InstructPix2Pix Image Editor

Text-guided image editing using InstructPix2Pix. Enables natural language based image modifications without manual editing tools. Supports batch editing, style transfer, and iterative refinement through an intuitive API and web interface.

Features

• Text-Guided Editing -- Modify images using plain English instructions (e.g., "make it sunset", "add snow to the mountains")
• Batch Processing -- Apply the same instruction across an entire directory of images
• Iterative Refinement -- Chain multiple editing instructions sequentially for complex transformations
• Style Transfer -- Apply artistic styles via natural language ("make it look like a watercolor painting")
• Parameter Optimization -- Grid search utility to find optimal guidance scales for a given editing task
• Side-by-Side Comparisons -- Automatically generate before/after comparison images
• REST API -- FastAPI-based service for programmatic access
• Gradio UI -- Interactive web interface for real-time editing

Architecture

instructpix2pix-editor/
├── src/
│   ├── editor.py          # Core InstructPix2PixEditor class
│   ├── pipeline.py         # End-to-end editing pipeline with pre/post processing
│   └── api.py              # FastAPI REST endpoints
├── examples/
│   └── demo.py             # Usage demonstrations
├── configs/
│   └── config.yaml         # Model and processing configuration
├── requirements.txt
└── README.md

Installation

git clone https://github.com/ln-bodicherla/instructpix2pix-editor.git
cd instructpix2pix-editor
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

Quick Start

Python API

from src.editor import InstructPix2PixEditor

editor = InstructPix2PixEditor()
editor.load_model()

result = editor.edit_image(
    image_path="input.jpg",
    instruction="make it look like a snowy winter scene",
    num_inference_steps=20,
    image_guidance_scale=1.5,
    guidance_scale=7.5,
)
result.save("output.png")

Iterative Editing

instructions = [
    "make it sunset",
    "add dramatic clouds",
    "increase the saturation",
]
result = editor.iterative_edit("input.jpg", instructions)
result.save("final_output.png")

REST API

uvicorn src.api:app --host 0.0.0.0 --port 8000

curl -X POST "http://localhost:8000/edit" \
  -F "[email protected]" \
  -F "instruction=make it look like an oil painting" \
  -o output.png

Batch Processing

from src.pipeline import EditingPipeline

pipeline = EditingPipeline()
pipeline.batch_process_directory(
    input_dir="./images",
    output_dir="./edited",
    instruction="convert to pencil sketch style",
)

Configuration

Edit configs/config.yaml to adjust model parameters, inference settings, and API configuration.

Parameter	Description	Default
model.model_id	HuggingFace model identifier	timbrooks/instruct-pix2pix
model.device	Compute device	cuda
editing.num_inference_steps	Diffusion steps	20
editing.image_guidance_scale	How closely to follow the input image	1.5
editing.guidance_scale	How closely to follow the text instruction	7.5
processing.max_image_size	Maximum dimension before downscaling	1024

Guidance Scale Tips

• image_guidance_scale controls fidelity to the original image. Higher values preserve more of the original.
• guidance_scale controls adherence to the text instruction. Higher values produce stronger edits.
• For subtle edits, use image_guidance_scale=1.8 and guidance_scale=5.0.
• For dramatic transformations, use image_guidance_scale=1.0 and guidance_scale=10.0.

Requirements

• Python 3.9+
• CUDA-capable GPU with at least 8GB VRAM (16GB recommended for high-resolution editing)
• PyTorch 2.0+

License

MIT License