The defendant argued that it did not develop the Tab AIGC model itself, but rather relied on a 3rd party vendor to generate such AIGC services by interfacing with the 3rd party application program, and therefore it should not be subject to the 2023 GAI Measures. The court however rejected the defendant's argument, quoting Article 22(2) of the 2023 GAI Measures, and ruled that "AIGC providers", by definition, also include those who provide AIGC services through interfacing with other application programs, thus finding the defendant qualified as an "AIGC provider" as defined under the 2023 GAI Measures.
Then the court acknowledged that the defendant had taken some measures, such as filtering certain keywords, to stop generating infringing pictures to some extent. However, during trial, one could still input other keywords related to Ultraman and generate pictures substantially similar to the prior copyrightable Ultraman images. Therefore, the court concluded that the defendant should take extra steps to the extent that its AIGC function no longer generates any pictures substantially similar to the prior copyrightable Ultraman works when its users input any Ultraman-related prompts.
In terms of the plaintiff's claim on training data, the court rejected its request for the defendant to delete the copyrightable Ultraman works from its training data base because the defendant itself did not train the AIGC model.