI should map their volume to the minute formula first, because the difference between casual karaoke work and a dubbing pipeline is measured in Fast minutes and API calls rather than seats. A hobbyist can test LALAL.AI quality for free before signing up, a weekly content creator will likely live on Lite, and a studio that needs the plugin, offline Lyra processing, batch jobs and REST access has to be on Pro or buy top-up packs. Voice Cloner access in LALAL.AI is also tied to subscription Voice Pack Slots, with one slot on Lite and three on Pro.