My standard usage rights language, built over years for traditional photo and video delivery, didn't map cleanly onto AI-generated work once that became a real part of my output, and I've had to genuinely rethink parts of my contracts rather than just swapping in new terminology on old boilerplate.
Who actually owns a generated image is a genuinely unsettled question
Traditional photography has clear, long-established norms around authorship and licensing that courts and industry practice have settled over decades. AI-generated content sits in a much less settled space, and I've stopped pretending my contracts can offer a client full legal certainty on a question that's still being worked out more broadly across the industry and, in places, in actual courts. What I can offer is clarity on what I'm actually granting and warranting, even where the deeper legal question remains open.
What I've actually changed in my contract language
I now explicitly disclose which deliverables involved AI generation as part of the delivery documentation, rather than treating it as an invisible part of my process, so a client has a clear record of what was generated versus captured. I've also added specific language around reference images used in generation, making clear when a client's own supplied reference material was used as an anchor, since that touches on their own rights to their existing brand assets in a way that's worth spelling out explicitly rather than leaving implied.
- Disclose which deliverables involved AI generation, don't leave it implied
- Spell out how client-supplied reference images were used in generation
- Be honest with clients that some underlying legal questions remain genuinely unsettled
Why over-explaining has worked better than confident boilerplate
I used to worry that flagging genuine uncertainty in a contract would make clients nervous. In practice, clients have responded better to honest disclosure of what's still unsettled than they would to confidently worded boilerplate that papers over a genuinely open question. Trust seems to build faster from acknowledged uncertainty than from false confidence, at least with the kind of clients I actually work with.