Stable Diffusion 3.5 Image Generation Script

This script generates images using Stable Diffusion 3.5 based on prompts from a CSV file. It's designed to help evaluate cultural representation in AI image generation.

Running Options

You can run this script either on your local machine or on a remote server. Choose the appropriate guide below based on your setup.

Option 1: Local Machine Execution

If you have a GPU on your local machine, follow these steps:

1. Set up Conda Environment

   # Create and activate environment
   conda create -n sd35 python=3.10
   conda activate sd35
   
   # Install dependencies from requirements.txt
   pip install -r requirements.txt
   
   # Or install packages manually if needed:
   pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
   pip install diffusers transformers safetensors pandas accelerate protobuf sentencepiece

2. Login to Hugging Face

   pip install huggingface_hub
   huggingface-cli login

3. Access the model on Hugging Face

   • Go to the https://huggingface.co/stabilityai/stable-diffusion-3.5-medium, accept the license to access the models.

4. Download Model

   # Create directory and download
   mkdir -p ~/stable-diffusion-models/stable-diffusion-3.5-medium
   python -c "from huggingface_hub import snapshot_download; snapshot_download(repo_id='stabilityai/stable-diffusion-3.5-medium', local_dir='~/stable-diffusion-models/stable-diffusion-3.5-medium')"

5. Prepare Script and Prompts

   • Modify paths in general_culture_SD3.py
   • Create prompts.csv with your prompts

6. Generate Images

   python general_culture_SD3.py

7. View Results

   • Check the output directory (or your specified output directory)
   • Images will be saved as PNG files with names based on the prompts

Option 2: Server Execution

If you're running on a remote server, follow these steps:

1. Connect to Server

   ssh username@server-address
   # Example: ssh [email protected]

2. Set up Environment on Server

   # Activate conda environment
   conda activate sd35

3. Transfer Files to Server

   # Upload script and prompts to server
   scp general_culture_SD3.py username@server-address:/path/to/destination/
   scp prompts.csv username@server-address:/path/to/destination/
   
   # Example:
   scp general_culture_SD3.py [email protected]:/home/chans/worldccub/

4. Run Script on Server

   python general_culture_SD3.py

5. Download Generated Images

   # Option 1: Download entire output directory
   scp -r username@server-address:/path/to/output ./local_output
   
   # Option 2: Zip and download
   # On server:
   zip -r output.zip output
   # Then download:
   scp username@server-address:/path/to/output.zip ~/Downloads/
   
   # Example:
   scp [email protected]:/scratch/chans/sd-data/output.zip ~/Downloads/

Prompt File Format

Create a file named prompts.csv with your desired prompts:

prompt
"Show me traditional Nigerian clothing Fullbody"
"Show me traditional Nigerian celebration"
"Show me traditional Nigerian food"
"Show me traditional Nigerian architecture"
"Show me traditional Nigerian art"

Command Line Arguments

• --model_path: Path to Stable Diffusion model
• --prompts_csv: Path to CSV file with prompts (default: "prompts.csv")
• --output_dir: Directory to save generated images
• --steps: Number of inference steps (default: 30)
• --guidance_scale: Guidance scale for generation (default: 7.5)

Troubleshooting

• CUDA errors: If you have GPU memory issues, try reducing the --steps value
• Model loading failure: Verify model path and Hugging Face login
• CSV file errors: Check that your CSV file has "prompt" header
• Server connection issues: Verify your SSH key and server address
• Permission denied: Check file permissions (use chmod +x if needed)

References

• Stable Diffusion 3.5 Medium
• Diffusers Documentation
• PyTorch Installation