— trained from their new OpenCLIP encoder, with higher resolution, a 4x upscaler, and a new depth-guided img2img model #