How an AI aimbot works: screen capture, vision models and human-like aim

An AI aimbot is aim assistance that never reads the game. Instead of pulling enemy positions out of memory, it looks at the screen — the same pixels you look at — finds the players in each frame with a trained vision model, and nudges your mouse toward them. That one design choice is why an AI aimbot can run on Valorant under Vanguard, on Rust and Apex Legends under EAC, and on Warzone under RICO, where anything that touches memory is the first thing to get flagged. This guide walks through the whole pipeline step by step, compares the 1 PC and 2 PC layouts, explains where the risk actually lives, and finishes with the settings that matter.

The short version

Every few milliseconds the software grabs a copy of your screen, runs it through a neural network that has been trained on thousands of hours of gameplay, gets back a list of boxes ("player here, head here"), picks the best target inside your field-of-view circle, and moves the mouse a small, smoothed amount toward it. Then it does it again. Nothing in that loop opens the game process, loads a driver or reads a byte of game memory — which is why we call it a vision aimbot rather than a memory aimbot.

1. CaptureA frame is copied from the screen — on the same PC, or streamed to a second PC over the network or a capture card.
2. DetectA vision model scans the frame and returns bounding boxes for every player and head it can see, each with a confidence score.
3. SelectBoxes outside your FOV circle are discarded; the rest are ranked by your priority rules and one target is chosen.
4. MoveThe mouse is moved a fraction of the remaining distance, shaped by smoothing and delay curves so the motion reads like a hand.
5. RepeatThe next frame arrives and the loop runs again — hundreds of times a second, with sub-0.5 ms spent on inference.

Step 1 — Capturing the screen

The model can only see what it is given, so capture is the foundation. On a 1 PC AI aimbot the frame is copied straight from the GPU's output on the machine you play on. On a 2 PC AI aimbot the gaming PC's video is sent to a second computer instead — over the LAN with our network capture, or through a physical capture card plugged into the second machine.

Some games make this harder than others. Valorant and the FACEIT client actively fight ordinary screen recording, which is why most vision aimbots tell you to buy a capture card for those two. Fave AI takes a different route — it captures through OBS, with a free UDP plugin, and needs no capture card on either build. The next section explains how. Resolution and refresh rate are not a constraint: 1080p through 4K, ultrawide and stretched, at 240 Hz and above, are all supported.

OBS UDP capture: no capture card, no ban risk

Most vision aimbots have a weak spot at Step 1. Either they hook the game's graphics output with their own capture code — which is precisely the kind of thing Vanguard and FACEIT look for — or they tell you to buy a capture card, an HDMI splitter and a cable and run the picture to the second PC by hardware. Fave AI does neither. It gets its frames from OBS Studio.

Why OBS?

OBS is the software every streamer runs. Riot, EAC, BattlEye and FACEIT all allow it explicitly — banning it would ban half of Twitch. So when Fave AI reads its frames from OBS, the gaming PC looks exactly like a streamer's PC: a whitelisted program doing a whitelisted thing. There is no custom capture hook for an anti-cheat to fingerprint, no unknown driver, and no unusual device plugged into the graphics card. Valorant and FACEIT, the two games that block ordinary screen recording, work out of the box, because OBS is the one recorder they were built to tolerate.

The free OBS UDP plugin

On its own OBS records to disk or streams to Twitch. Our plugin — free with every Fave AI license, not a separate purchase — adds one more output: it takes each frame OBS has already captured and sends it straight to the AI over UDP. On the 2 PC build the frames cross your LAN to the processing PC; on the 1 PC build the same OBS-based capture feeds the model on the machine you play on. Setup is three steps: install OBS, drop the plugin in, add the game as a source. No cables, no drivers, no hardware to buy.

Why UDP — and why this is the lowest-latency way to run an AI aimbot

UDP is the protocol real-time video was built on. Unlike TCP it never stops to confirm delivery or resend a lost packet: if one packet is dropped, the next frame is already on its way, which is exactly right when only the newest frame matters. The plugin is tuned for latency rather than picture quality — the model needs to know where a player is, not to produce a pretty recording — so over an ordinary gigabit Ethernet cable a frame is on the processing PC well within a single refresh at 240 Hz.

Compare the capture-card path: GPU → HDMI → the card's own encoder → USB → decoding on the second PC, with the card buffering one or more frames along the way. That chain adds several milliseconds before the model has even seen the picture, and every one of them is time the target spends moving. Put OBS UDP capture in front of Fave AI's sub-0.5 ms inference and the total time from "enemy appears on screen" to "mouse moves" is the shortest a vision aimbot can achieve — which is what people mean when they call Fave AI the lowest-latency AI aimbot.

