screenshot-stitcher 0.1.0

CLI for stitching vertically scrolling iPhone screenshots into one long image

screenshot-stitcher

English | Chinese

A small CLI for stitching multiple vertically scrolling iPhone screenshots into one long image.

It is designed for ordered screenshots from the same device and works best on list-like pages, app stores, settings screens, and similar vertical interfaces.

This is not an AI image generation tool. It is a deterministic image-processing utility powered by OpenCV and NumPy: it crops stable UI chrome, searches for overlapping regions, verifies alignment with visual features, and chooses a clean horizontal seam. Your screenshots stay on your machine; the CLI does not upload images or call hosted AI APIs.

Showcase

Full grouped sample assets live under examples/cases/.

Case	Inputs	Stitched Output
App Store Search 3 screenshots Full result
Xiaohongshu Profile 3 screenshots Full result
Apple Homepage 5 screenshots Full result

Why Not Just Use GPT-Image-2?

Long screenshot stitching is mostly a structural alignment problem, not an image generation problem. A general image model can produce plausible-looking output, but it may duplicate browser chrome, repeat page sections, or blur seams that should stay exact.

The comparison below uses the same Apple homepage flow:

screenshot-stitcher	GPT-image-2

Features

• Stitch screenshots in the exact order you pass on the command line
• Detect overlap between adjacent screenshots and avoid duplicated content
• Choose a cleaner horizontal seam instead of blindly cutting and stacking
• Tune top/bottom UI cropping and horizontal edge masking with CLI flags
• Keep working with a fallback mode when overlap confidence is not strong enough

Local Processing

All processing happens locally. screenshot-stitcher reads image files from disk, runs OpenCV/NumPy matching in the current Python process, and writes the stitched output back to disk. It does not send screenshots, filenames, or image metadata to any external service.

Installation

Requirements:

• Python 3.10 or newer
• pip
• macOS, Linux, or Windows with prebuilt opencv-python wheels available

Runtime dependencies are installed automatically:

• numpy
• opencv-python

Install the CLI:

pip install screenshot-stitcher
screenshot-stitcher --help

If your machine has multiple Python installations, use python -m pip to install into a specific environment:

python -m pip install screenshot-stitcher

Install The Latest GitHub Version

To install directly from the repository:

pip install "git+https://github.com/mate-matt/screenshot-stitcher.git"
screenshot-stitcher --help

Install From A Cloned Checkout

If you are an agent such as Codex or Claude working inside a checked-out copy of this repo, install it from the repository root:

cd /path/to/screenshot-stitcher
pip install .
screenshot-stitcher --help

For editable local development:

pip install -e .
python main.py --help

To verify the example assets work after installation:

screenshot-stitcher \
  examples/cases/app-store-search/inputs/01.png \
  examples/cases/app-store-search/inputs/02.png \
  examples/cases/app-store-search/inputs/03.png \
  -o /tmp/app-store-search-stitched.png

Usage

screenshot-stitcher img1.png img2.png img3.png -o output.png

You can also run it directly from the repo:

python main.py img1.png img2.png img3.png -o output.png

Common flags:

• --top-crop: override the default top crop
• --bottom-crop: override the default bottom crop
• --no-navbar: use only the status-bar height as the top crop
• --no-tabbar: use only the safe-area bottom as the bottom crop
• --x-margin: crop both left and right edges during matching, default 40
• --template-height: template height used during overlap matching
• --threshold: confidence threshold for accepting overlap matches

How It Works

screenshot-stitcher does not call GPT-image-2, an LLM, or any external vision API. The pipeline is local and OpenCV-driven:

• Convert screenshots to grayscale and edge representations with OpenCV
• Ignore configurable top/bottom UI chrome and noisy horizontal margins
• Estimate vertical overlap with multi-scale template and row-profile matching
• Re-score candidate matches with ORB feature support and local anchor consistency
• Pick a low-difference horizontal seam inside the overlap
• Stack the original image regions into one long screenshot

Current Scope

This project is intentionally narrow:

• Same-device iPhone screenshots
• Vertical scrolling only
• Identical width across all inputs
• Input order is controlled by the user

Known Limitations

• Repeated layouts can still confuse overlap matching
• Temporary overlays, badges, or floating UI can break otherwise valid overlaps
• Dynamic headers and browser bars may require tuning --top-crop, --bottom-crop, or --x-margin
• It is not a general panorama or arbitrary image stitching engine

Codex Skill

This repo also includes a Codex skill at skills/screenshot-stitcher/ for routing screenshot stitching requests to the CLI and suggesting the most relevant flags.

Because the CLI contract is simple and the prompting surface is small, the same skill pattern can be integrated quickly into Claude, Codex, OpenClaw, Hermes, and similar agent-driven toolchains.

Development

• examples/cases/: successful stitched outputs used in the README showcase
• scripts/build_showcase_assets.py: regenerates lightweight preview images for the showcase cases

License

MIT

Runtime dependencies are installed from PyPI and are not vendored in this repository:

• NumPy: BSD 3-Clause
• opencv-python package scripts: MIT
• OpenCV: Apache 2.0

These are permissive licenses and are compatible with keeping this project under MIT. If you redistribute a bundled binary application instead of a normal PyPI package, include the relevant third-party notices for those bundled dependencies.