Maybe add a not that `lighttaew2_1.pth` is needed in the `vae_approx` folder to get high-quality TAESD previews in ComfyUI?
Adding the one from here fixed the warning I initially got and made them work:
https://huggingface.co/lightx2v/Autoencoders/tree/main
I actually get an error if I copy "lighttaew2_1.pth" into VAE approx, the TAESD preview works without the file normally even if the console says:
"Warning: TAESD previews enabled, but could not find models/vae_approx/lighttaew2_1" which is kind of funny as that's the single thing that prevents the workflow from running.
ValueError: Input and output must have the same number of spatial dimensions, but got input with spatial dimensions of [1152, 896, 1] and output size of (512, 398). Please provide input tensor in (N, C, d1, d2, ...,dK) format and output size in (o1, o2, ...,oK) format
I actually get an error if I copy "lighttaew2_1.pth" into VAE approx, the TAESD preview works without the file normally even if the console says:
"Warning: TAESD previews enabled, but could not find models/vae_approx/lighttaew2_1" which is kind of funny as that's the single thing that prevents the workflow from running.
ValueError: Input and output must have the same number of spatial dimensions, but got input with spatial dimensions of [1152, 896, 1] and output size of (512, 398). Please provide input tensor in (N, C, d1, d2, ...,dK) format and output size in (o1, o2, ...,oK) format
the latent previews are the low quality LATENT2RGB ones without that in vae_approx, though. I don't get the error you noted on the very very very latest github pull of Comfy.