What you need — and what you do not

• OBS Studio (free) on the gaming PC
• The Fave AI OBS UDP plugin — included free with your license
• For the 2 PC build: both machines on the same LAN. Ethernet is recommended for the lowest latency; Wi-Fi works
• Not needed: a capture card, an HDMI splitter, a second cable or any extra hardware. A capture card is still supported if you already own one and prefer it.

Step 2 — Detecting targets with a vision model

This is the "AI" part. The captured frame is fed into an object-detection network — the same family of model that lets a self-driving car spot pedestrians or a security camera spot a person. The network has been trained on an enormous set of labelled gameplay frames: for each one, a human marked exactly where every player and every head was. Over training the model learns what "an enemy" looks like across every map, skin, lighting condition and pose, so at run time it can look at a frame it has never seen and answer, in well under a millisecond, "there are two players here, at these coordinates, and here are their heads."

Two numbers decide how good a vision aimbot feels:

• Latency — how long the model takes per frame. If inference takes longer than a frame, the aimbot is always reacting to where a target was. Fave AI's pipeline runs at sub-0.5 ms, which keeps up with even 240 Hz.
• Accuracy — how reliably the model finds real targets and ignores everything else. A good model does not lock onto teammates, corpses, ragdolls or your own gun; a poor one will. This is where the size and quality of the training set shows.

Because the model is a general-purpose "find the player" detector rather than a per-game memory offset, one build covers many games. That is why a single Fave AI license spans Valorant, Rust, Apex Legends, CS2, FACEIT, PUBG, Fortnite, Warzone, Rainbow Six, ARC Raiders and ABI, and why new titles can be added by training rather than by reverse-engineering.

Step 3 — Choosing a target

The detector hands back every player it can see. The aimbot then applies your rules:

• FOV — a circle around your crosshair. Only targets inside it are considered. Small FOV feels like a subtle nudge; large FOV feels like a magnet.
• Target priority — closest to crosshair, closest to you, or lowest health where the game shows it. Most players run "closest to crosshair", because that is what a human would naturally shoot.
• Bone — head or body. Head priority wins one-tap games like Valorant, CS2 and Siege; body is steadier in spray-heavy games like Rust and Apex.
• Stickiness — once a target is chosen, the aimbot keeps it until it leaves the FOV or dies, rather than flicking between two enemies every frame.

Step 4 — Moving the mouse like a human

A raw aimbot would jump the crosshair onto the head in one frame. That looks impossible on a killcam and produces mouse input no human could generate, so the movement is shaped instead:

• Smoothing divides the remaining distance into steps, so the crosshair glides in rather than teleporting. Per-target smoothing lets the first pull be firmer than the final correction.
• Micro-corrections add the tiny overshoots and settle-backs a real hand makes, instead of a mathematically perfect line.
• Delay curves insert a short, variable reaction time before the aimbot starts moving, so you are not "reacting" in 0 ms every fight.

Where the movement is delivered matters too. On a 1 PC build the mouse input is generated on the gaming PC. For games with stricter anti-cheat you can route it through a hardware device instead — a MAKCU, Ferrum or Arduino — which plugs in as a genuine USB mouse and turns the aimbot's commands into physical mouse signals. On a 2 PC build that is how it always works: the processing PC computes the correction and sends it back to the gaming PC as mouse input, so the machine that plays only ever sees a mouse.

1 PC vs 2 PC: which layout?

Fave AI comes in two editions with identical features and identical pricing. The difference is where the model runs.

If you are unsure, start with 1 PC. If you later add a second machine, your license and cloud configs move with you.

AI aimbot vs DMA cheat

The other product family on this site is DMA. Both keep the cheat out of the game process, but they get their information in opposite ways, and that changes what each can do.

They also combine well. A DMA second PC is already a capable machine sitting next to your gaming rig; the 2 PC AI build can run on it, giving you DMA ESP in one game and an AI aimbot in a game DMA does not cover.

Why kernel anti-cheats struggle with it

Vanguard, EAC, BattlEye and RICO are kernel-level: they load before Windows, watch every process, every driver and every read of the game's memory. Their entire model is "find the thing attached to the game". An AI aimbot is not attached to anything — the game process is never opened, no module is injected, no driver is loaded, and memory is never read. Screen capture is a normal Windows operation used by Discord, OBS and Game Bar, and a 2 PC build does not even do that on the gaming machine.

What remains visible to them is the mouse input itself. A perfectly straight, perfectly timed pull to the head every fight is a statistical fingerprint, and the newer anti-cheats do look for it. This is why the humanization layer in Step 4 is not cosmetic — smoothing, micro-corrections and delay curves are what keep the input inside the range a real player produces.

