Best ways to sell media and video archive data to AI companies
By Gareth HoyleLast updated How we research
In short
The best ways to sell media and video archive data to AI companies are a vetted broker such as Sell My Data To AI, a licensing platform such as Versos, or a data partner such as Protege. A Versos blog post reports OpenAI paying $1 to $4 per minute of video and Adobe about $3 per minute, without citing a source. Every rights layer in the footage needs clearing first.
Key takeaways
- Versos is best for large libraries that want to keep footage in their own storage.
- Protege is best for holders of video, audio and speech data.
- Shutterstock data licensing is best for contributors with stock footage.
- Sell My Data To AI wrote this list and ranks itself first. The method section explains the criteria.
- Every entry's facts come from the company's own published terms, checked November 2026.
Media companies and archive owners license footage to AI video model developers, and clips paired with transcripts, shot logs or descriptions give a model both the video and text about it. Our media and video archive page and video training data page list what buyers ask for.
Video is priced by the minute or hour, not by headcount. A Versos blog post reports OpenAI paying $1 to $4 per minute for directly licensed video and Adobe about $3 per minute, but cites no source, so treat both as indicative. Shutterstock's data licensing page claims more than 70 million videos available to AI developers.
Footage carries layered rights. Performers' consent under section 182 of the Copyright, Designs and Patents Act 1988, music, contributor releases, commissioned programs where the broadcaster kept rights and agency footage cut into programs each need clearing. Our video and audio licensing guide covers each layer.
- 1
Sell My Data To AI
Best for archive owners that need every rights layer checked before sampling
Sell My Data To AI values footage archives and brokers licenses with vetted AI buyers. Our eight-point seller vetting identifies third-party content such as music and agency footage, checks chain of title for commissioned programs and draws a random sample across the archive. You pay nothing to be assessed, approve every buyer, sample and license term, and agree the success fee in writing before any introduction. We do not take custody of your raw data.
- Fee to seller
- Nothing to be assessed. A success fee only when a deal completes, agreed in writing before any buyer introduction
- Seller vetting
- Eight-point standard, from authority to sign to manifest accuracy, before any buyer sees your data
- Buyer vetting
- Company identity, intended use, security and retention, onward transfer limits and, for larger deals, ability to pay
- Data custody
- No custody of raw data. Buyers see redacted samples under a mutual NDA
- Your control
- You approve every buyer, sample and license term, and can withdraw at any point before signing
- 2
Versos
Best for large libraries that want to keep footage in their own storage
Versos is a video licensing platform that connects to existing cloud storage such as S3, indexes and enriches video libraries in place, and delivers datasets to AI companies and frontier labs while tracking provenance through to revenue settlement. It publishes no rates or fees. A Versos blog post reports OpenAI paying $1 to $4 per minute and Adobe about $3 per minute, without citing a source.
- Published payout
- Not published for Versos. Its blog reports $1 to $4 per minute (OpenAI) and about $3 per minute (Adobe), unsourced
- Size requirement
- Not published
- Exclusivity
- Not published
- Fee to seller
- Not published
- Payment timing
- Not published
- 3
Protege
Best for holders of video, audio and speech data
Protege licenses real-world data from data holders to AI model builders, from startups to frontier labs, in healthcare, video, audio and speech, spatial and physical data and other domains. Its site covers pre-training, post-training and evaluation data, including healthcare benchmarks built on clinical documentation. It does not publish payouts, exclusivity, fees or payment timing, and data partners start through a contact form.
- Published payout
- Not published
- Size requirement
- Not published
- Exclusivity
- Not published
- Fee to seller
- Not published
- Payment timing
- Not published
- 4
Shutterstock data licensing
Best for contributors with stock footage
Shutterstock licenses image, video, music and 3D content to AI developers and claims more than 70 million videos and 640 million images on its data licensing page. Its library comes from more than 3 million contributors in more than 150 countries, who join through its contributor hub. The page does not describe contributor compensation for AI licensing, so check the contributor terms directly.
- What it buys
- Does not buy company data. Licenses its own contributor library of images, video, music and 3D to AI developers
- Size requirement
- Not applicable
- Fee to seller
- Not published
- Payment timing
- Not published
- 5
Defined.ai
Best for video, music and sound effects through a vetted marketplace
Defined.ai runs an AI data marketplace where enterprises buy ready-made or commissioned speech, text, image, video and multimodal training data. It cites more than 1.6 million contributors across more than 150 countries and more than 500 languages and locales, lists ISO 27001, 27701 and 42001 accreditation, and says only 30% of partners pass its ethical vetting. Seller revenue share and payment timing are not published.
