With AI image copyright, "uncertain" is the default
Between 2024 and 2026, use of AI image generators exploded — and copyright disputes grew with it. "Can I use an AI-generated image commercially?" The answer is not simple, and much of it is jurisdiction-dependent: the analysis below draws on US Copyright Office guidance, the EU AI Act, Korean copyright law, and the official terms of Adobe, Midjourney, and OpenAI — and the rules differ by country. Here is a guide to the major tools' licenses and the recent flow of disputes across the US, EU, and Korea: what is safe, and what is risky.

Fact 1: license comparison by tool
| Tool | Commercial use on free tier | Commercial use on paid tier | Output copyright |
|---|---|---|---|
| Midjourney | No (CC BY-NC) | Yes (Pro and above) | User-owned (with limits) |
| DALL-E (OpenAI) | Yes | Yes | User-owned |
| Stable Diffusion (open) | Yes | — | User-owned |
| Adobe Firefly | Yes (Creative Cloud) | Yes | User-owned + indemnification |
| Leonardo.ai | Partially | Yes | Varies by plan |
| Gemini Imagen | Yes (Google AI Studio) | Yes | User-owned |

Fact 2: the shift in US copyright registration
The US Copyright Office's position:
- 2023-2024: AI-generated portions cannot be registered; only the parts a human has "meaningfully transformed" qualify
- Late 2024: guidance clarified — a prompt alone does not amount to copyrightable human authorship
- 2025-2026: moving to case-by-case analysis; registrability turns on the degree of human editing, compositing, and selection
In short, a 100% AI-generated image cannot be registered for copyright in the US. But where a human meaningfully edits or composites, that contribution can be registered. Korea and the EU are trending in a similar direction — though the specifics remain jurisdiction-dependent.
Fact 3: the shadow of the training-data disputes
Midjourney, Stable Diffusion, and others trained on internet images, allegedly including large amounts of copyrighted material. The consequences:
- Getty Images v. Stability AI (UK) — proceedings ongoing into 2026
- New York Times v. OpenAI and related suits — some in progress
- A future risk that output from tools trained on copyrighted images could be treated as a derivative work
This risk is exactly why Adobe Firefly emphasizes training only on licensed data — it lowers dispute exposure.
Fact 4: five safe patterns (as practiced in the Korean market — the logic travels, but check your own jurisdiction)
Five patterns for using AI images commercially with lower risk:
Pattern 1: use only Adobe Firefly
The strongest legal backing. In a dispute, Adobe carries the responsibility. The cost is a Creative Cloud commitment ($20+/month).
Pattern 2: AI generation + 50%+ human editing
Take an AI-generated image and edit it by 50% or more — color, composition, compositing. This raises the odds that a human "creative contribution" is recognized.
Pattern 3: your own photo + AI transformation
Transform a photo you took yourself. You own the original's copyright; the copyright status of the AI-transformed portion is murky, but the dispute risk is low.
Pattern 4: clearly licensed tools with dispute indemnification
Beyond Adobe Firefly, some enterprise tools (Getty AI, Shutterstock AI) offer dispute indemnification. Costly, but safe.
Pattern 5: disclosed provenance + clear labeling
Label AI-generated images clearly as "AI-generated." Labeling mandates are advancing in some areas under advertising and consumer-protection law. A label lowers dispute risk.
Fact 5: five risk cases
Usage patterns with high dispute potential:
- Generating celebrity faces — near-certain infringement of publicity rights, in the US and Korea alike
- Mimicking a specific artist's style — images made with "in the style of [artist]" prompts; risky while the artist is living
- Variations on famous characters — Disney, Japanese anime characters, and the like: plain copyright infringement
- Logos and brand imagery — AI images resembling a company logo risk trademark infringement
- Regulated advertising (pharmaceuticals, finance) — regulated sectors impose separate limits on AI imagery
Recommendation: a safety guide for solo creators
General content (blog, social)
- Adobe Firefly, or your own photo + AI transformation
- Disclose the source and tool
- Never generate celebrities or famous characters
Commercial advertising
- Check the ad platform's guidelines in advance
- Only Adobe Firefly or other clearly licensed tools
- Legal review recommended if budget allows
Paid content (books, courses, SaaS)
- Main imagery from clearly licensed tools or a human designer
- Supporting imagery via Adobe Firefly and the like
- Be clear about who bears liability in a dispute (you do)
Checklist: five questions before using an AI image
- [ ] Have you read this tool's commercial-use license?
- [ ] Does the image avoid celebrities, famous characters, and known logos?
- [ ] Is it clear who bears liability in a dispute (the tool vendor? you)?
- [ ] Is this a domain that requires an AI-generated label (advertising, media)?
- [ ] Is the image's value proportionate to its dispute risk?
Conclusion
With AI image copyright, "uncertain" is the default state — and the rules are jurisdiction-dependent. The safest patterns are Adobe Firefly, or 50%+ human editing. The riskiest are celebrities, famous characters, and named-artist style mimicry. For a typical solo creator, Adobe Firefly plus source disclosure plus labeling is safe enough; commercial advertising and paid content warrant extra review.
One last line: The real price of an AI image is not the license fee but whether it is written down who bears liability in a dispute. A free tool missing that one line is itself the trap.
External references
Recommended primary sources on AI image copyright, tool licenses, and regulation:
- US Copyright Office, Compendium of US Copyright Office Practices + AI guidance (2023-) — primary source on US registration.
- EU AI Act (Regulation 2024/1689) — the European AI regulation text.
- The Korean Copyright Act + the Korea Copyright Commission's AI copyright guidelines.
- Adobe Firefly official license and indemnification pages.
- Midjourney Terms of Service.
- OpenAI DALL-E / image-generation official terms.
- Stability AI Stable Diffusion license + Getty Images v. Stability AI.
- Google Gemini Imagen official terms.
- Primary filings in New York Times v. OpenAI, Andersen v. Stability AI, and related cases.
- Korea Fair Trade Commission guidelines on AI labeling in advertising.




