What is a YOLO aimbot and how does the screen loop work?

Why this search is messy

Most people searching “YOLO aimbot” are trying to separate three tools that get mixed together in Discord and YouTube titles: an internal that loads into the game, a cloud overlay that streams frames off-box, and a local screen-crop tool that only sees pixels.

The useful question is practical. You want a clear definition, a working mental model of the loop, and enough detail to pick a file format without wasting an install. This guide answers that, then maps it onto the ClawAim Windows desktop app: NVIDIA TensorRT or ONNX Runtime, no models inside Setup.exe.

Using any aim assist in a competitive title usually violates that game’s rules. Riot Games, Activision, Valve, Epic, and others can ban accounts for it. ClawAim’s terms put that risk on you. Nothing below is a promise of safety, rank, or “undetectable” use.

What is a YOLO aimbot?

A YOLO aimbot is an AI aim tool that treats the game as video. YOLO (You Only Look Once) is a family of real-time object detectors from Ultralytics. The app grabs a center crop of the desktop, runs inference, draws player boxes, and converts a chosen box into mouse movement.

That is different from color aimbots that lock onto a single hue, and from internals that read bone positions from memory.

ClawAim’s product loop is the same four steps on every game pack: crop, detect, lock, move. The YOLO weights sit on the game pack. FOV, aim keys, smoothing, and macros sit on the character profile. Switch profiles in-game with Alt+1 through Alt+9.

How does a screen-crop YOLO aimbot work?

The useful picture is a pipeline, not a magic “AI aim” toggle.

1. 01
   Crop. The app captures a rectangle around the crosshair (BetterCam or MSS on Windows). Exclusive fullscreen often returns a black frame; borderless or windowed is the usual fix. See docs fixes.
2. 02
   Detect. A YOLO model runs on the crop. Live inference loads a TensorRT .engine or an ONNX .onnx. A raw .pt is export-only on ClawAim. It is not loaded in the live loop.
3. 03
   Lock. The app picks a box (closest to center, class filter, friend/foe if the model has those classes).
4. 04
   Move. Mouse output goes through the method you configured: normal hold, hold_release, auto_trigger, Win32 API, Logitech G HUB, Arduino, Makcu, or RP2040.

Nothing in that loop writes into the game process. That is the factual distinction versus an injector. Anti-cheat can still flag unusual mouse statistics, overlay behavior, or third-party hardware. Treat “not an injector” as an architecture claim, not a bypass claim.

YOLO aimbot vs injector vs cloud overlay

Opinion, labeled as such: if your requirement is “do not load a module into the title,” screen-crop is the honest category. If your requirement is “never get banned,” no public aim tool can offer that. Anyone selling certainty is selling a slogan.

TensorRT vs ONNX: which file should you use?

This is the decision that wastes the most time. Install docs cover the clicks. The rule of thumb is below.

What is a TensorRT .engine?

Fastest path on NVIDIA. Built for this GPU and this TensorRT version. ClawAim ships that stack as ClawRT. You cannot copy an engine to a friend’s PC and expect it to load. If both an engine and an ONNX exist for a pack, ClawAim prefers the engine.

What is ONNX for a YOLO aimbot?

Portable YOLO export. Import per game in Setup. Runs without TensorRT. Slower than an engine on the same card. This is the correct first file for most people. See ClawAim + ONNX.

Is a .pt file a live model?

No. A PyTorch .pt is a training or export artifact. On ClawAim it is not a live model. Export it to an engine on the PC that will run the app.

ClawAim’s installer does not bundle models. Community .onnx and .pt files come from MossyModels, then you import per game pack. Example packs: Marvel Rivals often pointed at an engine; Warzone / COD often starts as ONNX. Format is not locked to a title. Any pack can use either file.

When is TensorRT worth the extra setup?

Use an engine when you already have an NVIDIA GPU, you installed ClawRT, and you care about inference speed on a 320–640 crop. Rebuild after a GPU swap, a driver or TensorRT change, or an “engine will not load” error. Do not hunt for a downloaded engine from another machine.

When should I start with ONNX?

Use ONNX if you want a file you can re-import after a reinstall, you do not want CUDA plus TensorRT yet, or you are testing whether a pack’s classes match what you see on screen. You can export an engine later from the same weights.

Which games can a YOLO aimbot run on?

A screen detector does not need a game-specific memory layout. It needs a model trained on that title’s characters, viewmodels, and UI clutter.

ClawAim keeps a model path per game pack. Current packs include:

• Marvel Rivals
• Warzone
• Valorant
• Roblox
• Fortnite
• Apex Legends
• CS2
• Rainbow Six Siege
• Rust
• Overwatch
• The Finals
• Arc Raiders
• Battlefield
• PUBG
• Delta Force
• Deadlock
• Bloodstrike

Empty packs stay empty until you import a model. Friend/foe packs need matching class names (often a .classes.json sidecar). A general “person” model will box teammates too. That is a model problem, not an aim-key problem.

What do you need to run a Windows YOLO aimbot?

