Rights to the output and training on data

Four questions a person asks themselves after the tool has already made something: who owns what was made and whether it can be traded; whether models are trained on what I uploaded; whether that can be opted out of; whether the service marks the work of AI.

Here are the answers of 51 tools — as the tools themselves wrote them, with a link to the document.

This is not legal advice and not interpretation. We do not reduce a licence to «may / may not» where the document itself does not, and we do not decide for you what a given wording means.

Who owns the result

What matters here is not the tool but the plan. Of 44 answers, 27 say outright that the rights depend on the plan. A person tries it for free, does the work — and learns only later that it cannot be sold.

Rights depend on the plan — 27

Auphonic. Rights stay with the user: «you fully retain any rights that you have with regards to your Content, including copyright» (verbatim). The terms place no restriction on commercial use of the output, and there is no difference between plans in rights either. The practical difference of the free plan is technical rather than legal — an… source, checked 2026-08-07

Bria. Differs by plan, and this is written in the terms verbatim: «Certain Services, including but not limited to API access, advanced AI model features, and commercial use rights, are available exclusively to users with active paid subscriptions or valid licensing agreements». The pricing page adds for Enterprise «Commercial use rights for… source, checked 2026-08-07

Cleanvoice AI. Rights to the output stay with the user. Terms & Conditions clause 7.3 verbatim: «you also retain all right, title and interest in and to the Results generated through the Services». The licence to the service (clause 7.4) is granted solely to deliver the service ordered. There are no restrictions on commercial use of the output in what… source, checked 2026-08-07

Elai.io. Sources diverge: Here there is a contradiction between the product and the terms, and it has to be stated plainly. Terms of service, clause 2a: «Elai grants you a non-transferable, non-exclusive, revocable, limited license to access and use the Website and Services for you own personal, non-commercial use» and further «You shall not license, sell, rent… source, checked 2026-08-07

Fliki. Differs by plan, and that is the main point. The terms: the user «retain any and all of Your rights to any Content You create, post, or display on or through the Service» (verbatim). But the commercial licence is a paid feature: the FAQ on the pricing page says «with our subscription plans you get a commercial license to all the content… source, checked 2026-08-07

getimg.ai. Differs by plan: verbatim «Can I use created images for commercial projects? Yes, but only with our paid plans. Images generated under these plans can be used commercially, adhering to the CreativeML Open RAIL-M license»; about uploads the terms say verbatim «All materials uploaded remain yours» source, checked 2026-08-07

HeyGen. DIFFERS BY PLAN, and the difference is sharp. Paid: «you own all rights in your User Input or User Output», «HeyGen does not claim to own any of Your Content», commercial use expressly allowed. Free: only a «limited, non-exclusive, non-transferable, revocable license» for personal non-commercial use, and verbatim «User Output generated… source, checked 2026-08-07

HyperWrite. the terms do not allocate rights in the generated text directly; the licence to use the platform is granted «for your personal, non-commercial purposes» — verbatim, so commercial use is not permitted by this document. There is no difference between plans in the text of the terms source, checked 2026-08-07

Koala AI. the result is transferred to the user, verbatim: «Furthermore, we hereby assign to you all our right, title, and interest in and to the Content», where Content means input and output together; the same place says «You retain all rights to your Input». The service reserves the right to use Content to run the service, to comply with the… source, checked 2026-08-07

Krea. the terms do not call the user the owner of what is created directly. It says: "Krea may allow you to commercialize certain Content generated from the Services, subject to your adherence to our copyright dispute policy and any other restrictions or requirements (including pricing changes) imposed by Krea". At the same time what is… source, checked 2026-08-04

LALAL.AI. 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… source, checked 2026-08-07

Let's Enhance. The terms do not define rights in the output directly; about uploads they say verbatim «WE CLAIM NO OWNERSHIP RIGHTS OVER YOUR CONTENT». There is no difference between plans as to commercial use in the terms, but there is a practical one: on the free plan the result is delivered with a watermark (pricing page) source, checked 2026-08-07

Mubert. THE RIGHTS STAY WITH THE SERVICE, AND THE SCOPE OF THE LICENCE DIFFERS SHARPLY BY PLAN. The help centre verbatim: «All the copyright is owned by Mubert Inc. We license tracks for different use cases». The licence page: «Mubert owns all the rights to the tracks generated. Our Sync License allows you to use music in videos, podcasts, apps… source, checked 2026-08-07

