Another merge that turned out quite well, Has a baked in VAE.
Recommended settings: (use whatever you like, these work for me)
Sampling method: Euler Ancestral
Sampling steps: 30
Schedule Type: Automatic / DDIM
CFG Scale 5 - 7
Recommended positive prompt:
masterpiece, best quality, amazing quality, (all the rest of the stuff)
recommended negative prompt:
bad quality, worst quality, worst detail, sketch, text, signature, logo, username,
No need for overly long negative prompts, they just tend to make the quality worse. If you want something out of the picture, put it here, otherwise, can almost be left empty.
I have generated the sample images using plain txt2img, and then giving them couple rounds in img2img. The resizing is done in img2img also, with denoise set to 0.1 -0.3.
Description
FAQ
Comments (7)
Things I'm enjoying so far while playing with this:
Handles multiple characters better than many models, especially in terms of interactions
Full, bright colors
Very expressive faces
Simple enough style to avoid many of the usual AI gaffes
I not able to try it much do to all the page generation problems but the little I able to made with it looks good, I been looking a model that do some good cartoon style and this seems to make it well with the cartoon loras I use
Just when I found a great checkpoint for cartoon characters they undone all the changes and now is not available to site generator anymore, ohhh cruel world, and no my pc is not good enough to do it locally , ohhhh cruel world x2
use colab
amazing model! currently my top favorite
sorry to bother, do you have any future plans to update the model to illustrious v.2?
Maybe, maybe not.