• Windows 10 or 11 only.
• A license key, activated in the desktop app, not in the browser. One key binds to one PC.
• A YOLO detect model as .onnx or a .engine built on this GPU.
• NVIDIA GPU required for TensorRT. ONNX can run without it, with less performance.
• Capture stack: BetterCam or MSS. Admin launch if the crop is black.
• Config in %LOCALAPPDATA%\ClawAim\. Models in %LOCALAPPDATA%\ClawAim\models\<game>\.

Paid plans are duration licenses for the same app: $4.99 / day up to $99.99 / year. A short launch trial or offerwall key can be $0 for a limited window. Features do not change with price. Only the term does. See pricing or start free.

How do you set up ClawAim without wasting the first hour?

1. 01
   Install ClawAim-Setup.exe and activate the key on Setup. Walkthrough: app install.
2. 02
   Pick the game pack and character. Set an aim key (mouse-right is often 0x02).
3. 03
   Import an .onnx for that pack, or point Model at a local .engine.
4. 04
   Confirm system status is OK. Avoid exclusive fullscreen.
5. 05
   Start Aimbot, hold the aim key in-game. Switch characters with Alt+1–Alt+9.

If Start is blocked, the usual causes are: no model on this pack, key not activated, or capture still failing. Import again after you switch games. Packs do not share one file unless you copied it yourself.

Common mistakes that look like “the AI is bad”

• Using a .pt as if it were live. It will not load in the aim loop. Export or import ONNX.
• Copying a TensorRT engine between PCs. Rebuild on the machine that runs inference.
• Exclusive fullscreen. Black crop. Use borderless; try run_as_admin.bat.
• Wrong pack or wrong classes. Boxes on teammates, UI, or spectators. Fix the model, not smoothing.
• Boxes but no mouse move. Aim key, output method, COM/protocol for Arduino/Makcu/RP2040, or missing admin, not YOLO.
• Expecting one model to cover every title. Each pack has its own path for a reason.

Costs, alternatives, and when not to buy

Costs: free short key via launch trial or offers; paid from $4.99 (1 day) to $99.99 (1 year). You still need a model source and, for TensorRT, a compatible NVIDIA card.

Alternatives people actually compare: color aimbots (cheaper, brittle under new skins), internals (different risk model), and playing vanilla. A YOLO stack is heavier: GPU, file formats, capture quirks.

Skip this product if you need Linux or console, you refuse any terms risk, you do not have a model for your game, or you expected a kernel driver or an injector. If you are unsure whether a third-party tool is allowed, read that game’s official rules. That is the appropriate check, not a vendor slogan.

Key takeaways

• A YOLO aimbot sees pixels, not game memory. Crop → detect → lock → move.
• “Not an injector” describes architecture. It is not a ban-rate guarantee.
• On ClawAim, live files are .onnx or .engine. .pt is export-only.
• TensorRT engines are fast and machine-specific. ONNX is the portable start.
• Models are per game pack and are not inside Setup.exe.
• Competitive publishers can still ban screen-based assist. Use is your risk.

Frequently asked questions

Is a YOLO aimbot the same as an injector?

No. An injector loads code into the game process. A YOLO aimbot captures the screen and runs detection outside that process. Anti-cheat can still act on mouse data, overlays, or hardware. Architecture is not a ban-rate guarantee.

Does a YOLO aimbot work without an NVIDIA GPU?

Yes, if you run ONNX. TensorRT .engine files need an NVIDIA GPU plus the matching TensorRT stack (ClawRT on ClawAim). ONNX is slower on the same card but does not require CUDA or an engine rebuild.

Can I use one model for Valorant, Warzone, and Marvel Rivals?

You can point every pack at the same file. Detection quality drops on titles the weights were not trained for. Import a pack-specific .onnx or engine instead of sharing one general person model.

Why is my capture black?

Exclusive fullscreen and missing admin rights are the first two causes. Relaunch elevated, switch to borderless or windowed, and try MSS versus BetterCam. A black crop means YOLO never sees players.

Why do I see boxes but the mouse does not move?

The detector is working. Check the aim key, the mouse method (Win32, G HUB, Arduino, Makcu, RP2040), and whether that hardware path needs a COM port or admin. This is an output problem, not a YOLO problem.

Do I need to download models separately?

Yes, on ClawAim. Setup.exe does not bundle weights. Download .onnx or .pt from a model source you trust, then import ONNX or export a TensorRT engine on this PC.

Is ClawAim a cloud overlay?

No. Inference is local on your GPU or ONNX runtime. Frames are not sent to a remote box for the detect step. That keeps latency on your machine and avoids a cloud overlay account.

Will a YOLO aimbot get me banned?

It can. Most competitive games forbid aim assist. ClawAim does not inject, but publishers are not required to care about that distinction. There is no honest undetectable claim. Use is your risk.

How much does ClawAim cost?

The same desktop app is $0 for a short trial key (launch window or offers) or from $4.99 for one day up to $99.99 for a year. One key binds to one PC. Activation happens in the app, not the browser.

What is ClawRT?

ClawRT is ClawAim’s TensorRT installer path. You need it to build and run .engine files on this NVIDIA GPU. Skip it if you are staying on ONNX Runtime for portable inference.