kaigouthro/info.md

## info.md

      
    Raw
  

              info.md
            
          
    step 1.

make sure you have virtualenv

sudo apt install python3-venv
pthon -m pip install virtualenv

step 2.

create a virtualenv

cd ~
mkdir dreambooth
cd dreambooth
virtualenv -p python3 venv
source venv/bin/activate

step 3:

install hugginfaces autotrainer

pip install autotrain-advanced torchvision

step 3.

try running the autotrainer setup:

autotrainer setup

step 4.

if it fails (like it did for me  first time..)

check if ou need to use a slightly earlier transformers/bits and bytes from what ou have ( it noted in m error)

step 5.

put your imagss inn a /images folder, and run the autotrainer with the default hugginface settings..
hange the instance prompt string to your own prompt, and voila.
autotrain dreambooth --model stabilityai/stable-diffusion-xl-base-1.0 \
    --project-name my_dreambooth_project --image-path images/ \
    
    --prompt "photo of a <YOUR_PROMPT> <YOUR_CLASS>"    \
    
    --resolution 1024 --batch-size 1  \
    --num-steps 500 --gradient-accumulation 2  \
    --lr 0.00001 --fp16 --xformers  \
    --train-text-encoder --use-8bit-adam  \
    --gradient-checkpointing
    
and DONE yo.. sdxl 1.0  dreambooth in 60 seconds