Speculative Previews
Ghost-piece overlay showing the bot's next intended placements, updated with the latest pieces entering the queue.
17 training modes, 9 powered by Cold Clear. Frame-precise controls. Targeted drilling for every competitive skill.
You will learn to play like this
Skill is built in feedback loops of input, read, correct, and repeat. For decades, the traditional way to practice has been to play normal games and hope the situation you want to drill randomly surfaces. It's a fool's errand for deliberate practice. Liemtris isolates the loop you're training and runs it to exhaustion. 17 modes. Cold Clear analysis. Frame-precise controls.
Core Gameplay
The baseline modes: standard guideline play, survival under garbage pressure, live T-Spin pattern detection, and combo-training terrain.
Standard guideline play with 7-bag randomizer.
Full guideline rule set with SRS rotation, complete wall-kick tables, 7-bag randomizer, T-Spin detection via 3-corner rule, and back-to-back and combo scoring. The baseline mode all other training modes build on.
Cheese Race. Configurable garbage rate, trains smart downstacking.
Also known as Cheese Race. Garbage rows with single holes insert at a player-configured rate, pushing the stack toward the ceiling. Every block stacked on top of a hole costs ~2.5 pieces on average to dig back out, so Survival trains the discipline of clearing holes before they get buried. Row-type-aware hole generation produces realistic garbage patterns. Cold Clear is tuned here to minimize pieces per line of cheese — press the solution key to see what the bot is thinking.
Find the guaranteed T-Spin in N moves or fewer.
Like Chess's "find the mate in 2," Liemtris hands you a board and queue with a guaranteed T-Spin solvable in N moves. Find it. Toggle the Solution button to overlay the answer as ghost pieces. The most direct way to train attack and firepower. Every puzzle is generated at runtime, so the bank is unlimited and you never see the same one twice.
Pre-built 4-wide towers with open center channel for combo training.
Cold Clear reads the full visible queue and guides you to the continuations that keep the combo alive. Deadends are mathematically inevitable; when one is unavoidable, the system finds the fastest correction so you exit with the biggest attack the combo can still deliver.
The Engine
Cold Clear was trained on 100 million competitive guideline versus games for attack speed. It learns to hold T-pieces back for attacks rather than burn them in stack fill, maintains back-to-back, fires combos and spikes when the situation calls for it, and looks 20 moves ahead under pressure. This is the engine every top competitive streamer trains against, and it powers every AI-assisted mode in Liemtris.
Ghost-piece overlay showing the bot's next intended placements, updated with the latest pieces entering the queue.
Real-time placement analysis surfacing the bot's chosen move during gameplay.
Allow Cold Clear to use hold in its evaluation, or don't. Changes what the bot considers optimal.
A vibe thing. Watch the bot play at the speed your eye enjoys most.
Choose among General VS, Cheese Race, 9-0, 6-3, 2-7, 5-4 stacking, center 4-wide tower stacking, and combo-continuation settings.
Each AI-assisted drill uses the pretrained style that matches its skill target, from versus pressure to opener and combo study.
Just for fun. Press random keys and Cold Clear governs the actual movement, so it looks and feels like you're playing fast and smart.
Modes powered by Cold Clear
Opener & PC Training
Controlled-queue practice for named openers, continuation drilling from bag 2 onward, Cold Clear-powered PCO setups across 5 configurations, and Cold Clear's continuous PC loop.
Custom-opener entry, continuation drilling, and mirror support.
Enter any opener you want to practice into the dependency chart interface and set the piece dependencies yourself — anything you can describe, you can drill. Liemtris ships with Honey Cannon, Mountainous Stacking 2, TKI, DT Cannon, Gamushiro, and DPC, plus continuations and PC continuations for all of them. Practice bag 1 in isolation, then bag 2, bag 3, and deeper by selecting continuation nodes. Mirror mode doubles pattern coverage. Quiz mode presents a queue and asks whether it can produce the selected opener, training the mental calculation instead of the motor execution. Reference visuals stay onscreen the whole time.
All 5 canonical bag-1 PCOs — LOJI, STZ, Grace System — every solution, every ordering.
Every canonical bag-1 PCO to drill: the 5 LOJI setups, STZ, and Grace System — with every solution, every ordering, with or without hold. LOJI and STZ are included in mirrored form too, with and without an I-piece held at the end of bag 1. Win condition is the perfect clear. Built for drilling entry, stabilization, and the continuations that follow.
Watch an omniscient bot loop perfect clears endlessly. Just for fun.
Just for fun. Watch what perfect block stacking looks like when the mind has eleven pieces of foresight: continuous, back-to-back perfect clears, indefinitely. With that much queue knowledge a PC loop is always possible, and Cold Clear runs it as a persistent session — pre-solving the next clear while the current one plays. Step through a move at a time or let it run.
Finesse & Visualization
Input-precision and cognition drills: finesse sequences, piece-landing prediction, flash-based board reading, and infinite piece-shaped carvings.
Isolated finesse drills, filterable by piece, rotation, and column.
Focus on input mechanics alone, without the cognitive load of figuring out where the piece optimally goes. Turn on Cold Clear's solution preview and just put the piece where it belongs. Configurable finesse preview depth from 0 to 6 pieces — see only the next move, or read several ahead. Use more keys than optimal or misdrop and the info panel tells you what you did wrong; the piece resets to spawn so you can try the input again.
Endless finesse drills inside piece-shaped carvings.
Same loop as Finesse Trainer with a different visual: a stream of piece-shaped carvings appears on the board and you fill each one with the matching piece using optimal input. Misdrop or finesse fault and the piece resets to spawn so you can try again. Two sub-modes: pure drill (random piece, rotation, and column) and Game-Flow (carvings placed inside bot-guided realistic board states). Filter by piece, rotation, and column.
