Which video editors train AI on your videos? 13 terms, read
We read the terms and privacy policies of 13 video editors on 2 November 2026. Six say in their own documents that they train, or may train, AI on what users give them: CapCut, the Captions app, Riverside, OpusClip (individual users), VEED (its Free plan) and Canva (subject to a privacy setting). Descript gives users a setting. Four reserve a broad right to improve their services, and two do not mention training on uploads. We make Burncaps, a rival, so every row names its source and date.
- Read 2 Nov 2026
- Written by a rival, said up front
- Burncaps: video never uploaded
Live preview, rendered on this page by the same engine that burns your video.
13 editors, their own words on your videos
Each row is what the company's own terms, privacy policy or help pages said on 2 November 2026. The AI column uses five plain classes and never rounds up: says it trains or may train; opt-out offered; reserves a broad right to improve its services; says it does not; not mentioned. A right to “improve” a service is not counted as training.
| Editor | Your video goes to its servers | What it says about AI training | The licence you give it | How long it keeps your files | Source, dated |
|---|---|---|---|---|---|
| OpusClip | Yes, to “secure Google Cloud servers located in the United States” | Says it may, for individual users: it “may use your data to improve our AI models”, with personal identifiers stripped. Says it does not for enterprise customers. EU and EEA residents can opt out by email | Royalty-free, perpetual, sublicensable, irrevocable and worldwide, to operate its services; public submissions also for its marketing | Projects expire after 3 days on Free, 7 on the trial and 30 on Starter and Pro; its terms say complete deletion “may not be possible” | Terms and privacy, 17 November 2025; help centre, undated |
| Submagic | Yes: hosting, storage and processing on Amazon Web Services | Not mentioned: neither page speaks of training on what you upload. You can opt out of promotional use of your content by email | Worldwide, non-exclusive, to use, modify, reproduce and distribute it on and through its platform; its terms say you keep all rights | Destroyed if not asked for within 60 days after the agreement ends; its privacy policy says customer data is kept no longer than 2 years after the relationship ends | Terms and privacy, no date shown |
| CapCut | Yes, by “pre-uploading”, even before you save | Says it trains: to “train and improve our technology, such as our machine learning models”, on its legitimate interests in the UK and EEA. No opt-out found | Non-exclusive, royalty-free, transferable, sub-licensable, perpetual and worldwide; plus your username, image and likeness, including in sponsored content | “As long as necessary to provide the Services” | Terms and privacy, 15 April 2026; clause by clause |
| VEED | Yes | Free plan: says it may, using content “to train, test and develop any relevant artificial intelligence models”. Paid plans and enterprise: says it does not. To object on Free, you delete the account or upgrade | Worldwide, revocable, non-exclusive, royalty-free and transferable, in connection with its services; the Free plan's training licence is irrevocable while you stay on Free | “As long as reasonably necessary” | Terms, 28 August 2026; privacy, 6 January 2025 |
| Kapwing | Yes: it may “transfer, store, and process” your content in and out of the United States | Not mentioned for uploaded videos. Voice Cloning recordings are kept “for research and development purposes” | Non-exclusive, to publish it on Kapwing and to store, display, reformat and distribute it; it may use content for promotion and says it will never sell it without your explicit permission | A deleted project leaves its servers within 7 days | Terms, September 2024; privacy, 3 August 2022 |
| Descript | Yes | Opt-out offered: a “Share data with Descript” setting. Its terms allow training unless you opted out; its help centre says “Current AI models in production use no Descript user data” and research uses only users who opted in. Off by default for enterprise | Perpetual, irrevocable, nonexclusive, royalty-free, sublicensable and worldwide, including your name and likeness, to provide, maintain and improve its service | As long as necessary; voiceprints and avatar face data until their purpose is met, or up to three years after you last accessed the account | Terms, 25 February 2025; privacy, 14 April 2025; help, undated |
