dreambooth-for-diffusion

首个完整封装、一体化训练stable diffusion dreambooth的autodl镜像环境,可训练定制自己的独特大模型风格、人物,开箱即用,内含详细教程。

Stars
450

Dreambooth Stable Diffusion

autodl pip

autodlhttps://www.codewithgpu.com/i/CrazyBoyM/dreambooth-for-diffusion/dreambooth-for-diffusion

notebookwebuiDreamboothdreamArtistfix https://www.codewithgpu.com/i/CrazyBoyM/sd_dreambooth_extension_webui/dreambooth-dreamartist-for-webui

autodldreambooth-for-diffusion

autodlAA5000

gitREADME https://github.com/CrazyBoyM/dreambooth-for-diffusion

bhttps://space.bilibili.com/291593914

1.vscodesshautodlnotebook 2.()/root/dreambooth-for-diffusion/root/autodl-tmp/()

cd /root/autodl-tmp/dreambooth-for-diffusion

ckptdiffusers

  • sd_1-5.ckpt
  • nd_lastest.ckpt
#  
!python tools/ckpt2diffusers.py \
    --checkpoint_path=./ckpt_models/nd_lastest.ckpt \
    --dump_path=./model \
    --vae_path=./ckpt_models/animevae.pt \
    --original_config_file=./ckpt_models/model.yaml \
    --scheduler_type="ddim"
#  
!python tools/ckpt2diffusers.py \
    --checkpoint_path=./ckpt_models/sd_1-5.ckpt \
    --dump_path=./model \
    --original_config_file=./ckpt_models/model.yaml \
    --scheduler_type="ddim"

ckpt

diffusersckpt

python tools/diffusers2ckpt.py ./new_model ./ckpt_models/newModel_half.ckpt --half

float16--half

512*512

tools/handle_images.py center crop

python tools/handle_images.py ./datasets/test ./datasets/test2 --width=512 --height=512

testtest2 png/jpg--png

deepdanboorutags label.

!python tools/label_images.py  --path=./datasets/test2 

--path

deepdanbooru https://github.com/KichangKim/DeepDanbooru

other

pip install other/deepdanbooru-1.0.0-py3-none-any.whl 

pip install accelerate

NONO

accelerate config

train.sh

3~5

sh train_object.sh

Finetune 3000+

sh train_style.sh

A50008/1000

train/test_model.pymodel_pathprompt

python test_model.py

nohup sh train_style.sh &

:

nohup sh back_train.sh &

()

nohup.outvscode

tail -n 10 nohup.out

g

df -h

pip install diffusers
pip install ftfy
pip install tensorflow-gpu
pip install pytorch_lightning
pip install OmegaConf
... 

git

A
export http_proxy=http://100.72.64.19:12798 && export https_proxy=http://100.72.64.19:12798

A
export http_proxy=http://192.168.1.174:12798 && export https_proxy=http://192.168.1.174:12798

A
export http_proxy=http://10.55.146.88:12798 && export https_proxy=http://10.55.146.88:12798

xformers()

A5000

https://github.com/facebookresearch/xformers train/other pytorch1.12.x(~)

pytorch1.12.x

pip install torch==1.12.0+cu113 torchvision==0.13.0+cu113 torchaudio==0.12.0 --extra-index-url https://download.pytorch.org/whl/cu113

autodl

/root/autodl-tmp

https://www.autodl.com/docs/migrate_instance/

VScode

vscode

autodl

autodl /root/autodl-nas/,

https://www.autodl.com/console/netdisk .

cososs train/toolscos

autodlhttps://www.autodl.com/docs/scp/

diffusersdeepdanbooru nbardyPR tagscrosstyanNyanko LepsoniAUTOMATC1111 QQ455521885 by -