LALAL.AI

Tool · this deep dive is built from the official pages of the platform itself and of the tools that support it.

Overview

LALAL.AI separates and cleans audio. Vocal Remover takes off the vocal and the instrumental, Stem Splitter lays a recording out into vocal, instrumental, drums, bass, guitar, synthesizer, strings and wind instruments. Voice Cleaner removes background music, plosives and microphone rumble, and there are separate Echo & Reverb Remover and Lead/Back Splitter — the separation of lead and backing vocals.

Next comes the part for which this review is worth reading to the end: Voice Changer and Voice Cloner. The cloning builds a voice model, which the tool calls a voice pack, from uploaded recordings — up to five samples, as stated in the questions block of the voice cloning page — and that model is then applied in Voice Changer. The operator of the tool is OmniSale GMBH.

There is no separate list of languages. The product page declares: "It doesn't matter what language you speak or what accent you have - we support them all!" — verbatim. This is a claim on a product page without a list; confirmation for Russian, Arabic, Hindi and Chinese is not in what was read. The interface of the site itself switches to English, simplified and traditional Chinese, French, German, Italian, Japanese, Korean, Spanish, Dutch, Portuguese and Brazilian Portuguese — but that is the language of the site, not a confirmed coverage of the model.

Which platforms it works with

Social networks are absent from this list entirely. The points of work at LALAL.AI are these: the web, desktop applications for Windows, macOS and Linux, applications for iOS and Android, a VST plugin working locally inside a DAW, and an API with a published OpenAPI specification of version v1.

This differs by plan: both the VST plugin and the API are available only on the Pro plan — confirmed by the questions block of the pricing page. The balance of minutes, meanwhile, is common to all clients at once: spend on the phone and it decreases in the web.

Direct export to social networks, to streaming services or to podcast hosts is not in what was read. The output is a file that you distribute yourself.

Pricing

The prices are from the pricing page, the currency of the subscriptions is euros.

Lite — 6.75 EUR a month with annual payment, which gives 81 EUR for the year. The plan includes unlimited minutes in the Relaxed Queue, 90 minutes of Fast Queue a month, files up to 2 GB and batch processing.

Pro — 13.5 EUR a month, that is 162 EUR for the year: unlimited Relaxed Queue, 250 minutes of Fast Queue a month, files up to 2 GB, API access, the VST plugin, local processing in the desktop application and early access to new functions.

Both prices are shown in a monthly form with the note billed annually. A separate monthly payment scheme the page does not disclose — that is, 6.75 EUR is not "the price for a month" but the annual bill divided by twelve.

Besides the subscriptions there are one-off Fast Queue top-ups, verbatim: "Master with 750 minutes for $50, Premium with 3000 minutes for $190, and Enterprise with 5000 minutes for $300" — here the currency is already dollars. There is also a separate corporate offer.

The cheapest subscription step is 6.75 EUR a month with annual payment. The cheapest one-off purchase is the Master top-up for 50 USD, but that is not a subscription.

What is free

The Starter plan is marked "Always free" — permanent, not a trial.

It includes 10 minutes in the Relaxed Queue, a file size limit of 200 MB and — which is substantial — downloading of the results. That is, the free plan does not show the result merely for viewing but hands over the file.

What it does not include: Fast Queue minutes, batch processing, API access and early access to new functions.

Ten minutes is not ten minutes of work but ten minutes of deduction, and they are counted by the rules from the next section, where one and the same recording can cost a multiple of its own duration.

Restrictions

Minutes are deducted by the duration of the processed file, but not always one to one. For Stem Splitter the formula is direct: "total file length x number of stem separation types" — the duration is multiplied by the number of requested stems. For Voice Changer the minutes are counted by the duration of the generated result.

The quotas: Starter — 10 minutes of Relaxed Queue, files up to 200 MB. The paid plans — unlimited Relaxed Queue and a fixed Fast Queue: 90 minutes a month on Lite, 250 on Pro, files up to 2 GB.