Murf AI. DIFFERS BY PLAN. Commercial rights come with any PAID plan: «Yes, all paid plans offer commercial rights over the voiceovers generated in the Studio» (verbatim), including monetisation on YouTube and other streaming platforms. On the free trial downloading is not allowed, so there is no commercial use. The rights are perpetual… source, checked 2026-08-07

Pika. The pricing page marks «Commercial use» on all four plans, including the free Basic, so by the plan grid commercial use is not divided up. Who exactly owns the generated video the pricing page does not say, and the text of the terms of service could not be read source, checked 2026-08-07

Pixelcut. Differs by plan: the line «Commercial license» (verbatim) is part of the Pro and Business plans and absent from the free one; the terms of use do not address rights in the output at all — they contain only a non-exclusive licence to the service to process what is uploaded source, checked 2026-08-07

Playground. Sources diverge: The sources disagree. The pricing page divides the plans: the free one has «No royalty-free license», the paid ones a «World-wide, royalty free license» (verbatim). The terms, on the other hand, draw no distinction between plans and speak of all of them: verbatim «you own all Assets you create with the Services and (iii) we hereby assign… source, checked 2026-08-07

Postwise. rights in the content stay with the user, verbatim: «You retain any and all of your rights to any Content you submit, post or display on or through the Service»; publishing gives the service the right to use, display and distribute that content within the user's account. There is no separate clause about rights in the AI output in the… source, checked 2026-08-07

ProWritingAid. rights stay with the user, verbatim from the terms: «You retain ownership of all content you post, upload to, or otherwise share on the site», and from the privacy policy: «We do not retain any rights to any of the documents or other content uploaded into our system». It does not differ by plan — there are no caveats for Free or Premium… source, checked 2026-08-07

Recraft. the terms distinguish between the plans. The free one: "Assets generated by Free Tier users are Free Tier Assets", a licence "for your personal use" with the restrictions "(i) no commercial use of Free Tier Assets is permitted; (ii) Free Tier Assets are owned by Recraft and may not be sold, licensed or transferred; and (iii) Free Tier… source, checked 2026-08-04

Resemble AI. The terms do NOT contain a direct transfer of rights in the generated audio to the user: there is no wording about ownership of Output in the text. Resemble reserves rights in Aggregated Data — anonymised derivatives and metadata. The user is forbidden to use the output «to train, improve, or otherwise further develop your own or any… source, checked 2026-08-07

SOUNDRAW. the terms leave the rights with the tool outright: "All Intellectual Property Rights pertaining to the Service and the programs that comprise the Service belong to the Company. The neighboring rights of producers of phonograms to the sound sources used in the Generated Music belong to the Company. The Company shall grant the Registered… source, checked 2026-08-04

Suno. the terms split the rights by plan. The free one: "you covenant and agree that you will only use Outputs generated from Submissions made by you through the Service solely for your lawful, internal, personal and non-commercial purposes". The paid one: "Suno hereby assigns to you all of its right, title and interest in and to any Output… source, checked 2026-08-04

Taplio. rights in the content stay with the user, verbatim: «You retain any and all of your rights to any Content you submit, post or display on or through Service»; publishing content through the service, however, gives the company the right to «use, modify, publicly perform, publicly display, reproduce, and distribute such Content on and… source, checked 2026-08-07

Topaz Labs. Differs by plan, and it is written straight into the price table: Topaz Studio has «limited commercial use», Topaz Studio Pro and Topaz Image Web Pro have «Full commercial use» (verbatim). Who owns the processed image the terms on docs.topazlabs.com do not define source, checked 2026-08-07

Upscale.media. The terms do not define rights in the output: the user is granted, verbatim, «a limited, revocable, non-exclusive and non-transferable license» to use the service's content; there is no difference as to commercial use between the free and paid plans on the pages read source, checked 2026-08-07

Wordtune. rights in the text entered stay with the user, verbatim: «You retain all right, title, and interest in and to your User Content»; the licence to use the service, however, is granted «solely for your own personal purposes», and text derived from summarisation «shall be used by the User for personal use only». DIFFERS BY PLAN: the Terms of… source, checked 2026-08-07

The plan is not named — 17

The terms speak of rights the same way for everyone, or no difference was found in what we read.

