The Complete 2026 Guide to Using AI-Generated Images and Videos Commercially

TL;DR

• In 2026, AI-generated images and videos are commonly used in commercial work, but the rules are fragmented across jurisdictions and platforms. There is no single "yes/no" answer.
• US law: purely AI-generated visuals (no human creative input) cannot be copyrighted, but you can still use them commercially. Anyone else can also copy them — including your competitors.
• EU AI Act (in force since August 2024, with transparency obligations phasing in through August 2026): providers of AI systems that generate synthetic image, audio, video, or text content need machine-readable marking; deployers have disclosure duties for deepfakes and certain public-facing synthetic content.
• The bigger risks for most teams are not copyright — they are: (1) using AI images that depict real, recognizable people without consent, (2) generating outputs that reproduce trademarked brands or characters, and (3) violating a generation tool's commercial-use terms.
• Use the 6-question checklist below before publishing any AI visual to a paid context.

This guide is the long version. We also publish focused articles on each section linked at the end.

Two years ago, the question "Can I use this AI-generated image in a paid ad?" had no good answer. The law was in flux, the tools changed their terms every quarter, and most legal teams said "don't." In 2026, the answer is much clearer: you usually can, but four things need to be true at the same time — and the cost of getting any of them wrong is higher now than it used to be.

This is the guide we wrote for our own team at ImgIvy, and we publish it here because every week we get versions of the same questions:

• "Can I use a Midjourney image as the hero of my Shopify landing page?"
• "We trained a LoRA on our brand photos — does that change anything?"
• "Our agency wants to use an AI video as the opening shot of a TV spot. What do we need to clear?"
• "A founder I work with wants to use AI-generated headshots of fake testimonial customers. Is that legal?" (No, please don't.)

If you've ever asked one of these, this guide is for you.

Why AI-generated visuals are now a default option

Three things changed in the last 24 months:

1. Quality: the gap between the best AI images and licensed stock photos closed almost completely for most marketing use cases. AI video closed the gap for short cinemagraphs, B-roll, and stylized loops, though longer narrative video is still a few iterations behind.
2. Cost: a single Midjourney or Sora seat at $30–$60/month replaces what used to be hundreds or thousands of dollars in stock licensing for a small marketing team.
3. Time-to-shot: a campaign concept can go from brief to hero asset in under an hour. Stock libraries can't match that.

The trade-off is uncertainty. Stock licenses are boilerplate and well-understood. AI visuals come with newer, fuzzier rules — and the people you'd usually ask (lawyers, agency producers, ad platform reviewers) often don't know the answer either. The point of this guide is to give you enough structure to make decisions without needing a lawyer for every asset.

The 2026 legal landscape: what's settled, what isn't

Copyright: in most jurisdictions, AI-only outputs aren't protected

The US Copyright Office position, refined in a series of rulings between 2023 and 2025, is consistent: works that lack meaningful human creative authorship cannot be registered for copyright. A pure text-to-image output from Midjourney or DALL·E falls into this bucket. You can still use it commercially — there's no rule against using uncopyrightable material — but you can't stop competitors from copying it, and you can't license it to others as your own.

The practical consequence: if a visual is so important to your brand that competitors copying it would hurt you, don't rely on a pure AI generation. Either (a) add significant human creative input that you can document (compositing, painting, photography stages), or (b) commission a piece you can register normally.

The UK, Australia, and several other common-law jurisdictions take broadly similar positions. The EU's situation is more nuanced — some member states have older statutory provisions for computer-generated works — but for practical purposes, treat AI-only outputs as outside copyright protection across most of the Western market.

EU AI Act: marking and disclosure duties are phasing in

The EU AI Act entered into force on 1 August 2024, with many transparency obligations applying from 2 August 2026. The two provisions most relevant to AI visuals:

1. Article 50 requires providers of AI systems that generate synthetic image, audio, or video content to mark outputs in a machine-readable format (typically C2PA-style content credentials).
2. Deepfakes and certain AI-generated content presented to the public as real or informative must be clearly disclosed to viewers, unless an exception applies.

If you operate in or sell to the EU, plan now: keep an internal log of which assets are AI-generated, and add a small disclosure (caption, footer note, or alt text) on public-facing pages. Most ad platforms (Meta, Google, TikTok) have already added "AI-generated content" disclosure fields — use them.

Right of publicity: the biggest practical risk

This is where most lawsuits actually happen. The "right of publicity" (in the US, primarily state-level) and the related "personality rights" in Europe protect individuals from having their likeness used commercially without consent. AI doesn't create an exception — generating a face that looks like a specific real person, even if you never named them, is still actionable.

