What do AI tool terms say about training on uploads?
Of the 24, 13 say something. invideo, Vyond and SceneMixer keep uploads out of training; Runway, HeyGen, Higgsfield, Vidu, Story.com, Krikey AI and Descript may use them; Topview uses them unless you opt out; Luma's paid-use licence covers training; Pictory uses content in aggregate to improve its service. As of 2026-09-25.
| Source | Covers | What it says |
|---|---|---|
| Descript | Training | Voice training audio may be added, de-identified, to research datasets that train its voice technology |
| HeyGen | Training | Content may be used to train or improve its models |
| Higgsfield | Training | Content, inputs and outputs may be used to train its models; enterprise content is excluded |
| invideo | Training | User data not used to train or fine-tune its models |
| Krikey AI | Training | Content created from user inputs may be used to train an AI model |
| Luma | Training | Input from paid use may be used to train its models |
| Pictory | Training | Content used in a non-identifiable, aggregated way to evaluate and improve its services |
| Runway | Training | Inputs and outputs may be used to train and improve its models |
| SceneMixer | Training | Uploads and outputs not used to train models without separate consent |
| Story.com | Training | Customer data used to develop and train AI models, though not raw personal data |
| Topview | Training | Content used by default to train its models unless you opt out |
| Vidu | Training | Content may be used to improve its products, including AI training |
| Vyond | Training | Customer content and outputs not used to train language models or machine-learning algorithms |
Inclusion rule. A source is listed when its own text states what the question asks. Order. Sources in alphabetical order.
For a manuscript nobody else has read, this is the clause to check before anything else.
Silence is not a promise either way. Terms of service often point to separate privacy or data policies, which may say more.
Enterprise plans can carry terms of their own. Higgsfield's training clause excludes enterprise customers under separate agreements, and Descript's research use covers voice training audio.
1In their own words
Descript
as part of our research datasets to analyze, maintain, train, and improve our voice technology
Descript, terms of service
HeyGen
including to train or otherwise improve or modify our artificial intelligence and machine learning models
HeyGen, terms
Higgsfield
Your Content, Inputs, and Outputs may be used by Company to train, develop, enhance, evolve, and improve its (and its affiliates’) AI models
Higgsfield, terms of use agreement
invideo
Invideo will not use your User Data to train or fine-tune its artificial intelligence or machine-learning models.
invideo, terms and conditions
Krikey AI
we may use any Krikey Content, including Service Animations and Service Tracks created from User Inputs, for training an artificial intelligence model
Krikey AI, terms of service
Luma
create, test, improve, train, or otherwise develop the artificial intelligence or machine learning models
Luma, terms of service
Pictory
in a non-identifiable aggregated manner for the limited purposes of evaluating, improving, or enhancing the suite of services
Pictory, terms of service
Runway
You acknowledge that Inputs and Outputs may be used by the Company to train and improve its AI models
Runway, terms of use
SceneMixer
Unless you separately consent or the law requires it, we do not use your uploads or outputs to train models
SceneMixer, terms
Story.com
Developing and training artificial intelligence and machine learning models
Story.com, terms of service
Topview
we may by default use Your Content and Output to train, develop, and improve the artificial intelligence models and technology that power the Services
Topview, terms
Vidu
use User Content to improve and develop Vidu’s existing and future products, including artificial intelligence training
Vidu, terms of service
Vyond
Company will not itself use (nor will permit its third-party AI model providers and Subprocessors the right to use) Customer Content or Customer-Owned Output to create new products or services or to train any large language AI models
Vyond, terms of service
- AI speakersVoice training audio may be added, de-identified, to research datasets that train its voice technology
- Your contentContent may be used to train or improve its models
- Inputs and outputsContent, inputs and outputs may be used to train its models; enterprise content is excluded
- User dataUser data not used to train or fine-tune its models
- Krikey contentContent created from user inputs may be used to train an AI model
- Use of inputInput from paid use may be used to train its models
- Customer contentContent used in a non-identifiable, aggregated way to evaluate and improve its services
- Your contentInputs and outputs may be used to train and improve its models
- Your content and IPUploads and outputs not used to train models without separate consent
- Rights in customer dataCustomer data used to develop and train AI models, though not raw personal data
- Improving modelsContent used by default to train its models unless you opt out
- User contentContent may be used to improve its products, including AI training
- AICustomer content and outputs not used to train language models or machine-learning algorithms
Descript · HeyGen · Higgsfield · invideo · Krikey AI · Luma · Pictory · Runway · SceneMixer · Story.com · Topview · Vidu · Vyond · Do public-domain ebook files carry their own terms? · Do AI video tools make you promise you own the text?