Help fuel my passion! Even $1 makes a difference. ☕ https://ko-fi.com/citronlegacy

Please checkout all my Pokemon Loras: https://civarchive.com/user/CitronLegacy/models?tag=pokemon

Check out all the Art styles I’ve made: https://civarchive.com/user/CitronLegacy/models?tag=style

New v3 Version! Flatter colors with a desaturated tendency. The art looks kinda retro. Its really great for mixing with other styles because it gives a strong anime base.

• After collecting 890 images I realized most of Sugimori's art is retro looking. A lot of the art in this data set is from 1990-2000. Sure he still makes art but most of his art is from Pokemon gen 1 - 2

Shout out to Lykon. Not only has Lykon been and still is foundational in Anime AI generation but Lykon's Ken Sugimori Style Lora is one of the first Lora's I downloaded from CivitAI.

Old Version details

This is trained on a bunch of images by Sugimori so it kind of understands various Pokemon characters.

For example you can get a Marnie-lookalike with (marnie_(Pokemon):1), 1girl, short black hair, pink dress, black jacket,

The goal is to create the art style not characters but this is a nice bonus. Marnie is best girl.

Two versions trained on different datasets.

V1 is trained on the top 50 Sugimori images from Gelbooru

V2 is trained on the top 50 but with the no_humans tag applied meaning there were no humans in this dataset.

Suggested weight: 0.8

Please support the artist by checking out their page: https://twitter.com/SUPER_32X

Check out all the Art styles I’ve made: https://civarchive.com/user/CitronLegacy/models?tag=style

Training Data

v3

Trained on: 890 images

Training Model: AnyLoRA_noVae_fp16-pruned

flip_aug: False

Num of Repeats: 2

Unit is Epochs or Steps: Epochs

Number of Epochs or Steps: 10

Training Batch Size: 3

Total Steps: 5933

Resolution: 512

Network Dim: Dim 32 and Alpha 16

Lora Creation Process took: 01:30:59

v1 and v2

V1 and V2 Trained on 50 images

V1 and V2 Checkpoint: AnyLora

V1 and V2 Network Dim: 16

V1 Training time: 28 minutes

V2 Training time: 27 minutes (why? they had the same number of images and same settings? Maybe resolution played a factor?)