Can I use AI image and video generation commercially in my product?

For information only. This is not legal advice, and it is not a substitute for counsel reviewing your own use.

TL;DR

• AI-generated images and video can generally be used commercially, but provider terms, copyright law and disclosure rules create real limits.
• Purely AI-generated output has no US copyright, but your own creative contribution on top of it can be protected.
• If your users generate content inside your product, the compliance obligations sit with you.
• Runway Dev does not claim your outputs and does not restrict commercial use. Under an Enterprise agreement, it commits contractually not to train on your data, including on the models it hosts from other labs, and offers IP indemnification.

Commercial use of AI generation is five questions, not one. Sections 1 to 5 answer them in order.

1. Does the provider let you use outputs commercially?
2. Do you own what you generate?
3. Do you have the rights to your inputs?
4. What does the provider do with your data?
5. Do you have to disclose that content is AI-generated?
6. What changes when your users generate the content?
7. What should you check before you ship?
8. Commercial use on Runway Dev

1. Does the provider let you use outputs commercially?

Most providers permit commercial use of outputs. Four things in the terms decide whether yours does.

1. Ownership: most providers state plainly that they do not claim ownership of your inputs or outputs.
2. Plans: some providers permit commercial use only on paid tiers. Some restrict it above a revenue threshold. Free-tier terms are frequently different from paid ones.
3. API terms: providers frequently run separate terms for their developer platform. The developer terms are the ones that apply when your product calls the API.
4. Usage policy: commercial use is granted subject to compliance with the provider’s agreement and acceptable use policy. Output that breaches the policy is not licensed, and providers reserve the right to suspend accounts for violations.

2. Do you own what you generate?

The provider generally does not claim ownership. Whether copyright exists in the output is a different question, and at one end it is now settled.

2.1 Purely AI-generated output has no US copyright

The Copyright Office refused Stephen Thaler’s registration for an artwork he said his system created autonomously. The D.C. Circuit affirmed in March 2025. On 2 March 2026 the Supreme Court denied certiorari, without comment. Human authorship remains a requirement.

2.2 Prompts alone are not authorship

The Copyright Office’s January 2025 report concluded that prompts function as instructions conveying unprotectable ideas, and its registration guidance says the same. Jason Allen ran 624 prompts on “Théâtre D’opéra Spatial” and was still refused registration, because the AI determined the expressive elements.

2.3 Your own contribution is protectable

The Office has registered hundreds of works containing AI-generated material. The registration covers what the human did: creative selection and arrangement, or substantial modification. You disclose the AI-generated parts and claim the rest. The Office assesses each work individually, so keep a record of what you contributed.

2.4 Your output is unique in practice, but not exclusive in law

Generation is stochastic, and reference-driven work depends on your own inputs, so duplicate outputs are very unlikely. Uniqueness is not a legal right: stopping someone else requires copyright or trademark. For a logo or a recurring character, add your own creative work and register what you can.

3. Do you have the rights to your inputs?

Rights to every input are your responsibility, and no provider’s license grants them. Input rights are where most real problems start.

3.1 The output inherits every right in the input

Images you did not shoot, or shot under a license that does not cover derivative works, carry that limitation into the generation. The same applies to other companies’ products and logos in frame, to music you added rather than generated, and to a style taken from a living artist or an identifiable brand.

3.2 Likeness needs a release

A face, a voice or a recognizable likeness requires permission. Generating it from a reference does not create one.

3.3 Content your users upload becomes your problem

If your product accepts a photo, you inherit whatever arrives with it. If you cannot say where an input came from, remove it.

4. What does the provider do with your data?

Providers differ sharply on whether your inputs and outputs train their models. For unreleased product imagery, this is usually the clause that decides the vendor.

4.1 A commercial license does not tell you whether the provider trains on your content

Many providers reserve the right to use your inputs and outputs to improve their models. Establish whether that is a default or a contractual commitment not to.

4.2 A no-training commitment may not cover models hosted from other labs

A platform reselling another lab’s API cannot promise more than its own upstream agreement allows. Check whether a no-training commitment extends to every model on the platform or only to the provider’s own.

4.3 Retention is a separate question from training

Establish how long inputs and outputs are stored, and whether zero retention is available.