Unspent Fast Queue minutes do not carry over and are zeroed at the start of a new billing period.

The queue cannot be managed: switching it by hand does not work, processing goes through Fast Queue until the limit is exhausted and then passes automatically to Relaxed.

The balance is common to the web, the desktop, iOS and Android.

After cancellation the access and the minutes are retained until the end of the paid period, then the account returns to Starter and the Fast Queue remainder burns up. Verbatim about refunds: "The subscription is non-refundable".

Rights to the output

The answer is incomplete, and that incompleteness matters more than any quotation.

The terms do NOT transfer the rights to the output to the user by a direct wording and do not describe differences by plan. Only one thing is said: "You are solely responsible for the usage and distribution of uploaded audio files and end-result audio files" — verbatim, responsibility for the use and distribution of both what is uploaded and what is obtained lies with the user.

In the same document there is a general prohibition: "You agree not to reproduce, duplicate, copy, sell, trade, resell or exploit for any commercial purposes, any portion of the Service, use of the Service, or access to the Service unless you otherwise have an agreement with us which specifically grants you such right(s)" — verbatim. It is worded about the Service itself, the use of it and access to it, not about the output. But a separate permission for commercial use of the output is also not in the text, so the wording admits a double reading.

The privacy policy adds from the other side: "the OmniSale GMBH does not assume the rights to any uploaded audio and video data" — the operator does not appropriate rights to what is uploaded.

The upshot: the tool makes no claim on your files, but it does not formally issue you rights to the output either.

Do they train on your data

No — and this is the rare case where the tool answers directly.

The privacy policy verbatim: "the OmniSale GMBH does not assume the rights to any uploaded audio and video data, does not use user files for artificial intelligence training or other content". Training AI on user files is denied in the same sentence as the disclaimer of rights to them.

Where exactly this prohibition stands is material: in the privacy policy, not in the terms of use. In the Terms of Service of the 30.07.2026 edition the subject of training is not touched upon at all.

A mechanism for opting out is not described and by its sense is not needed: there is nothing to opt out of, since the training is denied. This is the absence of a procedure, not the absence of an answer.

How you earn with it

With commercial use of the output the situation is described above and in short looks like this: a direct permission is not in the terms, a direct prohibition on the output is not there either, and the prohibition that does exist, about commercial exploitation, is addressed to the Service and access to it.

What relates to earning directly is the duty of disclosure. The terms contain a section "AI Transparency, Synthetic Content, and Regulatory Compliance" with a reference to Regulation (EU) 2024/1689. It says that the output of the generative functions "may legally constitute deepfakes or synthetic content", and then verbatim: "You are solely responsible for clearly and distinguishably disclosing to the end audience that the audio content has been artificially generated or manipulated". It is you who are obliged to mark it before the audience.

Indirectly it also follows from there that some marks are embedded into the output: "You shall not remove, obscure, tamper with, or attempt to disable any machine-readable marks, metadata, or watermarks embedded into the outputs" — verbatim. Which marks exactly, and whether they differ by plan, is not described.

What it will not do

The main thing here is what is absent from the documents.

A SPECIAL RULE ABOUT THE CONSENT OF THE PERSON WHOSE VOICE IS BEING CLONED IS ABSENT FROM WHAT WAS READ. In the terms there is neither a requirement to obtain the speaker's consent, nor a prohibition on cloning the voice of a public figure or a celebrity, nor a description of any verification before a voice model is created.

Meanwhile the voice cloning page describes Voice Cloner as a model built "from your own recordings and audio samples", yet among the scenarios it directly offers "Clone the voice of a loved one to narrate personal stories or family history" — verbatim, cloning the voice of a close person — without stipulating that person's consent by a single word.

This is the absence of a rule, not a permission. From the document's silence it does not follow that cloning someone else's voice is allowed: it follows only that the tool has written nothing on the matter.

