This is the base SD 1.5 checkpoint, with the following changes:
CLIP model replaced with ZeroInt's ViT-L-14-BEST-smooth-GmP-ft
Baked VAE replaced with Furception VAE (excellent for fine details regardless of content type)
This is basically an opinionated set of changes I found after a bunch of testing to provide an overall better basis to work off of.
Description
Initial version.
FAQ
Comments (5)
Is there a way to make this better?
I found that using DPO and / or SPO in any combination actually was worse than just using ZeroInt's significantly better CLIP-L, in the particular case of SD 1.5.
How did you replace the CLIP in the SD1.5 model?
I just used "Checkpoint Saver" in Comfy. Original UNET, new Clip, new Vae as the inputs.
@diffusionfanatic1173 Genius idea
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.