Anyword. the terms say: "the Customer shall have all right and title in the Generated Content and Company shall not claim ownership with respect thereto"; at the same time the tool reserves the right to use the material "on an aggregate and anonymized basis, to improve the Anyword Platform" source, checked 2026-08-04

Colossyan. commercial use is permitted subject to payment and compliance with the rules: «Subject to the full payment of fees and compliance with the requirements of the Acceptable Use Policy, upon creation of a piece of Customer Created Content… Colossyan grants Customer a perpetual, limited, revocable… source, checked 2026-08-07

Gling. the terms say nothing about rights in the edited result or about commercial use of it: they are written as terms for the SITE and grant «a non-transferable, non-exclusive, revocable, limited license to use and access the Site solely for your own personal, noncommercial use» (verbatim), while the… source, checked 2026-08-04

Happy Scribe. rights stay with the user and are expressly extended to the output of the service. Verbatim: «you retain all rights, title, and interest in and to Your Content. This includes any content generated by the Services — such as transcripts, captions, translations, meeting recordings, or the notes and… source, checked 2026-08-07

LTX Studio. the terms (Lightricks Ltd.): "you own the Output (excluding any Stock Content or other third party materials therein)", but immediately with the caveat "you further acknowledge that pursuant to applicable law, it may be the case that you have no ownership rights, including copyrights, in the… source, checked 2026-08-04

Luma (Dream Machine / Luma Agents). the terms: "Customer owns and retains all right, title, and interest in and to the Output and Luma hereby assigns to Customer all of Luma’s right, title, and interest in and to the Output", but with the caveat "Customer acknowledges and agrees that it can only use the Outputs for commercial… source, checked 2026-08-04

Maestra. the terms do not speak directly about the rights to what is created. What is said: "As between the parties, all Content, except for Customer Data (as defined below), is owned by Maestra or its licensors", and as to what is uploaded the user warrants that "you own all rights, title, and interest in… source, checked 2026-08-04

Notta. rights in what is uploaded stay with the user. Verbatim: «Customer retains all ownership rights in Customer Data», and the service is granted a licence «to access, process, copy, export, and display Customer Data, only as reasonably necessary (a) to provide, maintain and update the Services» and so… source, checked 2026-08-07

Rask AI. the terms: "You are permitted to use the Content you create through our Service outside of the Service, including in commercial contexts, provided that such use complies with all applicable laws, regulations, and these Terms", "You retain full ownership rights to any Content created through our… source, checked 2026-08-04

Rytr. the terms say: "You retain any and all of Your rights to any Content You create, post, or display on or through the Service", while the user grants the tool a licence "to use, modify, publicly perform, publicly display, reproduce, and distribute such Content on and through the Service"; there is no… source, checked 2026-08-04

Sapling. DIFFERS BY CONTRACT, and the service says so outright. The restriction verbatim: «unless you have signed a service agreement with Sapling stating otherwise, you shall not license, sell, rent, lease, transfer, assign, distribute, host, or otherwise commercially exploit the Service, whether in whole… source, checked 2026-08-07

Speechify. the terms are silent about the rights to what is created; moreover, the licence to the tool itself is given thus: "Speechify grants you… a limited, non-exclusive, nontransferable, revocable license to stream, download and make personal non-commercial use of the Services for which you have… source, checked 2026-08-04

Sudowrite. the terms speak only about what the user has uploaded: "All rights in and to Your Content remain with you"; the rights to what has been created by AI and commercial use or publication of what is created are not addressed in the terms source, checked 2026-08-04

Synthesia. ownership is confirmed: «Customer will own all Customer Data, but excluding the components of the Services and Synthesia Content», and Output (video, images, audio) is classed as Customer Data. About commercial use the terms say nothing at all source, checked 2026-08-07

Udio. THE HARSHEST RESTRICTION IN THIS SET, AND IT IS THE SAME ON ALL PLANS. Terms of Service (rev. 12.11.2025), clause 1.2 verbatim: «You may not download copies of any Output or any other output generated by the Services», plus an undertaking not to reproduce, transmit, display, publish or distribute… source, checked 2026-08-07

WellSaid Labs. on the pricing page the commercial rights are tied to the plan: Trial has "3 download minutes per month (no commercial rights)", Starter, Pro and Business have "Full commercial rights", with the common subheading "All with unlimited generation and commercial rights built-in". In the terms… source, checked 2026-08-04

Wisecut. the terms: "Except as provided in this Section 8, we obtain no rights under this Agreement from you (or your licensors) to Your Content", that is, the rights to what has been uploaded and edited stay with the user; the terms contain no separate permission for or ban on commercial use; the user… source, checked 2026-08-04

Whether models are trained on what you upload

They train on yours — 11

The service says outright that it uses what you enter and what is created to train its own models.

