If running on GCP use the following configurations for your VM instance, Deep Learning VM for PyTorch 2.3 with CUDA 12.1, M125, Debian 11, Python 3.10, with PyTorch 2.3 and fast.ai preinstalled

Training data for dreambooth needs to be on Huggingface Hub for easy access from any system. The dataset used in the code below is at https://huggingface.co/datasets/bhuv1-c/valid-warehouses-dataset.

#!/bin/bash

# Clone the diffusers repository
git clone https://github.com/bhuvvaan/dreambooth-training.git

cd dreambooth-training

# Create a virtual environment
python3 -m venv diffusion-venv

# Activate the virtual environment
source diffusion-venv/bin/activate

# Install the diffusers package
pip install .

# Install accelerate
pip install accelerate

# Configure accelerate
accelerate config

cd examples/dreambooth

# Install requirements for the dreambooth example
pip install -U -r requirements.txt

# Login to Hugging Face
huggingface-cli login --token #your token here

export MODEL_NAME="CompVis/stable-diffusion-v1-4" # Model on the hub
export INSTANCE_HF_REPO="bhuv1-c/fruits-for-intent" # Dataset on the hub
export INSTANCE_DIR="fruits-for-intent" # Local dataset directory
export OUTPUT_DIR="md-intent-prediction-data-2" # Output folder on the hub

# Download dataset from Hugging Face Hub (only images)
python3 <<EOF
import os
from huggingface_hub import snapshot_download

# Download all files
repo_dir = snapshot_download(
    repo_id="bhuv1-c/fruits-for-intent",  # Correct string format
    local_dir="fruits-for-intent",
    repo_type="dataset",
    local_dir_use_symlinks=False  # Prevents extra cache files
)

# Filter only image files
valid_exts = {'.png', '.jpg', '.jpeg', '.bmp', '.gif', '.webp'}
for root, _, files in os.walk(repo_dir):
    for file in files:
        if not any(file.lower().endswith(ext) for ext in valid_exts):
            os.remove(os.path.join(root, file))  # Delete non-image files

# Check if dataset is present before training
image_files = [f for f in os.listdir(repo_dir) if f.lower().endswith(tuple(valid_exts))]
if not image_files:
    print("Error: No images found in dataset. Exiting.")
    exit(1)

print("Dataset downloaded and cleaned successfully!")
EOF

# Ensure dataset exists before proceeding
if [ ! -d "$INSTANCE_DIR" ]; then
  echo "Error: Dataset folder '$INSTANCE_DIR' does not exist!"
  exit 1
fi

# Remove any .cache directories
if [ -d "$INSTANCE_DIR/.cache" ]; then
  rm -rf "$INSTANCE_DIR/.cache"
  echo "Removed .cache directory."
fi

# Run training
accelerate launch train_dreambooth.py \
  --pretrained_model_name_or_path=$MODEL_NAME \
  --instance_data_dir=$INSTANCE_DIR \
  --output_dir=$OUTPUT_DIR \
  --train_text_encoder \
  --instance_prompt="a close-up photo of an apple, two lemon, one lime, one onion, three oranges and pear on a rustic wooden table"\
  --class_prompt="fruits on table" \
  --resolution=256 \
  --train_batch_size=1 \
  --gradient_accumulation_steps=4 \
  --learning_rate=3e-6 \
  --lr_scheduler="constant" \
  --lr_warmup_steps=0 \
  --max_train_steps=800 \
  --push_to_hub

Hyperparameters can be changed according to the users convenience. A helpful guide on hyperparameters can be found at https://huggingface.co/blog/dreambooth.

For original diffusers readme, refer to https://github.com/huggingface/diffusers