ghost
Sign In

Tokkyu

33
233
1
Updated: Oct 4, 2024
styleanimeportraits
Verified:
PickleTensor
Type
Checkpoint Trained
Stats
233
Reviews
Published
Nov 28, 2022
Base Model
SD 1.5
Training
Steps: 1,440
Epochs: 3
Trigger Words
(tokki girl)
Hash
AutoV2
1DD55BE3B1
default creator card background decoration
vcm07's Avatar
vcm07

I trained this DreamBooth model using EveryDream. The model was trained using the art from artist Tokkyu at 57 images, for 3 epochs in a total of 1440 steps. I trained it on a A100 (80gb Vram) with 768x768 native size/variable buckets, batch size 6, at around 68gb Vram per epoch using very high resolution training images non cropped. EveryDream handled the aspect ratios correctly. The images were captioned manually each with a little help from pharmapsychotic's excellent new 2.1 CLIP Interrogator.

There is a visual guide explaining several aspects of the model and how you can get the highest quality in your txt2img generations.

The model requires an Aesthetic Gradient to get the highest level of quality. It is in the HF page/Guide. Click show more to see the links.

Guide : https://huggingface.co/vcm07/Tokkyu_Model

Aesthetic Gradient: https://civitai.com/models/1120/tokkyu-model-aesthetic-gradient