5. Do you have to disclose that content is AI-generated?

Disclosure of AI-generated content is increasingly required by law, not only by platform policy. Three jurisdictions require it today.

Rule	What it requires	In force	Penalty
EU AI Act, Article 50	Providers mark generated output machine-readably. Deployers disclose deepfakes perceivably, at first exposure.	2 August 2026, marking grace to 2 December 2026	Up to 15 million euros or 3% of global turnover
New York Synthetic Performer Disclosure Law	Conspicuous disclosure in any visual or audiovisual ad reaching a New York consumer that features an AI-generated human likeness.	9 June 2026	$1,000 first violation, $5,000 after
California AB 853	Generative AI providers embed latent disclosures in generated images, audio and video.	2 August 2026	Set by the statute
Meta, TikTok and YouTube policies	Labels on realistic synthetic media that a viewer could mistake for real.	In force	Platform enforcement

5.1 Platforms require labels on realistic synthetic media

Meta, TikTok and YouTube all require disclosure where a viewer could mistake generated people, voices, places or events for real ones. Stylized output and production assistance generally fall outside. All three can read C2PA Content Credentials and label automatically.

5.2 The EU requires marking and disclosure under the AI Act

Article 50 has applied since 2 August 2026. Providers of generative systems must mark outputs machine-readably, and systems already on the market have until 2 December 2026. Deployers publishing deepfakes must disclose the artificial origin perceivably, at first exposure, not in a terms-of-service page. Penalties reach 15 million euros or 3% of global turnover.

5.3 New York and California require disclosure in advertising

New York’s Synthetic Performer Disclosure Law took effect on 9 June 2026. Any visual or audiovisual ad reaching a New York consumer needs a conspicuous disclosure if it features an AI-generated human likeness that is not a real person. Penalties are $1,000 for a first violation and $5,000 after. Audio-only ads, translation and expressive works are exempt. California’s AB 853 requires generative AI providers to embed latent disclosures in generated images, audio and video from August 2026.

5.4 Labeling does not cure a fabricated endorsement

FTC rules on fabricated testimonials still apply. An AI-generated customer endorsing your product is a fabricated endorsement whether or not the performer is labeled as synthetic.

5.5 In a product, disclosure is engineering

Decide where the label renders in your interface. Preserve provenance metadata rather than stripping it on re-encode. Store which model produced each asset.

6. What changes when your users generate the content?

When your users generate content inside your product, you carry most of the compliance obligations.

6.1 Your terms have to mirror the provider’s restrictions

Developer terms commonly require you not to permit your users to breach the provider’s restrictions, which makes those restrictions yours to enforce. Mirror them into your own terms of service, and build the rest into the product: moderation before publication, a record of who generated what, and a route to suspend users and take content down.

6.2 Load-bearing features need negotiated terms

Once generation is a product feature rather than an experiment, ask for the enterprise agreement, written training and retention terms, IP indemnity, and version pinning so behavior does not change under you. IP indemnity means the provider defends you, and covers the costs, if a third party claims your output infringes their rights. Read the indemnity carve-outs, which commonly exclude cases where you lacked rights to the input or knew infringement was likely.

7. What should you check before you ship?

1. Terms permit commercial use on the plan you are on.
2. The provider does not claim ownership of outputs.
3. You know whether your content trains models, and how long it is retained.
4. Rights are clean on every input: imagery, likeness, trademarks, audio, style references.
5. You are not relying on exclusivity you do not have.
6. If users generate, your terms bind them and you can moderate and take down.
7. Disclosure decided per platform, and for the EU, New York and California where you operate.
8. For anything load-bearing: enterprise terms covering training, retention, indemnity and version pinning.

8. Commercial use on Runway Dev

8.1 Ownership and commercial rights

• Input ownership: you keep ownership of what you upload.
• Output ownership: Runway does not claim ownership of your inputs or outputs.
• Commercial use: Runway does not restrict commercial use of your outputs, subject to your compliance with the agreement and the Usage Policy.
• Exclusivity: outputs may not be unique, and similar outputs may be generated for other users.

8.2 Data and training

Under a Runway Dev Enterprise agreement, Runway commits contractually not to train on your data and supports zero data retention. The commitment extends to the models hosted from other labs: Runway has negotiated commitments from every third-party model provider that they will not train on customer content, inputs and outputs included, and every model on the platform passes Runway’s vendor management program. See the Enterprise Services Terms and the Enterprise third-party model FAQ for the contractual scope.

