HPS+版本v1.2 (ID: 116415)

Introduction

Are you tired of models that make too much sense, are too reliable, or have a clear purpose or application?
Me too.

Instructions For Use:

counterfeit v3 and counterfeit v3's easynegative would work with this model, but I do not habitually use negative prompts, and i didn't render my test images with counterfeit.

try like, typing some words into the text prompt and seeing what they do.
fusing this with a basemodel and training another lora on that would be very funny. someone should do that.

7/22/2023 update: I now specially recommend using the 'dynamic thresholding' addon for automatic1111's ui.

I like a 'mimic CFG scale' of 12,

a 'mimic scale scheduler' of 'linear up' with 4 'minimum mimic scale scheduler',

and a 'CFG scale scheduler' of 'half cosine down' with 4 'minimum CFG scale scheduler'.

Self attention guidance is good too, with values of 1.0 , 1.0.

Cautionaries And Caveats:

This lora is a synthesis of multiple other Loras made by unknown authors using unknown datasets made at the middle of the night & tested on '1girl' before release.

It could do literally anything.

描述:

Okay for reasons beyond my understanding or control the pruned 'hps.1.1+' lora has worse compatibility with other full-strength loras than the original deltas used to extract the lora from its basemodel.
I now denounce 'hps.1.1+' (36k KB) and endorse 'hps.1.2+' (147k KB), which is made from the same lora fusion recipe.

训练词语: