- Status
- Offline
- Joined
- Mar 3, 2026
- Messages
- 546
- Reaction score
- 7
Anyone still trying to outrun BattlEye with pixel-based or AI methods knows the struggle—input lag, model inference times, and the constant battle with detection vectors. This is a complete C++ source for an AI-based aimbot utilizing DirectML, which is a solid alternative to the usual TensorRT setups if you're running AMD hardware or just want a different approach to hardware acceleration.
The project is essentially a functional framework that was used for months without catching a manual or automated ban, though it requires some minor refactoring on the input side. The previous mouse movement method relied on a Razer Synapse 3 exploit which has been patched out, so you'll need to swap in your own move-event logic (SendInput, a kernel driver wrapper, or a logic-based move).
Technical Feature Set:
Architecture & Performance
This isn't some slow Python script. It's built in C++ and uses a DLL for the overlay. On an RX 5700 XT, it pulls roughly 144-200 FPS with an inference response time between 3ms and 13ms.
The Razer Situation & Input Fix
The original author used a mouse movement method that Razer recently nuked. To make this UD and functional again, focus on the move-loop. If you have a KMBox or a custom HID driver, it's a trivial swap. If you're staying software-only, look into cleaner ways to send mouse packets without triggering BE's flag on synthetic input.
Verdict: It's a high-quality base for anyone looking to build a private AI external. The screenshotting feature alone is worth it for anyone serious about training a Yolo/ONNX model that doesn't get confused by gadgets or debris.
Anyone already ported this to a KMBox setup yet?
You cant view this link please login.
The project is essentially a functional framework that was used for months without catching a manual or automated ban, though it requires some minor refactoring on the input side. The previous mouse movement method relied on a Razer Synapse 3 exploit which has been patched out, so you'll need to swap in your own move-event logic (SendInput, a kernel driver wrapper, or a logic-based move).
Technical Feature Set:
- AI Aim Assistance: Includes smoothing and sensitivity scaling to look more human.
- Triggerbot: Supports both Full Auto and Semi-Auto modes.
- Automated Model Training: Built-in functionality to screenshot whenever shots are fired, allowing you to build a custom dataset specifically for Siege's player models.
- Visuals: ESP localized within the FOV circle and a real-time FOV customization system.
- Utility: Show distance, recoil modifier (pull-down logic), and confidence thresholds to avoid jitter.
- Persistence: Auto-saving config system that tracks settings changes in real-time.
- Control: All features are togglable via NUMPAD for easy in-game adjustment.
Architecture & Performance
This isn't some slow Python script. It's built in C++ and uses a DLL for the overlay. On an RX 5700 XT, it pulls roughly 144-200 FPS with an inference response time between 3ms and 13ms.
To get this compiling, you need to link the following packages:
- DirectML
- OpenCV
- ONNXRuntime
The Razer Situation & Input Fix
The original author used a mouse movement method that Razer recently nuked. To make this UD and functional again, focus on the move-loop. If you have a KMBox or a custom HID driver, it's a trivial swap. If you're staying software-only, look into cleaner ways to send mouse packets without triggering BE's flag on synthetic input.
Verdict: It's a high-quality base for anyone looking to build a private AI external. The screenshotting feature alone is worth it for anyone serious about training a Yolo/ONNX model that doesn't get confused by gadgets or debris.
Anyone already ported this to a KMBox setup yet?