| Captions app (Mirage) | Yes: uploads are transcribed, captioned and clipped by its services | Says it trains: “including to train our artificial intelligence/machine learning models”. It may also share information with third parties for their own purposes, including training their algorithms. No opt-out found | Limited, nonexclusive, royalty-free and sublicensable, to provide and improve its services and develop new ones; it says it will not commercialise your voice on its own without your written permission | As long as necessary; biometric data until its purpose is met, or three years after your last interaction with it | Terms, 4 September 2025; privacy, 10 November 2025 |
| Vizard | Yes | Reserves a broad right: its terms let it use machine learning to “improve algorithms” while processing what users provide. They do not say it trains on your videos. No opt-out found | Nonexclusive, perpetual, irrevocable, royalty-free, worldwide, transferable and sub-licensable, “solely for the operation and maintenance” of its website | Free projects expire after 3 days; on paid plans, unsaved projects over the storage quota after 28 | Terms and privacy, 31 October 2022 |
| Klap | Yes: Google Cloud in the United States; Google's Gemini receives a video that has no speech | Reserves a broad right: it uses data to “provide, improve, and maintain” its services, and contractors review samples of video frames. No training statement and no opt-out found | None: its terms grant it no licence over your content | Until you delete it or your account, plus up to 30 days in backups | Terms, 25 July 2024; privacy, 1 October 2026 |
| Canva | Yes | Says it may, under a privacy setting: it may use content and media uploads “to train our algorithms, models and AI products”, managed in your privacy settings. Canva Education content is excluded. The default for a personal account is not stated on the pages we read | Royalty-free and sublicensable, to provide its service; perpetual for designs you share | For a commercially reasonable time after the account ends, its terms say | Terms, 30 October 2026; privacy, 2 November 2026 |
| Riverside | Yes: a cloud platform for recording, editing, hosting and sharing | Says it trains: the licence includes “training, testing, developing, and improving” its AI models. Business plans can opt out by written agreement | Limited, non-exclusive, royalty-free and worldwide, covering recordings, video, audio and voice | As long as needed; voice and face data for its synthetic features destroyed once the feature is done | Terms, 5 October 2026; privacy, 15 June 2026 |
| Clipchamp (personal accounts) | No for editing and export: media is “processed locally on your machine”. Auto captions send the audio to Microsoft's Azure Cognitive Services; work and school accounts back up to OneDrive or SharePoint by default | Not mentioned on Clipchamp's pages. The Microsoft Services Agreement reserves a broad right: its licence includes improving Microsoft products and services | Worldwide and royalty-free, to the extent necessary to provide and protect the services and to improve Microsoft products | Caption data stays with your project; the video stays on your device unless you back it up | Help, undated; Services Agreement, effective 30 September 2026 |
| Zeemo | Yes | Reserves a broad right: the same machine learning clause as Vizard's, and it collects transcribed and edited text “to develop and improve this function”. No opt-out found | Nonexclusive, perpetual, irrevocable, royalty-free, worldwide, transferable and sub-licensable, solely to operate its app | Deleted on request | Terms, 2 March 2022; privacy, 18 February 2024 |
| Burncaps (ours) | No. The video is captioned and clipped in your browser and never uploaded. The sound goes to our speech service on a phone unless you switch it off, on a computer only if you choose it | We train no AI model of our own. Microsoft says it “does not use customer data to improve its Speech models”; Anthropic's terms say it “may not train models on Customer Content”. Microsoft lists MAI-Transcribe-2, the model it runs for us, as a preview, and its pages do not yet say whether these promises cover previews | None. Our terms: “Your files never reach us, and you keep all rights to them.” | The video never reaches us. Neither we nor, by Microsoft's statement, Microsoft keep the sound, and we do not store the transcript | Privacy and terms, 2 November 2026 |
Read on each company's own pages on 2 November 2026, from the UK; where a company shows different text by region (CapCut, the Captions app, Canva), the UK or EEA version is the one used. If a page has changed since, it is the authority. Product names belong to their owners.
