AI Art and Copyright: Who Owns What You Generate
Rachel Kim
17 followers
Ownership is weaker than you think
In Australia, a work with no human authorship gets no copyright protection at all. If you typed a prompt and a model produced the image with nothing else from you, that picture arguably belongs to everyone, including the person who lifts it from your site. Other countries draw the line slightly differently, but the direction is the same: more human input means more ownership.
The platforms handle it by contract instead. Their terms usually grant you a licence to use what you make, and that licence can be withdrawn, limited to personal use, or conditional on you not breaking the rules. Read the section titled commercial use before you plan a product around the output.
The input side is the bigger fight
Models were trained on billions of images, many scraped without permission, and creators are still litigating it. That matters commercially in two ways:
- Courts are deciding what training data was allowed to contain
- Outputs can resemble an existing work closely enough to trigger a claim
- Some services now offer indemnity, insurance, if you are sued, read what it excludes
What quietly goes wrong
Logos, brand characters and trademarked designs appear in outputs more often than people expect, and using them in advertising is a trademark problem regardless of who made the image. The same goes for images that look like a real person, a living artist's signature style used to sell something, or a photo that happens to match an existing stock image closely.
Marketplaces add their own rules on top. Print-on-demand sites, stock libraries and app stores each carry separate upload policies, and several now reject purely generated work outright or require a declaration at the moment you upload. Check that policy before you build a shop around a style, because a platform can change it overnight and leave a whole catalogue stranded.
A workable checklist
- Keep your prompts and edits, evidence of human involvement helps any future ownership claim.
- Disclose AI involvement where a client, platform or publisher asks for it.
- Reverse search anything you plan to sell before it goes live.
- Never paste client work, unreleased products or personal details into a prompt.
- Prefer models whose training and licensing terms you can actually read.
Where this lands
Treat generated images as raw material rather than finished assets. Modify them, combine them, crop and repaint and build on them until your contribution is obvious, and you move from the weak end of ownership to the strong end. Do that consistently and the grey zone stops being a risk, it just becomes part of your process.
Common questions
What quietly goes wrong?
Logos, brand characters and trademarked designs appear in outputs more often than people expect, and using them in advertising is a trademark problem regardless of who made the image.
Where this lands?
Treat generated images as raw material rather than finished assets. Modify them, combine them, crop and repaint and build on them until your contribution is obvious, and you move from the weak end of ownership to the strong end. Do that consistently and the grey zone stops being a risk, it just becomes part of your process.