Two patterns to avoid:

• Look-alike prompts: "a woman who looks like [famous person]" or "in the style of [photographer/director]." The resulting image may be infringing even if no specific source is reproduced.
• Synthetic testimonials: AI-generated "customers" with quotes implying real endorsement. Banned by the FTC in the US (2024 updates to the Endorsement Guides) and similar bodies elsewhere.

If the visual contains a human face, ask: would a reasonable viewer believe this is a specific real person? If yes — even if the answer is "we made them up" — use a different visual.

Trademark and trade dress: still binding

AI doesn't strip trademark protection. If your generation produces a Coca-Cola can, a Mickey Mouse character, or a recognizable BMW grille, the rights of those owners still apply. Most major generation tools now filter prompts that name brands or characters, but they don't catch everything, especially when the brand element appears as an "accidental" output (a generic "soda can" generation that happens to look like Coke).

The safe rule: if a viewer could identify a brand in your image, treat that image the way you'd treat a stock photo containing the same brand — incidental use in editorial is usually fine, central use in advertising your own product is not.

Training data lawsuits: the part that worries everyone

There are ongoing lawsuits (Getty v. Stability AI, the consolidated artist class actions, Reuters v. various) about whether training on copyrighted material without permission infringes the source works. As of mid-2026, none of these have produced a final ruling that says "downstream users of model outputs are liable."

Risk to downstream commercial users so far appears low, but not zero. The pragmatic response: prefer tools that have either (a) trained on licensed/owned data (Adobe Firefly, Shutterstock AI, Getty AI) or (b) offer commercial indemnification (Firefly, Microsoft Designer, some enterprise tiers of OpenAI). For high-stakes campaigns where a takedown would be expensive, indemnification is worth paying for.

The 6-question commercial-use checklist

Run any AI visual through these before publishing. If you can answer "yes" or "N/A" to all six, you're in the safe zone.

1. Tool license: Does the generation tool you used explicitly permit commercial use of outputs from your subscription tier?

Free tiers and trial tiers often forbid commercial use. The Pro/Standard tier is usually fine. Check the date — terms change.

2. Training data: Was the tool trained on licensed/owned data, or does it offer commercial indemnification?

If yes, you're better protected. If no, treat the asset as carrying some background risk.

3. People: Does the image contain a recognizable human face?

If yes, and the person could be mistaken for a real individual, regenerate without that face or use a clearly stylized depiction.

4. Brands: Does the image contain recognizable trademarks, logos, or trade dress?

If yes, decide whether they're incidental or central. Central = remove.

5. Disclosure: Are you using this in an EU-facing context that requires AI disclosure?

If yes, add a caption, footer note, or platform disclosure field.

6. Endorsement implication: Does the surrounding text or context imply the AI-generated person, place, or product is a real customer, real endorsement, or real result?

If yes, change the copy or the visual. Synthetic endorsement is illegal in most major markets.

If any answer fails, you have three options: regenerate, swap to a licensed stock asset, or change the use context to remove the failing condition.

Where to get AI visuals you can actually use commercially

Three sources, ranked by risk profile:

1. Free AI-visual libraries (lowest friction)

Curated libraries that aggregate AI-generated images and videos with clear commercial-use notes. ImgIvy falls in this category — asset pages show source, license, format, resolution, and commercial-use context where available. Browse the AI Visuals category to see what's available.

Pros: free, instant, license is pre-cleared, no generation skill required. Cons: less brand-specific (you can't fine-tune to your visual identity), competitors can use the same asset.

2. Generate your own (most control)

The major 2026 tools, ranked by commercial-friendliness:

The simple framework: if the project is high-stakes (paid ad spend, packaging, anything that would be expensive to take down), pay for a tier with indemnification. If it's lower-stakes (a blog header, a Slack background, an internal slide), the unindemnified tiers are usually fine.

3. Commission a human-augmented piece (highest protection)

For brand-critical hero imagery, the safest 2026 workflow is hybrid: an artist or designer uses AI as a starting point, then meaningfully modifies the output (compositing, painting over, integrating photographed elements). The finished piece has a documented chain of human authorship, qualifies for copyright protection, and avoids most of the AI-specific risks.

This costs more, but for the single most important visual on your homepage, it's worth it.