What Burncaps does with your video, sound and words
Your video
Never uploaded. Burncaps opens it, hears it, draws the captions or cuts the clips and writes the new MP4 inside your browser, so we never receive it and cannot keep it or train anything on it. You can watch this in your browser's Network tab: how to check, and every request the page makes.
Your sound
Only to our speech service: on a phone unless you switch it off, on a computer only if you choose it. It goes in pieces of up to 30 seconds, with nothing from the picture, through our server to Microsoft's fast transcription in Ireland. Microsoft's documentation says that for fast transcription it “does not retain or store the data provided by customers”, and that it “does not use customer data to improve its Speech models”. Microsoft lists MAI-Transcribe-2, the model it runs for us, as a preview, and its pages do not yet say whether these promises cover previews.
Your words
Only when you choose Clips, type names to spell right or pick a caption language other than the one spoken, the transcript text goes to Anthropic's Claude API. Anthropic's commercial terms say it “may not train models on Customer Content”, and its privacy centre says API inputs and outputs are deleted within 30 days, with exceptions such as content flagged under its usage policy. We do not store the transcript.
We train no AI model of our own. What our server keeps is counts, stored as salted one-way hashes, and for a paid plan your email, your purchase and up to three signed-in devices (ten with Captions + Clips). One exception, only on Captions + Clips and only if you add your podcast's public feed: our server keeps each new episode's transcript and clips for 30 days. Ticking Keep everything on this device stops every request that carries sound or words.
Microsoft: Data and privacy for Speech to text (last updated 26 August 2026), Speech service encryption of data at rest (28 February 2026) and MAI-Transcribe in Azure Speech (preview) (24 September 2026). Anthropic: Commercial Terms (effective 17 June 2025) and How long do you store my data? (1 July 2026). All read on 2 November 2026. Our own: privacy policy.
Other tools that keep the video on your device
Burncaps is not alone in this. These tools also do the work on your own computer or in your browser, by their makers' own account or by reports we name.
| Tool | What runs on your device | Source, and how sure we are |
|---|---|---|
| Microsoft Clipchamp (personal accounts) | Editing and export in the browser; auto captions send the audio to Microsoft's Azure Cognitive Services, much as Burncaps sends the sound to its speech service on a phone | Microsoft's help page: the vendor's own words |
| VDoCipher Caption Generator | Speech recognition in the browser and burned-in captions; it says “Your video and audio are never uploaded, transmitted, or stored on any server” | Its tool page: the vendor's claim; we did not check its code |
| Whisper Web | Open-source speech recognition in the browser: transcripts, no burned-in captions | Its code on GitHub: open source |
| Shotcut | A free desktop editor with speech to text on the computer since version 24.10 | Its release notes |
| Adobe Premiere Pro and Apple Final Cut Pro | Speech to text reported to run on the computer | An Adobe forum answer and press reports: less certain |
How we read the terms
On 2 November 2026 we downloaded each company's current terms, privacy policy and, where they answer the question, its help pages, and searched each word by word for training, models, machine learning and improving. Canva's terms and trust page, and part of Riverside's pages, only show in a browser and were read through a tool that summarises; the Canva training words quoted above come from its privacy policy, read in full.
Every quote is the company's exact words. A row says a company trains only where its own document says it trains or may train; a general purpose to improve or develop a service is listed as a broad right, never as training. Where documents disagree, as Descript's terms and help centre do, the row says so. We will read every row again by 2 December 2026, because these terms change: CapCut's changed twice in a year, and Canva updated its privacy policy on the day we read it.
We make Burncaps and compete with these tools. If you work for one of them and a row is out of date or wrong, write to hello@burncaps.app and we will correct it.
Questions, answered plainly
Which video editors train AI on my videos?
Of 13 editors whose terms we read on 2 November 2026, six say in their own documents that they train, or may train, AI on what users give them: CapCut, the Captions app, Riverside, OpusClip (individual users), VEED (its Free plan) and Canva (subject to a privacy setting). Descript gives users a setting for it. Vizard, Zeemo, Klap and Clipchamp reserve a broad right to improve their services, and Submagic and Kapwing do not mention training on uploaded videos.
