
There's one player in your game you can't beat — you know it the moment they sit down. Master Poker's Rival Clone builds an AI that plays like them, from their own hands, so you can drill against that exact opponent any time you like. Here's how it works.
Most poker AI is a solver — it computes how to play optimally, the same way against everyone. Your friend doesn't play a solver's ranges; they play their ranges — the hands they open, the sizes they bet, the spots they can't fold, the bluffs they love.
A Rival Clone is the opposite of a solver. It's an imitation: point Master Poker at a rival's real hand history — games you've played with them on Master Poker, or a PokerNow / n8 log you import — and it trains a model that reproduces that individual's style. Not "a good player." Them.
The hard part of cloning a real player is that you rarely know what they held. In a real game you only see a rival's cards when they show at the end of a hand — never when they fold. Naïvely trained, an AI would conclude "this player never folds" simply because every hand it can learn from is one they didn't fold. That's a trap, and Master Poker is built specifically to avoid it:
The result plays the way the real player does in the spots that matter: it 3-bets you light if they do, it over-folds to aggression if they do, it won't lay down top pair if they never would.
Fidelity answers one question: how much does this clone play like the real person? It's measured on hands the model never saw during training, across a few axes — does it enter pots at the same rate, size its bets like they do, and get its aggressive/passive decisions right. A higher score means a closer impression; a lower one usually just means "feed me more of their hands."
It is an honest number, not a marketing one. It can even drop when you add hands — that's the score re-scoring itself against a harder, larger sample, which is exactly what you want from a metric you can trust.
At least 500 hands of that rival — that's the minimum to build. A thousand or more sharpens the read; past the floor, the clone simply gets closer to the real person the more of their history you import.
No. It decides from public information only — the board, the action, the sizes — exactly like a real opponent across the table who can't see your hand. That's a deliberate design choice, not a limitation.
Never. Practice tables are a separate, play-chips-only mode for study. They don't touch your real games, your history, or your analysis.
No — Rival Clones are for studying opponents. Your own identity is marked "you" and kept out of the clone list, so the feature stays pointed at the rivals you actually face. (To study your own game, use your analysis dashboard — your HUD stats, leaks and hand replays live there.)
Build a Rival Clone with Pro First clone included · learned from real hands · play chips onlyFormat, chips, three house rules, and keeping the money straight.
Cash stakes tables, the 100BB rule, and a tournament clock that ends on time.
A ledger that sums to zero and payouts with the fewest possible transfers.