LALAL.AI

Starter includes 10 minutes in the Relaxed Queue and accepts files up to 200 MB. You can upload a track, pick a separation type and listen to previews of the resulting stems. Downloading the full result is not included on Starter, according to the homepage and desktop app FAQs, so the free plan is a listening test rather than a way to obtain a finished file.

The refund policy says audio quality on free and paid versions is identical, and it asks buyers to judge the preview first. That makes the free plan the right place to decide which separation type and neural network setting suit a given mix. The public pages do not say whether a preview alone uses Starter minutes, so check the balance before and after.

Minutes are deducted as total file length multiplied by the number of separation types selected. One separation type returns a pair of stems, such as vocals and instrumental, or drums and a drumless track. The homepage example is a 5-minute file split into Drums, Piano and Vocal/Instrumental, which costs 15 minutes.

Voice Changer is charged differently: minutes follow the duration of the generated output. For stem work, choose only the separation types you will actually use, because a three-way split of a full album quickly outgrows a monthly Fast allowance. Uploading a trimmed section of a song rather than the whole file also reduces the charge.

Both paid plans include unlimited processing in the Relaxed Queue, which runs as server capacity allows. The difference is priority time: Lite includes 90 Fast Queue minutes a month and Pro 250. Processing starts in the Fast Queue at the beginning of each period and switches to Relaxed automatically once the Fast minutes run out; you cannot choose the queue manually.

Editorial assessment

Lite suits someone who downloads a few separations a week and can wait when the queue is busy. Pro is justified by features rather than minutes alone: the VST plugin, offline Lyra processing and API access are Pro-only. For a single large batch, a one-time Top-Up of Fast minutes can be cheaper than moving up a tier for a month.

• Starter: previews only, 10 Relaxed minutes, 200 MB per file.
• Lite: downloads, batch processing, 90 Fast minutes a month, 1 voice pack slot.
• Pro: 250 Fast minutes, VST plugin, Lyra local processing, API access, 3 voice pack slots.

Lyra is the on-device model in the desktop app, available to Pro subscribers. The desktop FAQ says local processing works without an internet connection, does not upload audio to LALAL.AI servers and does not deduct processing minutes. An internet connection is still needed occasionally to validate the subscription.

This matters for unreleased music, client stems or confidential dialogue that should not leave the machine. By default the desktop app processes in the cloud like the website, so switch to local mode deliberately before loading sensitive material. The desktop FAQ presents it as a desktop advantage over the browser version.

The privacy policy says OmniSale GmbH does not assume rights to uploaded audio and video, does not use user files for AI training and does not share uploaded media with third parties. The terms place responsibility for using and distributing both the uploaded files and the resulting stems entirely on the user.

Separating a commercial song is technically possible, but publishing an isolated vocal, a remix or a karaoke track still needs permission from the rights holders. The terms also describe voluntary compliance with the DMCA takedown process. Voice Cloner can copy a real voice, so clone only your own voice or one whose owner has given explicit permission.

The refund policy treats payments as generally non-refundable once minutes are used. Refunds may be granted at the provider’s discretion for unresolved technical errors, duplicate purchases with unused minutes or payment processing errors. Processing a file in full counts as accepting the quality shown in the preview.

The statutory 14-day withdrawal right applies only before the plan is activated; uploading a file for processing starts the service and ends that right. Cancelling a subscription keeps paid features until the end of the billing period, after which the account returns to Starter and any remaining Fast minutes expire.

1. Note the Starter minute balance before uploading.
2. Upload the file and preview the vocal and instrumental stems.
3. Change the neural network in the widget settings and regenerate the previews once.
4. Listen for vocal remnants in the instrumental and instrument bleed in the vocal, especially in the reverb tails.
5. Record the minute balance again after both previews.

Accept previews where the instrumental carries no audible lead vocal and the balance change matches the formula. Reject previews with clear vocal bleed, or a balance drop that the formula does not explain. No separation quality or minute result is reported until the exercise is run.

For cleaning spoken recordings rather than music, compare Adobe Podcast and Auphonic. For generating new synthetic speech rather than transforming an existing recording, see ElevenLabs.