I just tried generating on the new 1.1 pro model using the camera raw file name hack, and it spews something you would expect in a persons phone camera roll (I

I just tried generating on the new 1.1 pro model using the camera raw file name hack, and it spews something you would expect in a persons phone camera roll (I put these images here https://vm.tiktok.com/ZGdJr9T5g/ https://vm.tiktok.com/ZGdJr9T5g/ ). Along with occasional watermark. I wonder how much this will work on such distilled/undistilled/redistilled models (if they will release dev/schnell 1.1 version at all), and how much about training data can be inferred this way.