doodad-v1-3-ckpt

作者 doohickey · 下载 0 · 点赞 2 · 更新 2023-11-01

stable-diffusion text-to-image

Doodad.ckpt

This is (A converted version of) the 1st part of a 2 (3?) part project meant to be used with Doohickey

The model was trained with Dreambooth-Stable. It mixes especially well with the style included in Doohickey. It's a finetuned version of the Trinart-Waifu-diffusion-50-50 included in this organization's models and was trained on 48 images from the author's (crumb's) Pinterest feed.

Parameter Value
resolution 512
train_batch_size 1
gradient_accumulation_steps 2
learning_rate 5e-6
num_class_images 120
max_train_steps 1200

Example outputs:

"portal to another dimension" doodad-v1-3-ckpt 预览 1

"a portrait of a dying god" doodad-v1-3-ckpt 预览 2

"photograph" doodad-v1-3-ckpt 预览 3