Practical workflows by use case

Marketing landing pages

• Hero image: use either a curated free AI asset or generate on Firefly/Designer (for indemnification). Run the 6-question checklist.
• Below-the-fold supporting imagery: free AI assets from libraries like ImgIvy are usually the right cost/quality balance.
• Customer logos / social proof: never AI. Always real.
• Disclosure: if EU traffic is significant, add a small "Imagery includes AI-generated content" note in the footer.

Paid social ads (Meta, TikTok, Google)

• Static creatives: AI visuals are widely used and accepted. Disclose via the platform's AI-content toggle.
• Video ads: AI video is fine for short cuts (3–6 seconds), B-roll, and stylized transitions. Avoid AI-generated faces in lead-gen creatives — auto-moderation is aggressive about anything that looks like a fake testimonial.
• People shown in ads: prefer obvious illustration over photorealism if generating humans. Photorealistic AI humans trigger more moderation flags and audience suspicion.

YouTube B-roll and thumbnails

• B-roll: AI video (Runway, Sora, Pika) is now standard. Length is the limit — keep AI clips short and intercut with real footage.
• Thumbnails: AI imagery dominates here. The only real risk is generating something that could be mistaken for a real event ("Tom Cruise reveals secret recipe!" + AI face = strike risk under YouTube's manipulated content policy).
• Monetization: YouTube's policies (updated 2024–2025) require disclosure of "altered or synthetic content that could mislead viewers." Use the disclosure toggle in the upload form when required.

Print and packaging

• Lower risk than digital because reach is smaller and takedown is less common, but higher cost of error (you can't recall printed boxes).
• Always use an indemnified tier (Firefly, Adobe Stock AI, Shutterstock AI) for anything that goes to print.
• Always archive the prompt, model version, and generation timestamp alongside the final asset for at least 7 years.

Email newsletters and blog headers

• Lowest-risk environment. Free curated AI libraries are usually the right fit.
• The same 6-question checklist applies, but the failure consequences are smaller (an email goes to a list, not to a billboard).

The mistakes that get teams into actual trouble

In the cases we've seen:

1. Synthetic testimonials: AI-generated "customer" photos with quotes implying real endorsement. Banned by the FTC. Recall + fines.
2. Look-alike prompts: prompts that explicitly named living people, then used in commercial campaigns. Right-of-publicity claims.
3. Reproduced trademarks: AI outputs that closely resembled protected character designs (Disney, Nintendo, automotive logos). Cease-and-desist + asset takedown.
4. Free-tier outputs used commercially: most common mistake. Designer used Midjourney free trial for a paid client, the resulting brand assets had no commercial license.
5. Ignoring EU disclosure obligations: pre-2026 launches that didn't add disclosure are being retrofitted now. Plan for it from day one.
6. No paper trail: when a takedown notice arrives, teams can't show what was generated when, by whom, on what model version. Always log this.

What's coming in 2026–2027

Three trends worth planning for:

1. C2PA / Content Credentials become standard. Major camera makers, generation tools, and platforms are converging on a shared metadata standard for provenance. By 2027 expect "view content credentials" to be a one-click action on most platforms. Future-proof your workflow by preserving metadata through your asset pipeline.
2. Platform-side moderation gets stricter. Meta, TikTok, and Google have all expanded their AI-content moderation in 2025. Expect false-positive rates to drop and false-negative rates to drop too — meaning more accurate but more invasive checks.
3. The "AI-trained on licensed data" market segment grows. Adobe, Shutterstock, Getty, and several startups now offer "fully licensed" generation. Premium positioning, premium pricing, but the only path for many enterprise buyers post-2026.

How ImgIvy approaches this

We curate free AI-generated images and videos around three operating principles:

• Clear source and license context: asset pages show source and license notes so users can understand how a visual may be used.
• Content review before scale: we avoid publishing obvious recognizable people, trademarks, and brand-like outputs as commercial-safe assets.
• Provenance where available: when upstream files include useful provenance or generation metadata, we preserve it where the asset pipeline supports it.

We're not a substitute for your legal team on high-stakes work, but for the daily reality of marketers, founders, and creators making 50 visuals a week, we're built to make the safe choice the easy choice.

Related guides

• Are AI-Generated Images Copyright Free? What the Law Actually Says in 2026
• Can You Use Midjourney, DALL·E, and Stable Diffusion Images Commercially?

This guide reflects how ImgIvy interprets the legal and platform landscape as of May 2026. It is provided for general information only and is not legal advice. For high-stakes campaigns, packaging that ships at scale, or use cases in regulated industries, please consult a qualified attorney in your jurisdiction.