Does OpusClip use my videos to train AI?
OpusClip's help centre says that for individual users it “may use your data to improve our AI models”, with personal identifiers stripped, and that it does not use enterprise customers' data to train AI models. Residents of the EU and EEA can opt out by email. Read on 2 November 2026; the help article shows no date.
Does CapCut use my videos to train AI?
Its privacy policy (last updated 15 April 2026) says it uses what it collects, including your content, to train and improve its machine learning models, on the basis of its legitimate interests in the UK and EEA. We found no way to opt out of the training. Our page on whether CapCut is safe quotes each clause.
Does VEED train AI on my videos?
On its Free plan, VEED's terms (28 August 2026) take a licence to use your content to develop its applications, including to train AI models, and say that licence is irrevocable while you stay on Free. They say it does not apply to paid plans or enterprise customers. To object on Free, VEED says you must delete your account or upgrade.
Can I opt out of AI training?
In some editors. Descript has a Share data with Descript setting, Canva has a control in its privacy settings, OpusClip takes opt-outs by email from EU and EEA residents, VEED's Free users can upgrade or delete their account, and Riverside's business plans can opt out by written agreement. In CapCut, the Captions app, Vizard, Zeemo and Klap we found no opt-out.
Is there a private AI video editor that does not train on my data?
Yes, if the video never reaches the company. Burncaps, which we make, captions and clips the video in your browser, so we never receive it, and we train no AI model of our own. Clipchamp's personal accounts also edit on your computer, VDoCipher says its caption generator never uploads the video, and Whisper Web runs speech recognition in the browser.
Does Burncaps train AI on my videos?
No. The video never leaves your browser, and we train no AI model of our own. On a phone the sound goes to our speech service unless you switch it off, and Microsoft says that service does not use customer data to improve its speech models; Microsoft lists MAI-Transcribe-2, the model it runs for us, as a preview, and its pages do not yet say whether these promises cover previews. When you choose Clips, names to spell or a translation, the transcript text goes to Anthropic's Claude API, whose commercial terms say Anthropic may not train models on customer content.
Burncaps facts, checked 2 November 2026
What it is: a browser tool that burns animated word-by-word captions into a talking video, or finds the best moments of a talk, podcast or interview and gives them back as ranked, captioned 9:16 clips.
Price: free for 2 videos every 30 days, each one recording with its captioned video and up to 3 clips, no mark and no sign-up; then the free 7-day Captions + Clips trial, no card (your email only); then Captions + Clips, $29 a month, gives unlimited clips and captioned videos, or Captions, $9.98 a month or $49 once, gives unlimited captioned videos. Captions paid once is not a subscription.
Privacy: the video never leaves your browser. The sound goes only to our speech service, on a phone unless you switch it off, on a computer only if you choose it. Because the video never reaches us, we cannot keep it or train AI on it, and we train no AI on anything you make.
Files: MP4 or MOV in, MP4 out. Captions on the whole video: up to ten minutes. Clips: up to three hours, on a phone or a computer. SRT, VTT, ASS and text files are free without limit for recordings up to ten minutes; a longer recording, up to three hours, gives its script and those files (the first ten minutes free, all of it in the trial or on either paid plan).
Clips: about one clip for every two minutes of talk, up to 30, ranked, the ones we would post first circled, each showing the words it opens and ends on and the reason it was chosen; each cut to 9:16 with ums and dead air removed, an opening card and captions, framed on whoever is talking. Every clip of the whole recording is found and shown, free and with no sign-up; each free video exports up to 3 of them, and the 7-day trial and Captions + Clips export every clip, unlimited.
Languages and styles: captions in the language spoken, twelve Indian languages heard on your device, translation into 59 languages, 60 caption styles.
After it listens: correct any word (one click changes it everywhere), add or cut words, one to six words at a time, position, size, text and highlight colours, and names this browser remembers, all free and without listening again.
Made by: Jordan Labs Ltd, UK company no. 17358646.
Against other tools: clip tools compared, from the vendors' own pages.