- What it buys
- Lists speech, text, image, video, music and sound effects, and multimodal datasets from partners
- Size requirement
- Not published. Partners pass an ethical vetting process, and Defined.ai says 30% are approved
- Fee to seller
- Not published
- Payment timing
- Not published
- 6
Labelbox Prism
Best for archives with consented likeness and voice
Labelbox Prism lists consented likeness and voice among the categories with buyer interest and says that data needs explicit, timestamped consent. It publishes deals from around $50,000 for smaller targeted datasets to more than $1 million for large enterprise partnerships, pays an upfront licensing fee at signing plus a revenue share on downstream use, and makes licenses exclusive by default.
- Published payout
- From around $50,000 (smaller targeted datasets) to more than $1 million (large enterprise partnerships)
- Size requirement
- Not published. Prefers high-volume B2C companies for support logs
- Exclusivity
- Exclusive by default, tailored to scope, sensitivity and strategic value
- Fee to seller
- Not published. Scoping is free and non-binding
- Payment timing
- Upfront fee at signing plus revenue share. Prism says first call to first wire takes under 21 days
- 7
Opendatabay
Best for packaged video datasets priced by the seller
Opendatabay is a marketplace for licensed AI training and fine-tuning datasets, showing 111 licensed data providers and 655 data products on its homepage across text, image, audio, video, code, agentic trajectories and synthetic data. Listing is free, sellers set the price and keep 70% to 95% depending on the sale price, and self-serve payouts arrive within 30 days of the end of the month.
- What it buys
- Lists text, image, audio, video, code, agentic trajectories, 3D, tabular, time-series, human feedback and synthetic data
- Size requirement
- Not published
- Fee to seller
- Free to list. Commission from 30% (sales up to $100) to 5% (sales above $250,000), set per transaction
- Payment timing
- Within 30 days of the end of the month for self-serve sales. Balances under $50 carry forward
- 8
FileYield
Best for owners who want to set license type and exclusivity
FileYield, operated by Dployers LLC, is a brokered marketplace where sellers list AI training datasets free and negotiate price, license type and exclusivity directly with buyers. Its taxonomy covers 2,566 data types across 36 categories. FileYield does not touch the data, process payments or hold escrow, and its terms say any facilitation fees are disclosed when you list, request or accept a deal.
- What it buys
- Lists datasets across 36 categories and 2,566 data types, from structured records to audio, medical imaging, code and video
- Size requirement
- Not published. Listings are reviewed before going live
- Fee to seller
- Free to list. Any facilitation fees are disclosed when you list, request or accept a deal
- Payment timing
- Arranged directly between seller and buyer
How we ranked this list
Sell My Data To AI wrote this list and ranks itself first. We are a broker with a commercial interest in the result, and some companies below offer a service similar to ours. Read the order with that in mind.
For media companies and archives we gave extra weight to platforms that name video, audio or likeness data in their published brief, and to those that track provenance and exclusions.
| Criterion | What we checked |
|---|---|
| Fit for the data type | Whether the company names this kind of record in its own published brief |
| Published terms | Whether payouts, size requirements and license terms appear on the company's own pages |
| Vetting | What checks the company says it runs on sellers, buyers or datasets before a sample changes hands |
| Exclusivity | Whether licenses are exclusive, non-exclusive or perpetual, and whether that is published |
| Fees | Whether the seller pays a fee or commission, and whether the rate is published |
| Payment timing | When sellers are paid, as stated by the company |
| Transparency | How much of the above you can confirm before a first call |
Every fact about another company comes from our profile of it in the data buyers directory, checked on that company's own pages in November 2026. Where a term is missing we write "Not published" rather than guess. After first place, the order reflects fit for a company selling or licensing its data, not a quality score, and we do not rate any company negatively.
We rank Sell My Data To AI first because this is our list and our service is built for this reader, not because we scored ourselves against the others. We are independent of the buyers we work with, and our terms are on our fees and how we vet pages so you can compare them line by line.
Frequently asked questions
How much is a minute of video worth to AI companies?
Do performers need to consent to AI training use of footage?
Is it better to use a licensing platform or license video directly?
Sources
- Copyright, Designs and Patents Act 1988, section 182: consent required for recording of live performance, legislation.gov.uk, accessed November 2026
- Versos homepage, Versos, accessed November 2026
- Sell video content as AI training data, Versos, accessed November 2026
- Protege homepage, Protege, accessed November 2026
- Data providers, Protege, accessed November 2026
- Data licensing, Shutterstock, accessed November 2026
- Defined.ai homepage, Defined.ai, accessed November 2026
- AI marketplace partnerships, Defined.ai, accessed November 2026
- Prism, Labelbox, accessed November 2026
- Opendatabay homepage, Opendatabay, accessed November 2026
- Fees and pricing, Opendatabay, accessed November 2026
- FileYield homepage, FileYield, accessed November 2026
- How it works, FileYield, accessed November 2026
- Terms of service, FileYield, accessed November 2026