Official Answer to the Mystery of Copyright Ownership in AI-Generated Text-to-Image

AI-generated images shine on every timeline of social media, yet countless creators toss and turn late at night—who truly owns this painting? As algorithms condense the styles of millions of artworks into a few seconds of generation, the boundaries of copyright seem shrouded in a thin mist, while the official reply is like a faint light pointing us toward which direction to go.

Within the framework of copyright law, the protected object has always been the intellectual achievement of a ‘person.’ AI itself has no will, no emotion; it merely recombines the statistical features from its training data. Consequently, many legal scholars point out that if no natural person makes substantive choices during the creation process—such as carefully crafting the prompt, selecting and post‑processing the generated result—then the image is unlikely to be recognized as a copyright‑protected ‘work.’ On the contrary, if the user merely clicks ‘one‑click generate,’ it is more likely to be regarded as the AI model’s output, with the model’s owner potentially holding the relevant rights.

Major platforms have already clarified this gray area in their terms of service. Take a well‑known AI drawing service as an example: its user agreement explicitly states that users own the full copyright of content generated via the service, while the platform retains the right to use, copy, and redistribute that content. Other platforms adopt a more cautious stance, declaring that the copyright of generated images belongs to the ‘provider of the prompt,’ while simultaneously reserving in the fine print an intellectual property claim over the model itself. This dual declaration lets users enjoy convenience while harboring potential legal risks.

For independent illustrators, the rise of AI is a double‑edged sword. On one hand, it lowers the barrier to creation, allowing those without formal art training to quickly obtain visual results; on the other hand, the flood of AI‑generated images saturates the market, driving down commission prices and even snatching orders that once belonged to human artists. Even more troubling, some AI models secretly absorb copyrighted works during training, causing familiar brushstrokes or compositions to appear in the generated images without traceable sources. This leaves original creators feeling that their hard work is silently consumed, yet they find no clear path for enforcement.

Real‑world conflicts have already surfaced. Last year, an international image supplier sued an AI company, alleging that its model had unlawfully copied millions of protected photos during training, thereby infringing reproduction and distribution rights. The court noted in its deliberation that although the AI does not directly ‘copy’ a single image, its internal parameters in fact carry the expressive characteristics of the original work, and such indirect copying may likewise constitute infringement. In China, artists have also uploaded their own works to an AI creation platform, only to discover that the generated images closely resemble the originals; they then sent infringement notices to the platform, which denied liability on the grounds that the prompts were user‑uploaded, leaving a suffocating deadlock.