Where to sell video footage for AI training: Troveo, Protege, Defined.ai, Shutterstock and Wirestock compared

Video licensing platforms compared

Terms are as each company or a named report publishes them, checked November 2026.

Troveo: how it works and what it pays

Troveo licenses video libraries from production companies, channels, broadcasters and independent shooters to teams building video-generation and world models. It licenses privately and directly to labs under NDA, pays a royalty on every licence and keeps licences non-exclusive by default. There is no public rate card: price depends on hours, resolution, variety, demand for the categories and how clean the rights are. It prefers 1080p or 4K masters and excludes third-party music unless the music licence covers AI training. Processing takes one to two weeks.

It launched in November 2024 with $4.5 million of seed funding, The Hollywood Reporter reported. By March 2025 it had paid about $5 million for one million hours of content, and by October 2026 it said it had paid rights holders more than $20 million. One creator earned $33,000 in a year from about 1,000 hours of old YouTube videos through Troveo, Semafor reported.

Source: Troveo: how to sell your videos for AI training (Oct 2026)

Protege: media, audio and health data

Protege licenses media content, audio, de-identified health records and medical imaging to AI developers, and pays data partners a revenue share each time their data is used. It says it works with most of the largest US technology companies and that its partner network grew to hundreds of organisations in 2025. It raised a $30 million Series A extension led by a16z in January 2026, bringing its total to $65 million.

Its audiovisual unit, Protege Media, licenses scripted and unscripted content, final cuts and raw footage such as unused B-roll, Luminate reported. Pre-training datasets have been paid as flat upfront fees, it has done a first usage-based deal, it secures non-perpetual licences, and its contracts bar training on licensed film music or reproducing actors' faces and voices. In July 2026 Dailymotion put a catalogue of about 40 million videos on Protege. Pricing and take rates are not published.

Source: Protege: $30M Series A extension (Jan 2026)

Defined.ai: a data marketplace

Defined.ai lets data partners license text, audio, image, video and multimodal datasets through its marketplace and earn licensing fees from sales. Commercial terms vary by dataset and use, and the partner approves them; there is no fixed minimum dataset size. Onboarding reviews quality, volume, organisation and metadata, plus licensing rights, legal, compliance and security. It requires consent, documentation and the right to commercialise the data for AI training.

It reports dataset sales up 319% year on year, average partner revenue above $600,000 and more than five partners above $1 million, and said in January 2026 that demand is strong for multimodal data for robotics, computer vision and healthcare. Reported price points in 2024 were $2 to $4 per short-form video and $100 to $300 per hour of longer films, The Decoder reported citing Reuters.

Source: Defined.ai: partnership programs

Shutterstock: the contributor fund

Shutterstock groups contributors' images, video and 3D models into datasets licensed for training machine-learning and computer-vision models. Contributors earn an average 20% royalty on data-licence revenue, shared in proportion to how much of their content is in the datasets sold, and can switch data licensing on or off separately for images and video. Its data revenue was $120.3 million in 2024, and its initial deals with large technology companies were $25 million to $50 million each, most later expanded, VentureBeat reported.

Source: Shutterstock: data licensing and the contributor fund

Wirestock: datasets from creators

Wirestock supplies image, video, design, gaming and 3D datasets to AI labs from more than 700,000 creators, with $40 million of annual run-rate revenue and $15 million paid to contributors, TechCrunch reported in May 2026. Under its terms of use, creators can be paid a pro-rata share of 50% of what Wirestock receives from a dataset buyer; submitted content joins its Dataset Deals Program with no opt-out, and model releases are required for identifiable people.

Source: TechCrunch: Wirestock raises $23M (May 2026)

Company footage is a different sale

These platforms are built mainly for media libraries and creators. Footage a company records for its operations, such as dashcam, CCTV, worksite, production-line or aerial survey video, is in demand from robotics and world-model developers, but it raises different questions: faces, recorded audio, what workers were told and who owns client-commissioned captures. See selling company video footage to AI and robotics companies, and the video data prices tracker for what buyers have paid.

Guides for other record types

What buyers want, published payouts and the rules for each kind of record:

• Work orders and field service records
• Slack, Teams and email archives
• Source code and its history
• Call recordings and transcripts
• Patient records
• Medical imaging and diagnostics
• Deal, credit and portfolio data
• Payments and transaction data
• Security alerts and incident cases
• AI usage logs and agent trajectories
• Lab notebooks and R&D records
• CAD files and engineering data
• Video footage for robotics and physical AI
• AI video training data prices tracker

Next step

Count your hours by type and year, pull ten sample clips and check the rights. Origin can tell you which buyers would pay for the footage and which route fits.

Origin Data Partners works with companies that own data and helps get it bought. We are independent of the services on this page, and a buyer may pay us when a deal closes (how Origin is paid). See what your data is worth.