Auphonic. YES, training on user recordings is stated. Terms of Service: «your Content may be viewed and/or listened to by an Auphonic employee or any person contracted by Auphonic to work on our audio processing algorithms» (verbatim); individual audio fragments may be stored indefinitely for the sake of improving the… source, checked 2026-08-07

HeyGen. they train on both the input and the output, and say so directly: the user grants a licence «to access, use, host, cache, store, reproduce, transmit, display, publish, distribute, and modify Your Content to operate, improve, promote and provide the Services and to develop new services and products, including to train… source, checked 2026-08-07

HyperWrite. yes, verbatim: the service may use the data «to improve the Software, including to train its machine learning algorithms applicable to its general user base and systems, and related products»; the content of letters and the nature of interaction with the service are named separately source, checked 2026-08-07

Luma (Dream Machine / Luma Agents). the terms directly permit training on what is entered and created: with paid use the licence is given, among other things, in order to "create, test, improve, train, or otherwise develop the artificial intelligence or machine learning models, systems, architecture, weights, or related technology used by Luma", and… source, checked 2026-08-04

Murf AI. There is no direct statement in the terms about training product models on user texts and audio. There is only a caveat about quality control: Murf staff «may use your Cloning Voice to create ... utterances, solely for internal quality assurance purposes» (verbatim), and the service may listen to audio samples «to… source, checked 2026-08-07

quso.ai (formerly vidyo.ai). Yes, uploaded video goes to AI models. The privacy policy: «To enhance these features, we share anonymised data with these AI models. This include the video you uploaded to our app» (verbatim). Which model providers exactly is not named source, checked 2026-08-07

Sapling. yes, verbatim from the terms: the user grants Sapling a licence to use their content «solely for the purposes of providing the Service and to improve the Service and develop new products and services»; the privacy policy confirms: «We collect text entered or otherwise transmitted by you» and «Your content is stored by… source, checked 2026-08-07

Suno. training on what is entered is built into the terms: the user warrants that no additional permissions are required "arising out of or related to our use of your Submissions, including to create your Output or your Voice Model and/or to train, develop, fine-tune or otherwise improve the Service and any related… source, checked 2026-08-04

Udio. YES, stated directly. Terms of Service clause 1.1 verbatim: «You acknowledge and agree that we may use your Input Content and any Output to train or otherwise improve our Services, including our artificial intelligence and machine learnings model(s) related to the Services» (the source's spelling preserved). Input… source, checked 2026-08-07

Wisecut. the terms directly require consent to training: "You consent to our use of Your Content and any associated data to train and improve our artificial intelligence models. In such circumstances, we will not publicly post or share any of Your Content"; the terms describe no opt-out from such use source, checked 2026-08-04

Wordtune. yes, the licence over User Content includes the right to use it to «administer and make improvements to the Software and Services (including, to improve the algorithms underlying the Software and the Services)» — verbatim; the licence is described as perpetual and irrevocable source, checked 2026-08-07

Not trained on — 8

The service denies this outright. The same section covers cases where training is possible only with your consent: without consent they do not train, and for the reader that is the answer «no».

