metadata
license: creativeml-openrail-m
Guizmus' test models
Those are medium to low quality models, single focus usualy, that can bring some fun but not really do significant things.
I'm not sharing the diffusers, just the ckpt and some examples.
TMNT v1.0
- training: 11 pictures, 1000 steps on 5e-6 polynomial LR with 8 bit adam
- Activation token :
TMNT person - Model link
Bendstact v1.0
- training: 100 pictures, 7500 steps on 1e-6 polynomial LR
- Activation token :
Bendstract Style - Model link
Sid v1.0
- training: 10 pictures, 2500 steps on 1e-6 polynomial LR
- Activation token :
Sid - Model link
Kirby v1.0
- training: 10 pictures, 2500 steps on 1e-6 polynomial LR
- Activation token :
Kirby - Model link
Simsbooth v1.0
- training: 122 pictures, 18k steps on 1e-6 polynomial LR
- Activation token :
Simsbooth person - Model link
Hogleg Style v2.4
- training: 5k steps lr 1e-6, 122 samples, then 2k more steps on a lower lr and a rebalanced dataset
- Activation token :
HogLeg Style - Model link
Illusion v1.0
- training: 1k2 steps, 12 samples
- Activation token :
Illusion drawing - Model link
NikeAir v1.0
- training: 1k2 steps, 10 samples
- Activation token :
NikeAir Shoes - Model link