Hardware: what you actually need

Inference runs on the GPU, so the GPU is the number that matters.

• 1 PC — Windows 10 or 11, an Intel i5-10400 class CPU or better, and an RTX 2060 or better. Weaker cards still work, but the model runs at a lower rate and you will feel it as slightly later reactions. Optional: a MAKCU, Ferrum or Arduino for hardware mouse output.
• 2 PC — the same i5-10400 / RTX 2060 spec, but on the processing PC. The gaming PC needs only Windows, the game and OBS. Both machines on one LAN — Ethernet for the lowest latency, Wi-Fi if you must — with frames travelling over the free OBS UDP plugin. No capture card, unless you already own one and prefer it.

Put your money where the model runs. A modest gaming laptop next to a strong processing desktop is a better 2 PC setup than the other way round.

Settings that matter

Every game and rank is different, but these starting points keep the aim useful without looking like a script. Tune from here and save the result as a cloud config.

• FOV — start small, around 3–5% of the screen. Increase only if you find yourself missing targets the model clearly saw.
• Smoothing — medium. Set it so a full pull takes several frames, not one. Lower it a little for close-range games, raise it for long-range ones.
• Reaction delay — a short, variable window. A fixed 0 ms is the easiest thing to spot on a review.
• Bone — head in Valorant, CS2 and Siege; body in Rust, Apex, PUBG and Warzone, where sprays and armour make chest hits the better trade.
• Priority — closest to crosshair. It matches what a human would shoot first and avoids the tell-tale "flicked to the wrong guy" moments.

The community config browser on Discord has per-game presets other players have tuned; they are a faster start than guessing.

Supported games

One license covers every title the model is trained on: a Valorant AI aimbot, Rust AI aimbot, Apex Legends AI aimbot, CS2 and FACEIT, PUBG, Fortnite, Warzone, Rainbow Six Siege, ARC Raiders and ABI, with new games added as they release. The per-game notes — which anti-cheat each runs under and what settings suit it — are on the 1 PC and 2 PC product pages.

Frequently asked questions

What is an AI aimbot?

Aim assistance that finds targets by looking at the screen with a trained vision model instead of reading game memory. It captures frames, detects players in them, and moves the mouse toward the chosen target with humanized motion. Because it never touches the game process, it works under kernel anti-cheats such as Vanguard, EAC, BattlEye and RICO.

Is an AI aimbot detectable?

There is nothing for anti-cheat software to detect on the game side: no injection, no driver, no memory reads. The only observable output is your mouse input, which is why humanization settings — smoothing, micro-corrections, reaction delay — matter. Play with believable settings and there is no fingerprint to find.

Does an AI aimbot work in Valorant?

Yes — it is the game most people buy one for. Vanguard is extremely good at catching anything that touches the game, and a vision aimbot does not. Fave AI captures Valorant through OBS, which Riot explicitly allows, so no capture card is needed on either the 1 PC or 2 PC build.

What is OBS UDP capture?

Fave AI's way of getting the picture to the model without a capture card. OBS Studio — the streaming software every anti-cheat already permits — captures the game, and our free plugin sends each frame to the AI over UDP, the protocol built for real-time video. On a 2 PC build the frames cross your LAN to the processing PC. The plugin is included with every license at no extra cost.

Why is Fave AI called the lowest-latency AI aimbot?

Because both halves of the pipeline are built for speed. OBS UDP capture delivers a frame well within one refresh at 240 Hz with no capture-card buffering in the way, and inference takes under 0.5 ms. The result is the shortest possible gap between an enemy appearing on screen and the mouse moving.

Does it see through walls?

No. A vision aimbot only knows what is on screen, so it cannot provide ESP or radar. If you want wall information, that is what a DMA cheat is for; the two can run side by side on a 2 PC setup.

Do I need a second PC for an AI aimbot?

No. The 1 PC build runs entirely on the machine you play on and needs no extra hardware. A second PC is optional: the 2 PC build moves inference off your gaming rig for zero FPS loss and a completely clean gaming machine.

Does an AI aimbot cost FPS?

On a 1 PC build with an RTX 2060 or better the impact is small. On the 2 PC build it is zero, because the gaming PC runs nothing but the game.

Which is better, an AI aimbot or a DMA cheat?

Different tools. DMA gives you ESP, radar and per-game features but needs a card, firmware and a second PC, and covers one game per build. An AI aimbot needs no hardware, covers ten-plus games on one license and cannot see through walls. Many players run both.

Is a capture card required?

No. The free OBS UDP plugin replaces it on both builds, including for Valorant and FACEIT. If you already own a capture card and would rather use it on a 2 PC setup, it is still supported — but you do not need to buy one.