8.3 What the Enterprise plan adds

Runway provides IP indemnification on the Enterprise plan for covered claims that its services infringe a third party’s intellectual property rights, subject to the exclusions and conditions in the Enterprise Services Terms. The obligations cover the models hosted from other labs as well as Runway’s own. The plan also includes SOC 2 Type II, contractual data terms and built-in content moderation, with compliance documentation available through the Trust Center.

8.4 You are responsible for what you and your users generate

You are responsible for the content you and your registered users make available through the service. Clearances for likeness, trademarks and third-party material are yours to obtain, and a Runway license does not grant them. If your product gives your own users access to generation, each of them has to be bound by terms no less protective than Runway’s.

FAQ

Can I use AI image and video generation commercially in my product?

Most providers permit it. Runway does not claim ownership of outputs and does not restrict commercial use of them, subject to compliance with its agreement and Usage Policy. Provider permission is separate from owning copyright, from the rights to your inputs, and from any duty to disclose.

Can I use AI-generated content commercially on a free plan?

That depends on the provider. Some permit commercial use only on paid tiers, and some restrict it above a revenue threshold. Free-tier terms are frequently different from paid ones, so check the terms for the plan you are actually on.

Do I own AI-generated content?

Providers generally do not claim ownership of outputs. Holding copyright is a separate matter: the Supreme Court declined to hear Thaler v. Perlmutter on 2 March 2026, leaving the human authorship requirement in place, so purely AI-generated output gets no US copyright. Your own creative selection, arrangement or modification can be protected.

Can I copyright an AI-generated image or video?

You can register the human contribution. The Copyright Office has registered hundreds of works containing AI-generated material, covering what the human author added. Prompts alone are not enough: Jason Allen ran 624 prompts on “Théâtre D’opéra Spatial” and was refused. You must disclose more than a trivial amount of AI-generated material and disclaim it.

Can two companies get the same AI-generated output?

Duplicate outputs are realistically very unlikely. Generation is stochastic, and reference-driven work depends on your own inputs. Providers do disclaim any uniqueness warranty, but that is a disclaimer rather than a prediction. Exclusivity is separate, and comes from copyright or trademark rather than from the file.

Do I have to disclose that a video is AI-generated?

Disclosure is required in a growing number of cases. Meta, TikTok and YouTube require labels on realistic synthetic media. In the EU, Article 50 of the AI Act has applied since 2 August 2026, with machine-readable marking due by 2 December 2026 for systems already on the market, and penalties up to 15 million euros or 3% of global turnover. New York has required disclosure in ads featuring AI-generated human likenesses since 9 June 2026, at $1,000 for a first violation.

Does labeling an AI actor make the ad compliant?

Labeling does not make a fabricated endorsement compliant. FTC rules on fabricated testimonials apply regardless, so a synthetic customer endorsing your product is still a fabricated endorsement.

Does Runway train on my images and prompts?

Under a Runway Dev Enterprise agreement, Runway commits contractually not to train on your data and supports zero data retention. These protections cover the models hosted from other labs as well as Runway’s own. Review the Enterprise Services Terms and your applicable agreement for the protections that apply to your account.

Does Runway offer IP indemnification?

Runway provides IP indemnification on the Enterprise plan for covered claims that its services infringe a third party’s intellectual property rights, subject to the exclusions and conditions in the Enterprise Services Terms. The obligations cover the models hosted from other labs as well as Runway’s own.

Can I use a competitor’s product or a real person’s face in a generated ad?

A provider license does not grant those rights. Likeness needs a release and trademarks need permission, because the output inherits the rights position of your inputs.

What if my users generate the content instead of me?

You are running a service on top of the model. Providers typically require your end users to be bound by terms no less protective than the provider’s own, so the obligation sits in your terms of service, your moderation and your takedown process.

Do a provider’s API terms differ from its app terms?

Many providers run separate terms for their developer platform, and those govern generation inside your product. Check which set your API access falls under.

Is AI copyright law settled?

One point is settled: no human author, no US copyright. How much human input is enough is assessed case by case, with no bright line. Training-data fair use remains unsettled.