The only norm bearing on the case is a general prohibition: "You are strictly prohibited from using Services to impersonate individuals without legal authorization or to falsely represent synthetic audio as authentic human speech" — verbatim. Forbidden are impersonation without legal authorization and the passing off of synthetic audio as authentic human speech. Uploading someone else's voice as such is not brought under this clause.

A separate list of prohibited content topics is ABSENT from what was read.

What the tool forbids directly: removing and tampering with machine-readable marks, metadata and watermarks on the output, and failing to observe applicable law.

Conceal from the audience that the audio is synthetic it will not let you either — but by your own effort: the duty of disclosure it shifts onto you.

Publish the result for you it will not be able to — export to social networks and to streaming the tool does not declare.

Refund the subscription it does not promise: "The subscription is non-refundable".

Verified data

The checked data this deep dive rests on.

About the platform

Separating and cleaning audio: Vocal Remover (removing vocals and instrumental), Stem Splitter (vocals, instrumental, drums, bass, guitar, synthesiser, strings and wind), Voice Cleaner (removing background music, plosives and microphone rumble), Echo & Reverb Remover, Lead/Back Splitter (separating lead and backing vocals), Voice Changer and Voice Cloner — building a voice model (a voice pack) from uploaded recordings, up to five samples, then applied in Voice Changer. The operator is OmniSale GMBH

The product list is taken from the site navigation, and the description of Voice Cloner and the five-sample limit from the FAQ block of the cloning page

source, checked 2026-08-07

There is no separate list of languages. The product page claims «It doesn't matter what language you speak or what accent you have - we support them all!» (verbatim) — that is a claim on the product page with no list of languages, and there is no confirmation for Russian, Arabic, Hindi or Chinese in what was read. The site interface switches to English, simplified and traditional Chinese, French, German, Italian, Japanese, Korean, Spanish, Dutch, Portuguese and Brazilian Portuguese

The interface languages are taken from the language switcher in the site header; this is the language of the site, not a confirmed coverage of the model

source, checked 2026-08-07

Platforms

Clients and integration points: web, desktop for Windows, macOS and Linux, apps for iOS and Android, a VST plugin that runs locally inside a DAW (Pro plan only), and an API (Pro plan only) with a published OpenAPI v1 specification. The minute balance is shared across all clients. There is no direct export to social networks, streaming or podcast hosts in what was read

The VST plugin and API being limited to the Pro plan is confirmed by the FAQ block of the pricing page; the specification is at https://www.lalal.ai/api/

source, checked 2026-08-07

Pricing

Starter — «Always free»: 10 minutes in the Relaxed Queue, a 200 MB file size limit, downloading results available. Not available: Fast Queue minutes, batch processing, API access and early access to new features

The composition was read from the plan comparison table on the pricing page

source, checked 2026-08-07

Lite — 6.75 EUR a month paid yearly (81 EUR a year): unlimited minutes in the Relaxed Queue, 90 Fast Queue minutes a month, files up to 2 GB, batch processing. Pro — 13.5 EUR a month (162 EUR a year): unlimited Relaxed Queue, 250 Fast Queue minutes a month, files up to 2 GB, API access, the VST plugin, local processing in the desktop app, early access to new features. One-off Fast Queue top-ups, verbatim: «Master with 750 minutes for $50, Premium with 3000 minutes for $190, and Enterprise with 5000 minutes for $300». There is a separate corporate offering

The Lite and Pro prices are shown per month with the note billed annually; no separate monthly billing scheme is disclosed on the page

source, checked 2026-08-07

6.75 EUR a month — the Lite plan paid yearly, which comes to 81 EUR a year. No separate monthly price is shown on the page that was read

