Welcome to AutoCV¶

AutoCV is a Windows-first computer vision automation toolkit for capturing game or desktop windows, analyzing frames with OpenCV, and steering Win32 input back into the client.

Features¶

Win32 window discovery and capture via WindowCapture / Vision
Template matching, contour detection, and color filtering helpers
Human-like mouse motion and keyboard/mouse message sending via Input
Optional OCR via PaddleOCR (get_text())
Interactive tuning tools (AutoColorAid, ColorPicker, ImagePicker, ImageFilter)

Installation¶

pip install autocv

OCR requires PaddlePaddle; install one of the extras:

pip install "autocv[paddle-cpu]"
# or (Windows x64)
pip install "autocv[paddle-gpu]"

Requirements¶

Windows (AutoCV uses pywin32 for capture/input)
Python 3.10+

OCR Notes¶

The first call to get_text() may download OCR models and can require network access. If your environment blocks PaddleOCR model-host checks, set DISABLE_MODEL_SOURCE_CHECK=True before importing/using AutoCV, or pass disable_model_source_check=True when constructing Vision.

Reading the Documentation¶

The class structure and inheritance within the documentation is organized as follows:

AutoCV inherits from Input → Vision → WindowCapture.

This means AutoCV has access to all methods and properties of its parent classes. The reference pages are structured to reflect this hierarchy.

Example¶

from autocv import AutoCV

autocv = AutoCV()
autocv.set_hwnd_by_title("RuneLite")

# Optionally walk down to an inner canvas handle.
while autocv.set_inner_hwnd_by_title("SunAwtCanvas"):
    pass

# Patch GetCursorPos checks (requires the bundled `antigcp` extension).
autocv.antigcp()

autocv.refresh()

contour = autocv.find_contours((0, 255, 0), tolerance=50, min_area=50).first()
autocv.move_mouse(*contour.random_point())
autocv.click_mouse()