Amberscript. NO, expressly denied in the privacy policy. Verbatim: «The audio and transcripts that are being generated, will never be used for training purposes of the language models that are used». It is separately stipulated that when human proofreading is ordered, the file and the draft transcript are visible to in-house… source, checked 2026-08-07

Cleanvoice AI. NO, training is expressly forbidden by the service itself. Terms & Conditions clause 11.7: stored content is not used «for training artificial intelligence models or for any purpose unrelated to the provision of the Services» (verbatim) source, checked 2026-08-07

Colossyan. the most direct denial of the three: «Colossyan will not use… Customer Materials to train any AI model or algorithm, with the sole exception that… Colossyan may use Recordings to train the AI model that creates and operates Customer’s Custom Created Avatar. Such AI model is exclusively used by Colossyan to provide the… source, checked 2026-08-07

Happy Scribe. NO by default: training is possible only with the user's explicit consent. Verbatim: «If you have not opted in, Your Content uploaded on or after the Effective Date is not used to train or maintain any machine-learning models». With consent switched on, uploads are used to «develop, train and improve machine-learning… source, checked 2026-08-07

LALAL.AI. 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 source, checked 2026-08-07

LTX Studio. the terms give Lightricks a "worldwide, non-exclusive, fully-paid, royalty-free, irrevocable, sublicensable and transferable license to use the Input you provide, the Output and your Account information including your or any third-party’s name, voice, image, likeness", including "for Lightricks’ internal purposes to… source, checked 2026-08-04

ProWritingAid. no, verbatim: «When using ProWritingAid, your texts will not be used to improve the quality of our services»; the same place says that after a grammar check «all submitted texts and their improvements will be deleted», and that the algorithms are trained on texts proofread by in-house professional copy editors source, checked 2026-08-07

Sonix. NO, expressly denied in the privacy policy. Verbatim: «Sonix does not use your content or Customer Data to train our machine learning or generative AI models». It is separately stipulated that when a third-party AI application is connected through OAuth the service passes on «only the data necessary to fulfill the… source, checked 2026-08-07

They train, but with reservations — 2

One document contains both: training happens but is bounded. Reducing this to «yes» or «no» would be lying in both directions.

Recraft. the terms: "You acknowledge and agree that Recraft may use your Inputs you provide and the Assets produced by the Services to train, fine-tune, and otherwise improve Recraft’s artificial intelligence models and systems. This training use is enabled by default, but you may opt out at any time by adjusting your… source, checked 2026-08-04

Synthesia. improvement using usage data does happen, but with explicit caveats: improvements «will not result in… (d) the improvement or fine tuning of any artificial intelligence components unless integrated with the Services being used by Customer; or (e) the creation or development of any artificial intelligence models that… source, checked 2026-08-07

The answer does not reduce to a single word — 6

The wording does not read unambiguously as either «yes» or «no». It is not interpreted — it is printed verbatim.

Anyword. the terms give the tool "a non-exclusive, perpetual, worldwide right and license to use, copy and reproduce the Content solely to (i) provide the Customer the Services; and (ii) provided that the Content is first aggregated and anonymized" — that is, what is uploaded is used to improve the platform after it has been… source, checked 2026-08-04

Bria. About its own models the service writes verbatim «Every asset in our training pipeline comes from commercial agreements with 30+ data partners» and «No scraping. No synthetic data. No gray areas»; about user uploads it says only that during fine-tuning «our architecture keeps your data isolated and auditable»… source, checked 2026-08-07

Let's Enhance. The privacy policy says the service applies automated systems to uploads verbatim «for security purposes, fraud prevention, compliance with legal and regulatory requirements, investigations of potential misconduct, product development and improvement, research, and customer or technical support»; there is no wording… source, checked 2026-08-07

Playground. The terms grant the service a perpetual licence over prompts, uploaded images and results: verbatim «a perpetual, worldwide, non-exclusive, sublicensable, no-charge, royalty-free, irrevocable copyright license to use, copy, reproduce, process, adapt, modify, publish, transmit, prepare Derivative Works of, publicly… source, checked 2026-08-07

Resemble AI. The terms permit internal training on derived data: Content is used «for the purposes of having Resemble AI synthesizing and generating an AI voice model» (verbatim), and Aggregated Data «may be used for ... internal learning and training» (verbatim), as well as for debugging, fixing errors and product development.… source, checked 2026-08-07

SOUNDRAW. the terms say nothing about training on what the user submits and describe no opt-out. The service answers a different question — about the provenance of ITS OWN training data: «SOUNDRAW is an ethical AI music generator, trained using input from our in-house music team. Since we never use stolen material for training… source, checked 2026-08-04

Whether you can opt out

Amberscript. no separate opt-out mechanism is described and none is needed: training on uploaded recordings is denied by default for all users source, checked 2026-08-07

