No, I'm very much talking about the Google models. From the original link: "We have taken multiple steps to minimize these concerns, for example in internal tr

No, I'm very much talking about the Google models. From the original link: "We have taken multiple steps to minimize these concerns, for example in internal trials, we apply input text prompt filtering, and output video content filtering. However, there are several important safety and ethical challenges remaining. Imagen Video and its frozen T5-XXL text encoder were trained on problematic data. While our internal testing suggest much of explicit and violent content can be filtered out, there still exists social biases and stereotypes which are challenging to detect and filter."