This isn't a gray area anymore. The EUIPO's position aligns with the CJEU's longstanding requirement that a work reflect the "author's own intellectual creation" — meaning personality-driven, free creative choices. When a model predicts pixels or tokens based on statistical weights, no human author is making those micro-decisions. The prompter sets constraints; the model fills the space. That gap matters legally.
Where it gets messy: hybrid workflows. If you paint over Midjourney output, curate 50 generations into a deliberate sequence, or use AI as one tool in a broader human-directed process — those human choices can create protectable elements. But the AI-generated portions themselves remain public domain. You're essentially collaging uncopyrightable material with your own expressive layer. The protection only covers your layer.
Practical implication: if you're building a commercial asset library with Stable Diffusion or DALL-E 3, you have zero exclusivity on the raw outputs. Competitors can legally scrape and resell them. Your moat shifts to:
Curation and sequencing— the editorial judgment in selecting and arranging** Post-processing**— manual retouching, compositing, style transfer you control** Prompt-to-workflow pipelines**— proprietary systems that encode your creative intent into reproducible processes** Brand and distribution**— the only defensible IP around pure AI output
The UK and US reach similar conclusions via different routes. UK law requires "skill, labour and judgment" — prompt iteration might qualify if it involves genuine selective judgment, not just trial-and-error. US Copyright Office refuses registration for AI-generated material but allows protection for human-authored elements in mixed works. China's Beijing Internet Court
didgrant copyright to an AI-assisted image where the plaintiff demonstrated extensive prompt refinement and parameter tuning — an outlier worth watching.
For developers building AI creative tools: document the human decision points. Log every manual edit, selection, and compositional choice. That audit trail becomes your evidence if enforcement ever matters. And stop marketing "copyright-free AI art" — the output isn't copyright-free, it's uncopyrightable. Different legal category, same practical outcome. The policy debate is whether this gap needs legislative filling. Rightsholders want sui generis protection for AI outputs. Open-source advocates argue the public domain should absorb them. Either way, the current framework treats AI as a camera, not a painter — and the person pressing the shutter only owns the photo if they framed the shot.
Team messaging that actually remembers why you built that feature 2h ago
The hype cycle promised mass adoption by 2024 — reality check 1d ago
[Where exactly does AI-assisted become AI-written 1d ago](/en/news/6966/)
[Why most companies will miss the AI-native shift entirely 1d ago](/en/news/6912/)
AI is making every side project feel like meaningless slop 2d ago
AI is killing the spirit of hackathons if we keep letting LLMs 3d ago
Next Why I still lose sleep over alignment even though I build with →
these real-world AI monetization case studies, with plenty of directly applicable cases.