BLOG
AI vision capture for poker: why GTO MAX AI reads the table faster and more accurately than any RTA
AI vision capture is the part of a poker RTA that turns your poker window into a hand the solver can work on: your cards, the board, the pot, every stack and bet, the button, whose turn it is. GTO MAX AI's capture does this from the pixels of a shared window in a fraction of a second, learns a room's cards once and recognises them forever, and confirms every card before it trusts it. This is what it reads, how fast, and why that reading is the most accurate available.
1. What the vision capture sees
Share the poker window once, press Start, and the capture reads the whole table from the pixels: your two cards, every board card as it lands, the pot, each player's name, stack and bet, the dealer button, who has folded and whose turn it is. It reads the table the way you do, by looking at it, which is why it runs on GGPoker, Natural8, BetKings, ClubGG, PokerStars and PokerNow without touching the client, and on any other room once you mark the table yourself.
The read is complete, not partial. The live table beside the shared window redraws every seat with its position, stack and bet, the board, and the pot, from the latest read, so you see exactly what the coach is working from. Tick "show what it reads" and the capture draws its boxes over your shared window too. Nothing is hidden behind a loading spinner.
2. How fast it is
Speed is what makes real-time assistance real. From a customer's session on a GG network table this week, the capture read a changed frame in a tenth to a third of a second on the server, with cards and digits it already knew recognised in milliseconds. A design it had never met took about one second, once. From then on those cards were read locally, instantly, at no cost.
That is fast enough for fast-fold poker. The board shows the moment the client draws it, your pair shows the moment it is dealt, and the solver has already worked ahead on the street, so the advice is on screen when your action buttons light up. Rush formats with a short clock were the design target, not an afterthought.
3. Why it is the most accurate reading available
Three design choices do the work.
- It learns your room's cards once and keeps them. After the first sight of a deck design, every card is matched on your own frames, pixel for pixel, rather than guessed by a general-purpose model each time. The result is the same card every frame, with no drift between hands.
- Every card is confirmed before it is trusted. A card is accepted only when two reads of the same pixels agree, and the colour of the pixels has the final say on the suit. A card caught halfway through its deal animation cannot be mistaken for another card, which is why the board never flickers and your hand never changes mid-hand.
- Numbers are read twice. Stacks, bets and the pot are read by learned digits and by an on-device reader, and the two must agree. Per-room rules handle big blinds versus currency, separators, and whether the room's pot already includes the bets in front of the players, so a 3.3 big blind bet is 3.3, never 33, and the pot is never counted twice.
Behind this sits a test bench of hundreds of real frames from every supported room, including the public GGPoker desktop set, run on every change. A reading rule added for one room cannot ship if it moves a single read on another. That discipline is why the capture behaves the same on a Monday as it did on the Friday before.
4. Where the AI sits, and where it does not
The word AI covers two very different things in GTO MAX AI's capture, and it is worth being exact. A vision model is used for sight-reading: the first time the capture meets a deck design or a new player name, it asks the model what it is looking at, learns the answer, and does not ask again for those pixels. Everything after that first sight is deterministic, local and instant. The piece on where AI sits in an RTA explains the rest of the stack, including why the solver and the opponent model are not language models.
This split is what keeps the capture both fast and cheap. A tool that sends every frame to a model pays in seconds and in dollars on every hand, and still gets a different answer when the model has a different day. A tool that learns once and recognises forever pays nothing after the first hand and gives the same answer every time.
5. Built for the way you actually play
Up to four tables at once, each with its own shared window and its own reader, the table where it is your turn lit up and its advice in front. Between hands the capture goes quiet: it reads when the table changes and does nothing when it does not, so an empty seat or a sit-out costs nothing. A Re-read button forces a fresh read of every box if you ever want to double-check a spot, and a three-minute recording button sends support exactly what the capture saw, so a new room skin is supported within days rather than months.
It runs in Chrome or Edge on Windows and Mac, with nothing installed and nothing injected into the poker client. The detectability piece covers what that means in practice: the capture looks at a shared window the way a video call does, and the poker software has no process to find.
6. What it feeds
The capture exists to feed the coach. Every read becomes a precise hand state: positions, effective stacks, the action so far, the price to continue. From that the coach shows two solver tables side by side, the Pure GTO line and the Exploit line adjusted for the player across from you, with the reason in words. Our player profiling page explains how opponent reads are built from showdowns and sizings over sessions, and the poker RTA guide puts the whole stack in context.
A solver is only as good as the hand it is given. Fast, complete, confirmed reads are the reason the advice can be acted on with the clock running, and the reason the same spot gets the same answer every time it comes up.
7. Try it on your table
The quickest proof is your own room. Open the app, share the table window, press Start, and watch the live table fill in beside the shared window. Within a hand you will see every seat, stack and card as the capture reads them, and the two solver tables the moment your action buttons appear. The guides cover each supported room and marking a table for any other.
GTO MAX AI: the RTA that knows the player.
GTO line, exploit line and the reason, live, on up to four tables, in Chrome. GGPoker, Natural8, ClubGG, PokerNow, PokerStars, and any room you mark yourself. Try it free for 30 minutes: no sign-up chat, no card.
Start a free 30-minute trialCoaching and study tool. Use it where real-time assistance is allowed: study, home games, private clubs, play money. Most online sites forbid real-time assistance in their terms.
More from the blog
- Is poker RTA detectable? How sites catch it in 2026
- GTO vs exploitative poker: when to leave the equilibrium, and by how much
- AI in poker: what RTA, bots and solvers actually do, and what they do not
- Playing GTO with real-time assistance: where the edge is, and where it is not
Longer reads: the poker RTA guide and all guides.
Open the app