PowerAim

A modern, AI-powered aim alignment tool for accessibility, training, and fun.

Note

If you enjoy PowerAim, please consider giving the project a star ⭐ — it really helps. Thanks!

---

PowerAim is a universal AI-based aim alignment tool. It captures the screen, runs a YOLOv8 ONNX model on the frame, and nudges the mouse towards the detected target — fully configurable, with a clean Fluent UI built on .NET 10 and WPF.

PowerAim started as a fork of Babyhamsta/Aimmy but has since been heavily reworked: a decoupled service architecture, a complete trigger-system overhaul, a Fluent-styled UI, gamepad / AutoPlay support, localization in 9 languages, dynamic model sizes, and a much faster capture & inference pipeline.

PowerAim is 100% free: no ads, no key system, no paywalled features. It is source-available but not open source — please do not make commercial forks.

---

📖 Documentation

Full, searchable documentation is published via GitHub Pages:

https://fgilde.github.io/AI-Ming/

The docs cover installation, every feature in detail, model training, configuration reference, and troubleshooting. They are also bundled with the app and shipped offline — click the Help button in PowerAim's title bar to open them locally without an internet connection.

Quick links:

• 🚀 Getting Started
• 🎯 Features
• 🎮 Controller Mapping
• 🤖 AutoPlay
• 🧠 Training Your Own Model
• 🔧 Troubleshooting

---

Table of Contents

• Purpose
• How it works
• Features
• Setup
• Trigger System
• Performance Tools
• Web Model & Training
• Contributing Models
• Credits

---

Purpose

PowerAim was designed for gamers who are at a real disadvantage relative to able-bodied players:

• Physically or visually impaired gamers
• Players without access to a separate HID for controlling the pointer
• People practicing reaction time / hand-eye coordination
• Anyone training their FPS aim mechanically
• Long-session players who develop fatigue or sweaty hands

It is also a great research / debugging tool for anyone interested in real-time object detection on the desktop.

How it works

flowchart LR
A[Game on screen]
C[Screen Capture<br/>DXGI / GDI]
B[YOLOv8 ONNX<br/>DirectML / CUDA]
F[Prediction Filter<br/>Multi-Class + Confidence]
S[Sticky-Aim Selector]
T[Trigger System<br/>multi-trigger, charge, key operators]
P[Prediction<br/>Custom Kalman + Velocity]
M[Mouse / Gamepad output]

A --> C --> B --> F --> S --> P --> M
S --> T --> M

Loading

Each block is an independent service — the capture loop, the inference pipeline, the trigger logic, the aim/output loop. They communicate through clear contracts (ICapture, IPredictionLogic, IAction).

Features

Detection & inference

• DXGI Desktop Duplication capture with automatic GDI fallback (≈6× faster than GDI alone)
• Dynamic ONNX input-size support — no more hardcoded 640×640
• Multi-class YOLO models with per-class filtering
• LUT-based byte→float tensor conversion (lower GC pressure)
• Built-in Performance Benchmark that recommends the optimal model size for your hardware
• Optional inference FPS cap

Aim

• Custom 2D Kalman filter with lead-time prediction
• Velocity-based Shalloe & WiseTheFox prediction methods (no longer the broken upstream versions)
• Sticky Aim target lock between frames — no flicker between overlapping detections
• Movement-path selector: Cubic-Bezier, Lerp, Exponential, Adaptive, or Perlin-noise jitter

Trigger system

• Multiple independent triggers per profile, each with its own keys and behavior
• Charge mode with BeginIntersectionCheck + ExecutionIntersectionCheck
• AND/OR operators for trigger keys and anti-trigger keys
• Sequential vs simultaneous action execution
• Configurable head-area sub-region

UI / UX

• Fluent-styled UI on .NET 10 (Mica backdrop, light / dark / system-follow)
• Hamburger sidebar navigation
• Localization in 9 languages (en, de, es, fr, it, ru, tr, uk, zh)
• Modern in-app MessageDialog (slides down from the window header)
• Live monitor / window picker with thumbnail previews and on-hover overlay highlights
• Gamepad Test page with virtual vJoy + AutoPlay system

Anti-Recoil

• OpenCV crosshair-tracking based anti-recoil (replaces the original simple recoil compensator)

Mouse backends

• SendInput, MouseEvent, LG HUB, Razer Synapse, ddxoft

Setup

1. Install the x64 version of .NET Runtime 10
2. Install the x64 version of the Visual C++ Redistributable
3. Download the latest PowerAim release from the Releases page
4. Either run the Installer.exe, or extract the .zip and run the bundled .exe
5. Pick a model in the Models tab and click Active — that's it

For CUDA acceleration, download the _cuda variant of the release.

Trigger System

PowerAim's trigger system is a complete rewrite of the original Aimmy autotrigger.

• Each trigger is an ActionTrigger with its own name, active state, keys, actions, intersection checks, and timing.
• Trigger Keys / Anti-Trigger Keys support AND or OR operators — combine LMB AND Shift, or LMB OR Q, or NOT (R OR Tab) to block firing while reloading.
• Charge Mode lets the trigger pre-aim while a button is held: enters when the target enters the configured begin head-area, executes when it enters the execution head-area.
• Sequential / Simultaneous action mode controls whether multiple actions are sent in order or all at once.

Open Aim Tools → Triggers → Edit to configure visually with live previews.

Performance Tools

• Run Benchmark (Models tab) measures FPS / inference time / GPU% across a set of image sizes (320 / 416 / 512 / 640 / 800) and recommends the largest size that still hits ≥60 FPS on your hardware.
• Max Inference FPS (Prediction Config) lets you cap the loop — useful for laptops where you want to keep thermals in check.
• Image Size Override (Models) is used for ONNX models with dynamic input shapes.

Web Model & Training

The repo contains a TFJS export under Universalv3_web_model/. It is intended to help auto-label new training data via MakeSense.ai. Load your images, pick Object Detection, run the AI locally with YOLOv5, upload the web-model files, label, and export.

A short walkthrough video for training your own model:

Want to contribute a model?

See MODELS.md for the full step-by-step guide. PowerAim's in-app downloader merges models from this fork and from the upstream Babyhamsta/Aimmy repo — newer commit wins on a name conflict, fork wins on a tie.

Credits

PowerAim is built on the shoulders of Babyhamsta/Aimmy by BabyHamsta, MarsQQ and Taylor — without their original work and ONNX/DirectML wiring this project would not exist. Thank you. ❤️

Model creators (kept from upstream):

• Babyhamsta — UniversalV4, Phantom Forces
• Natdog400 — AIO V2, V7
• Themida — Arsenal, Strucid, Bad Business, Blade Ball, LGHub check
• Hogthewog — Da Hood, FN
• Ninja — MarsQQ's emotional support

PowerAim is source-available (see SourceAvailable.md). Commercial forks are not permitted.