The cheapest one-off purchase is the Master top-up of 750 minutes for 50 USD, but that is not a subscription tier

source, checked 2026-08-07

Limits and restrictions

Minutes are deducted by the length of the processed file; for Stem Splitter the formula is «total file length x number of stem separation types», and for Voice Changer minutes are counted by the length of the generated result. Starter — 10 minutes of Relaxed Queue, files up to 200 MB. Paid plans — unlimited Relaxed Queue and a fixed Fast Queue: 90 minutes a month on Lite, 250 on Pro, files up to 2 GB. Unspent Fast Queue minutes do not carry over and are zeroed at the start of a new billing period. The queue cannot be switched manually: processing goes through Fast Queue until the limit is used up, then automatically through Relaxed. The balance is shared across web, desktop, iOS and Android. After cancellation, access and minutes remain until the end of the paid period, then the account reverts to Starter and the Fast Queue balance is lost; «The subscription is non-refundable»

The rules on queues, carry-over and cancellation are taken from the FAQ block of the pricing page

source, checked 2026-08-07

Restrictions

WHOSE VOICE: THERE IS NO SPECIFIC RULE ABOUT THE CONSENT OF THE PERSON WHOSE VOICE IS CLONED IN WHAT WAS READ. The Terms of Service contain neither a requirement to obtain the speaker's consent, nor a ban on cloning the voice of a public figure or celebrity, nor a description of any check before a voice model is created. The product page describes Voice Cloner as a model built «from your own recordings and audio samples», yet among its use cases it openly suggests «Clone the voice of a loved one to narrate personal stories or family history» (verbatim), without stipulating that person's consent. The only relevant provision is a general ban: «You are strictly prohibited from using Services to impersonate individuals without legal authorization or to falsely represent synthetic audio as authentic human speech» (verbatim) — that is, impersonation without legal authority is forbidden, not uploading someone else's voice as such. In addition it is forbidden to remove or tamper with machine-readable marks, metadata and watermarks on the output, and applicable law must be observed. THERE IS NO separate list of forbidden content topics in what was read

The wording about a loved one is taken from https://www.lalal.ai/voice-cloning/ and is cited as an indication that cloning a third party is openly offered with no caveat about consent

source, checked 2026-08-07

Legal

NO. The privacy policy verbatim: «the OmniSale GMBH does not assume the rights to any uploaded audio and video data, does not use user files for artificial intelligence training or other content». The Terms of Service in the 30.07.2026 revision do not address training at all

The prohibition is recorded specifically in the privacy policy, not in the terms of use

source, checked 2026-08-07

The terms do NOT transfer rights in the output to the user in so many words, and describe no difference between plans. All they say is: «You are solely responsible for the usage and distribution of uploaded audio files and end-result audio files» (verbatim). The same document contains a general prohibition: «You agree not to reproduce, duplicate, copy, sell, trade, resell or exploit for any commercial purposes, any portion of the Service, use of the Service, or access to the Service unless you otherwise have an agreement with us which specifically grants you such right(s)» (verbatim) — it is worded about the service itself and access to it, not about the output, but neither is there any separate permission for commercial use of the result in the text. The privacy policy adds that «the OmniSale GMBH does not assume the rights to any uploaded audio and video data»

Marked partial: the terms contain no provisions on rights in the output, and the wording about commercial purposes admits two readings; the second quotation comes from https://www.lalal.ai/privacy-policy/

source, checked 2026-08-07

The duty to label is placed on the USER. The section «AI Transparency, Synthetic Content, and Regulatory Compliance» refers to Regulation (EU) 2024/1689 (the EU AI Act): the output of the generative features «may legally constitute deepfakes or synthetic content», and further, verbatim: «You are solely responsible for clearly and distinguishably disclosing to the end audience that the audio content has been artificially generated or manipulated». The presence of technical marks is indicated only indirectly: «You shall not remove, obscure, tamper with, or attempt to disable any machine-readable marks, metadata, or watermarks embedded into the outputs» (verbatim) — but exactly which marks are embedded, and whether they differ by plan, is not described

The duty to disclose is confirmed verbatim; the source does not disclose whether marks are embedded or what they consist of

source, checked 2026-08-07

Catalogue section: all similar See also: catalogue index · scheduler comparison · find by situation · platform restrictions