Commercial use, copyright, and indemnification.

The most basic question: does the agreement you signed (clicked through, more often) when you set up the account permit you to use the outputs commercially? Most paid plans on most generators say yes; many free plans say no, or say yes with carve-outs. The carve-outs are where people get into trouble.

What the licence typically grants in 2026:

• Reproduction, make copies of the image.
• Distribution, publish it to web, print, social, etc.
• Modification, edit, crop, composite into other work.
• Sublicensing, sometimes, for use cases like agency-to-client delivery; check the clause specifically.

What free plans typically restrict: commercial use entirely (some); commercial use only if outputs are watermarked (some); commercial use only if not displayed publicly (some); commercial use granted but you grant the vendor a perpetual licence to use your prompts and outputs as training data (most). Read the "feedback" or "data use" section of the terms specifically, it is where this lives.

What to read on a vendor's page:

1. The plan-tier table on the pricing page (commercial use is usually listed per tier).
2. The terms of service section titled "Content", "Output", or "Your Generated Material".
3. The community guidelines for additional restrictions on output type (e.g., no public-figure likenesses).

For platforms commonly discussed in this category, primary licence pages include Adobe's gen-AI user guidelines verified April 2026, OpenAI's terms verified April 2026, Stability AI's community licence verified April 2026, and the relevant pages on Midjourney, Ideogram, Leonardo, and others, link from /compare. Read them yourself; we don't paraphrase.

Whether the AI-generated image is itself copyrightable, that is, whether you can prevent others from copying it, varies by jurisdiction and by how much human creative input went in.

United States

The US Copyright Office issued a policy statement in March 2023 verified April 2026 reaffirming that human authorship is a prerequisite for copyright registration. Purely AI-generated material is not registrable. Material that contains AI-generated elements may be registered for the human-authored portions and arrangement, with the AI portions excluded. This is the Zarya of the Dawn precedent. The Office's subsequent Report on Copyright and AI verified April 2026 develops the position further, including for visual arts.

Practical implication: registering an AI-generated image as your copyrighted work in the US is not currently possible for outputs that are largely or entirely AI-generated. Where the use case demands enforceable copyright protection, for example, a brand asset you intend to defend, significant human creative input (selection, arrangement, editing, compositing) is required, and only those human-authored elements are protected.

United Kingdom

The Copyright, Designs and Patents Act 1988 contains a specific provision for computer-generated works at section 9(3): the author of a work generated by a computer in circumstances such that there is no human author is taken to be the person by whom the arrangements necessary for the creation of the work are undertaken. The protection term is 50 years rather than the usual life-plus-70.

The UK government consulted on AI and copyright in 2025 verified April 2026. Reform proposals include changes to the s9(3) regime and a separate text-and-data-mining exception.

European Union

EU copyright law generally requires the work to reflect "the author's own intellectual creation", a standard articulated in CJEU case law (Infopaq, Painer). Purely autonomous AI outputs are unlikely to meet this standard. Human creative input through prompting, selection, and editing may suffice, but case-by-case.

The EU AI Act verified April 2026 imposes transparency obligations on general-purpose AI providers, including providing summaries of training data. The Copyright Directive verified April 2026 contains the text-and-data-mining exception under Article 4 with the rights-holder opt-out mechanism.

USPTO and inventorship

Distinct from copyright, the US Patent and Trademark Office published inventorship guidance in February 2024 verified April 2026 on AI-assisted inventions. The Federal Circuit's Thaler v Vidal decision held that an AI cannot be a named inventor under the Patent Act. The two regimes (copyright and patent) ask similar questions about non-human creators with separate answers.

Indemnification is the vendor's contractual promise to stand behind the outputs of their model. If a rights holder sues you alleging that an AI-generated image infringes their copyright (because, for example, the model memorised and reproduced training imagery), an indemnification clause obliges the vendor to defend the case and pay damages, subject to defined limits.

Indemnification is the difference between "the licence permits commercial use" and "commercial use is safe for me to undertake". The first is necessary; the second is what the indemnification provides.

Vendors that have publicly offered IP indemnification

Adobe states that Firefly outputs on enterprise plans are covered by IP indemnification verified April 2026, subject to the enterprise contract. The positioning is part of Adobe's broader "commercially safe" marketing.

OpenAI announced a Copyright Shield verified April 2026 covering paid ChatGPT Enterprise and API customers; verify the current scope on the linked page.

Microsoft has published Copilot Copyright Commitment verified April 2026 language for some Copilot products; coverage depends on the product and tier.

Stability AI, Midjourney, Leonardo, Ideogram, and most consumer-tier generators do not generally indemnify consumer customers. Some offer enterprise contracts with tailored terms. Always confirm on the vendor's current legal page; coverage shifts.

What to ask a vendor before relying on indemnification

1. Which outputs are covered? (All outputs vs only outputs from a specified model version, only outputs that pass certain safety filters, only outputs generated under specified conditions.)
2. What is the cap on liability? (Per-claim, aggregate, multiple of fees paid.)
3. What conduct on my part voids the indemnification? (Bypassing safety features, providing reference images you don't own, using outputs to train your own model.)
4. What jurisdictions are covered?
5. Is the indemnification automatic on the plan I am buying, or is it conditional on signing an enterprise agreement?