CLI-Only Image Generator 🎨

Pure command-line image generation using Stable Diffusion. No UI, no web server - just a simple Python script.

Features

✅ Pure CLI - no web interface needed ⚡ SDXL Lightning - ultra-fast 4-step generation 🎮 Optimized for RTX 4090 (works on any NVIDIA GPU) 🚀 1-2 second generation time 📦 Uses HuggingFace Diffusers (industry standard)

Quick Start

1. Install Dependencies

cd imgen_cli
pip install -r requirements.txt

Optional but recommended for RTX 4090:

pip install xformers

2. Generate Your First Image

python generate.py -p "a beautiful sunset over mountains"

That's it! Your image will be saved in the outputs/ folder.

Usage Examples

Basic generation

python generate.py -p "a cyberpunk cityscape at night, neon lights"

With negative prompt

python generate.py -p "portrait of a cat" -n "blurry, low quality, distorted"

More steps for better quality

python generate.py -p "detailed fantasy landscape" -s 8

Different resolution

python generate.py -p "abstract art" -w 768 -h 768

Reproducible results (same seed)

python generate.py -p "red sports car" --seed 42

All options combined

python generate.py \
  -p "majestic dragon flying over castle" \
  -n "blurry, low quality" \
  -s 8 \
  -w 1024 \
  -h 1024 \
  --seed 12345 \
  -o my_images

Command-Line Options

Option	Short	Description	Default
--prompt	-p	Image description (required)	-
--negative	-n	Negative prompt	""
--steps	-s	Number of steps	4
--width	-w	Image width	1024
--height	-h	Image height	1024
--seed	-	Random seed	Random
--output	-o	Output directory	outputs
--model	-m	Model to use	SDXL-Lightning

Performance

With RTX 4090:

• First run: ~30 seconds (model download + loading)
• Subsequent runs: ~1-2 seconds per image
• VRAM usage: ~8-10 GB
• Resolution: Up to 1024x1024

Models

Default: SDXL Lightning (Recommended)

python generate.py -p "your prompt"

• Ultra-fast (4 steps)
• Excellent quality
• Best for most use cases

Alternative Models

You can use other models from HuggingFace:

# Stable Diffusion XL Base
python generate.py -p "your prompt" -m "stabilityai/stable-diffusion-xl-base-1.0" -s 20

# Stable Diffusion 2.1
python generate.py -p "your prompt" -m "stabilityai/stable-diffusion-2-1" -s 30

Troubleshooting

Out of memory errors

• Reduce resolution: -w 768 -h 768
• Use fewer steps: -s 2

Slow generation

• Make sure you have CUDA installed
• Install xformers: pip install xformers
• Check GPU usage: nvidia-smi

Model not found

The model will auto-download on first run. Make sure you have internet connection and ~7GB free disk space.

Tips for Better Images

1. Be specific in prompts: "portrait of elderly wizard with long white beard, magical staff, fantasy art" is better than "wizard"

2. Use negative prompts: Add "-n 'blurry, low quality, distorted'" to avoid common issues

3. Experiment with steps: SDXL Lightning works great at 4 steps, but try 6-8 for even better quality

4. Use seeds for iteration: When you get a good image, note its seed and reuse it with variations of the prompt

5. Batch generation: Create a simple loop in bash/PowerShell to generate multiple variations

System Requirements

• Python 3.10+
• NVIDIA GPU with 8GB+ VRAM (RTX 4090 recommended)
• CUDA 11.8+ or 12.x
• ~10GB free disk space (for models)

License

Uses HuggingFace Diffusers library. Model licenses vary by model.