Then, we upscale it by 2x using the wonderfully fast NNLatentUpscale model, which uses a small neural network to upscale the latents as they would be upscaled if they had been converted to pixel space and back.
Then, we upscale it by 2x using the wonderfully fastNNLatentUpscalemodel, which uses a small neural network to upscale the latents as they would be upscaled if they had been converted to pixel space and back. This results in a pretty clean but somewhat fuzzy 2x image: ...