/vt/ - Virtual Youtubers

did something retarded just for the sake of doing it, since the legend is probably hard to read:
X Values (VAE): "None, Automatic, novelai.vae.pt, vae-ft-mse-840000-ema-pruned.vae.pt"
Y Values (Checkpoint Name):
>"animesfw-final-pruned.ckpt [22fa233c2d]
Exactly what it says it is. It surprisingly seems to want to load the SD VAE on its own??? >not_actually_infringement_VAE-2129-2982-0000.safetensors
NAI Curated, VAE baked in, fp16 pruned >not_actually_infringement_VAE.safetensors
before fp16 prune
>Gekijōban Shinseiki Dai Rantō HLL Mix 1.0 + 3.1 You Can (Not) Proompt II HD Remix &IRYS.safetensors
because launching the colab downloaded it...might as well

1. Even if you have a VAE baked into the model, you can override it. Baked in VAE are only active when you use "None" or "Automatic"
2. Baking in a VAE...doesn't change the size of the model? Weird.

>>44686341
>spoiler
Setting the VAE to automatic is entirely situational. If you set it to automatic, and you have a .vae.pt with the same filename as the model in your /models/Stable-diffusion folder? It loads that. Set it to automatic and you already had a VAE loaded before - as in it was set to something else at launch time? It keeps those VAE weights loaded. Set it to automatic with a VAE baked into the model? It uses that. How aptly timed...

>>44686933
xformers on/off

>>44687013
>old commit and you are in a more recent one. Could this be the cause? maybe idk.
Unless it's fucking ancient? The only thing I can think of for a "within the past month" change would be the "fix" to any SDE sampler that now makes it function the same in/out of batch mode. You have to toggle a setting to get the "old" behavior back batch size >1 output was different from batch size=1

>>44685291
>>44685345
kek, novel AI could/would do it: https://www.pixiv.net/en/artworks/102817042

>>44685542
I've been prompting lewd La+ for 2 days now. kek.