city96 / sd-latent-upscaler Goto Github PK

It works great, but it's inconvenient to have to network to huggingface to find the model every time, and the model files aren't too big when I look at them, can't they just exist locally? My internet access is extremely low, it's clear that the process is fast, but waiting for the internet connection takes a long time.

It shows successful loading in the console, but no related plugin node is found when using it in the browser interface.

I'm sure I'm doing something wrong. I'm trying to integrate the latent upscaler into imaginairy but the latents look like this after upscaling. I'm using latent-upscaler-v2.1_SDxl-x2.0.safetensors. Input latent is [1, 4, 128, 128]

Any ideas how I messed this up?

original image:

Not sure what your training process is however were you training with the actual 1.0 vae?

or did you use the 0.9 vae? - it's known that an "invisible" (quite visible) watermark is baked into the 1.0 vae. (this is in your relation to the artifacts found with the SDXL VAE)