Happy Scribe. yes: consent is given voluntarily and can be withdrawn at any time. Verbatim: «withdraw your opt-in at any time from your account settings or by emailing dataprotection@happyscribe.com» source, checked 2026-08-07

HyperWrite. yes, verbatim: «you shall have the right to opt-out of the general systems»; after opting out the data is used only for models trained for that particular user; the request is sent by email to info@othersideai.com, and there is a warning in the same place that after opting out the user does not… source, checked 2026-08-07

Krea. DIFFERS BY PLAN, and that is the main finding. The general Terms of Use (rev. 20 May 2024) say NOTHING AT ALL about training models on user data, although they grant Krea a «worldwide, non-exclusive, perpetual, royalty-free, fully paid, sublicensable and transferable license to use, edit, modify… source, checked 2026-08-07

LTX Studio. There is an opt-out, but a partial one and only for personal data. Terms of Service §6.2 grant Lightricks a licence over Input, Output and account data, including «name, voice, image, likeness», for purposes including «(iii) for Lightricks' internal purposes to improve our products and services and… source, checked 2026-08-07

quso.ai (formerly vidyo.ai). Consent is given by a checkbox at upload: «By checking this box, you allow us to share anonymised data to enhance app features and functionality» (verbatim). That is, you can decline by not ticking the box. No separate procedure for withdrawing consent already given is described in the policy source, checked 2026-08-07

Recraft. Everyone has an opt-out and it is switched on in the profile settings. Terms of Service, the «Licenses to Recraft» section, verbatim: «You acknowledge and agree that Recraft may use your Inputs you provide and the Assets produced by the Services to train, fine-tune, and otherwise improve Recraft's… source, checked 2026-08-07

Sonix. no separate opt-out is required, since training on customer data is denied by default. A withdrawal mechanism is described only for third-party connections: verbatim you may «revoke a connector's access at any time through your Sonix account settings or by contacting us» source, checked 2026-08-07

Sudowrite. There is nothing to opt out of: the service states that it does not train models on users' texts at all. In the FAQ, to the question «Does Sudowrite use my writing to train models?» the answer is verbatim: «We are constantly listening to writers' feedback to improve Sudowrite. However, we do not… source, checked 2026-08-07

WellSaid Labs. there is nothing to opt out of: the services agreement expressly forbids training on customer data — verbatim «For clarity WellSaid will not use Customer Data to train its or any third party's AI models» (the Customer Data section); Customer Data is defined as Input and Output; about a user's own… source, checked 2026-08-07

Whether the service marks the work of AI

Platforms require AI-made content to be labelled; whoever publishes answers for the label. What the tools themselves say about it is below.

2short.ai. The service promises no watermark on the output on any plan: the features block says «Say goodbye to watermarks and hello to 1080p quality» (verbatim), and the plans, by the service's own statement, differ only in hours of analysis. There is no separate rule for the free plan. Metadata, C2PA and… source, checked 2026-08-07

Auphonic. There is no AI labelling: the Terms of Service and the pricing page contain no mention of a watermark, metadata or C2PA marking machine processing. A difference between plans does exist, but of another kind — an audio insert: in the comparison table the row «Includes Auphonic jingle» is Yes for… source, checked 2026-08-07

Bria. The service embeds attribution in the output: the terms verbatim forbid you to «remove, obscure, or alter any attribution, watermarks, or metadata embedded in Outputs by Company's attribution technology»; no difference between plans is stipulated in this clause, and C2PA is not mentioned by name source, checked 2026-08-07

Checksub. on the free option video is exported with the service's watermark («With Watermark» in the comparison table), and on all paid plans there is no mark («No Watermark»). This is a brand mark, not a notice that the subtitles or dubbing were made by AI; there is no requirement to label the result as AI… source, checked 2026-08-07

Colossyan. there is a watermark on the free plan, and its removal is sold from the Professional plan (pricing page). The terms say nothing about watermarks, metadata or C2PA; they do, however, expressly forbid passing off what AI made as made by a human source, checked 2026-08-07

Dubverse. there are no requirements to label the result as AI content, and no metadata or C2PA, in the terms that were read or on the pricing page. The opposite is there: on both paid plans «No Water Mark» is stated outright, that is, there will be no watermark on the exported video source, checked 2026-08-07

Fliki. A visible watermark differs by plan: on Free «Includes Fliki watermark» (verbatim), while on Standard and Premium the comparison table shows «No watermark». About metadata, C2PA or any other hidden marker of AI origin neither the pricing page nor the terms says anything source, checked 2026-08-07

Happy Scribe. no labelling of the output as AI content is claimed in the terms of use or on the pricing page. There is SOMETHING ELSE: on the free plan video export carries the service's watermark — «Export files in DOCX, TXT, SRT and MP4 (watermarked)» (verbatim), while paid plans allow you to «Remove watermark… source, checked 2026-08-07

HeyGen. there is a watermark on the free plan, and its removal is sold with the paid ones (pricing page). The terms themselves, however, say nothing about watermarks, metadata or C2PA, and the moderation rules require labelling only for 18+ content: «clearly labeled at the outset of the video as mature content» source, checked 2026-08-07

LALAL.AI. 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… source, checked 2026-08-07

Let's Enhance. AI-origin labelling (metadata, C2PA) is not mentioned in the terms or on the pricing page; the watermark on the free plan is presented as a plan limitation, not as an AI notice source, checked 2026-08-07

Luma (Dream Machine / Luma Agents). the terms: "Luma reserves the right, in its sole discretion and without prior notice, to embed watermarks, content credentials, provenance metadata, or other identifying information in or associated with Output, including as may be required by applicable Laws", removing or circumventing them is… source, checked 2026-08-04

Mubert. There is a watermark, but only until download. Verbatim: «Mubert watermark is added to prevent fraudulent activity. The watermark will be deleted once you download the track to your device» — so no mark remains in the downloaded file. There is no requirement to label the result as AI-made. The… source, checked 2026-08-07

Pika. The service places no visible watermark on any plan: the line «Download videos with no watermark» (verbatim) appears on all four plans, including the free one. This does not differ by plan. About metadata, C2PA and other hidden markers of AI origin the pricing page says nothing source, checked 2026-08-07

Pixelcut. A notice of AI creation (a watermark, metadata, C2PA) is nowhere mentioned on the pages read; the pricing page, meanwhile, expressly promises free export without a watermark source, checked 2026-08-07

quso.ai (formerly vidyo.ai). The service speaks of a watermark only by negation, and only on the paid plans: on Lite «Share crisp, watermark-free videos», on the mobile Plus «No watermarks on exported videos» (verbatim). In the description of the free plan a watermark is mentioned neither as present nor as absent. Metadata… source, checked 2026-08-07

Recraft. the terms contain a section "Transparency and Marking of AI-Generated Content": the tool "may (i) embed machine-readable metadata, watermarks, or other provenance information into Assets generated by the Services; (ii) make available or link to a tool that enables users and third parties to detect… source, checked 2026-08-04

Resemble AI. The PerTh watermark is Resemble's own technology; for the open models labelling is claimed to be mandatory and automatic: «Every clip is watermarked with PerTh at generation, before it leaves the model» (verbatim, Chatterbox Multilingual); for DramaBox — «Every output watermarked with PerTh at… source, checked 2026-08-07

Spikes Studio. The visible watermark differs by plan outright, and the service says so without evasion. The FAQ: «Do the videos have watermarks? Yes, our free plan includes a watermark. To remove it, you'll need to subscribe to any of our paid plans» (verbatim). The free plan card repeats «Has watermark», while… source, checked 2026-08-07

Synthesia. the acceptable use policy forbids removing marks: it is forbidden to be «Removing, deactivating or disabling any «watermarks» or other mechanisms of the Services that are designed to help validate provenance or differentiate between human-generated and AI-generated content» (verbatim). What exactly… source, checked 2026-08-07

How to use this

Check the plan, not only the service. Rights to what you made depend more often on which plan you worked on than on which tool. 2. A document's silence is not permission. If nothing is said about rights, training or opting out, there is no answer; reading it in your favour is exactly as dangerous as reading it against yourself. 3. The requirement to mark AI work comes from the platform, not from the tool. See what platforms require: the one who publishes is the one who will answer for the mark.

On method. Every line is a fact with a reference to the tool's document. The direction of the answer («trains» / «does not train» / «with caveats») is sorted by a rule with six control examples; whatever the rule could not sort is put in a separate section and printed verbatim rather than filed under «yes» or «no».