Predict the resulting terrain from a sequence of inputs without ever seeing them placed.
Trains your mind's eye to picture the future state of a board so you can instinctively read several pieces ahead instead of being anchored to the current state. The mode hands you a terrain plus finesse movements for 1–6 pieces; your job is to visualize the resulting terrain without placing them. Press the Solution button to check.
Trains instant recognition of gap widths in your stack.
Identify the flat parts of the stack from the prompt in the panel. The point is to instantly see how many cells you need to fill — once you know a gap is 5 wide, the piece combinations that lay flat into it are immediate (a flat L and Z, or a flat I and an upside-down J, and so on). Trains the count-at-a-glance recognition every top player runs in the background.
Pattern Recognition
Cold Clear-backed pattern training: C4W construction, live T-Spin sandbox, overhang puzzles, asymmetric height-difference setups, and forecastable-pattern drills under garbage pressure.
Center 4-wide opener and side towers, tuned across every bag.
Cold Clear has been tuned to open and maintain C4W against any bag sequence. The opener builds a 3-block residual in the center 4 columns, choosing residual shapes with the most continuations rather than obscure ones, then constructs the two side towers around it. Side selection for the left or right 4-wide stack is coming soon. Practice anticipating the bot's placements, or let it play and study its construction directly. Best paired with a visualization habit: the 3-wide towers leave almost no room for error if you haven't read the queue ahead, so foreseeing the resulting terrain before you place is non-negotiable here.
Every 1–5 piece overhang setup, exhaustively drilled.
Most players know the basic single-piece overhang setups and not much else. Across 1 to 5 pieces, there are more than 550 clean ways to combine pieces into an overhang. Knowing them means you can build the T-Spin into the stack instead of leaving an ugly residual after taking it with the only solution you happened to spot. Every setup, drilled to recognition.
Every way to resolve every left/right height delta.
The foundation of a T-Spin is resolving the height difference between flanks, because there is no twist until the setup is level. More than 2,500 ways exist to resolve a height delta, and Liemtris has them all. Filter by side (left, right, both) and by available pieces to focus on the cases you're weakest at.
Drills every 1-, 2-, and 3-piece placement across 3-wide terrain.
Pattern Recognition isolates the core read behind C4W stacking. It trains every possible way to stack 1, 2, and 3 pieces inside a 3-wide terrain, with the expectation that you visualize the resulting terrain before you commit. Use the solution key to check your work. Each terrain cycles into the next set of piece fits, so the drill builds the same forward-visualization skill as Visualization mode, but narrowed to 3- and 4-wide columns. That focus is small enough to drill exhaustively and general enough to transfer across contiguous columns on the full board.
Pre-insert attacks under your stack and learn what discovered attacks look like.
Most players neglect that the stack beneath the current terrain feeds future terrains. Pre-emptively planted overhangs can expose a T-Spin the moment intermediate lines clear, but setups like these are a pain to find in the wild. Forecast lets you insert garbage and attacking opportunities directly under your stack by hand, so you can viscerally see what a discovered attack looks like on your board and train the eye to spot it in real play.
Drill T-Spin calculation against fresh queues on a familiar terrain.
If a board accepts a T-Spin, the terrain that got it there has usually followed the right principles — good height management, clean residuals, the things a strong stack does anyway. Sandbox keeps that terrain fixed so you can take advantage of it, then hands you a fresh queue each round and lets you work out, in your mind's eye or by trial, every way to land a T-Spin with the pieces you have. Generate as many queues as you want against the same board.
Roadmap
Walkthroughs of every SRS kick, including rare and asymmetric kicks.
Every SRS kick presented with the terrain that produces it, the rotation input, and the resulting finesse. Includes the asymmetric I-piece kicks and the rare kicks most players never encounter in normal play.
The Settings Catalogue
Every setting surfaced in Liemtris, organized by system. Tier indicators mark standard, niche, and novel features relative to other competitive stackers.
Per-device input tuning: DAS, ARR, soft drop, and tap-back DAS independently configurable for keyboard, physical joystick, and physical d-pad.
DAS / ARR / Soft Drop Speed
DAS, ARR, and soft drop speed independently configurable down to 0ms.
Tap-Back DAS
Reverse-jerk inputs aren't perfectly discrete in real play. After a charged slide one direction, a fast opposite flick overshoots or lingers — your thumb travels an arc, the d-pad transition takes time, and at high speed nerves and contact noise blur the release. Regular DAS triggers too soon on this kind of input and flings the piece back across the board. Tap-Back DAS gives intent a short forgiveness window so the meant-to-be-a-tap reversal wins over the hardware noise, which is exactly why it pays off most in high-precision, high-speed play.
3 physical, independently tunable handling profiles
Keyboard, physical joystick, and physical d-pad, each with their own DAS, ARR, tap-back, and soft-drop settings.
Stick activation + release hysteresis
Hysteresis on stick activation and release eliminates input chatter near center. The result is the feeling that the system understands your intent — pieces move when you mean them to, and don't when you don't.
Per-cardinal stick zone angles
Set the arc the system uses to qualify up, down, left, and right on the joystick. If your thumb doesn't land precisely every time, widen the zones; if you want a tight competitive feel, narrow them. Your leniency, your call.
Rotation-stick hysteresis
For sticks with loose tolerance that snap past center on release — the snap-back doesn't count as a movement in the other direction against you.
Stick debug overlay
Real-time angle, magnitude, and zone readout for calibrating the thing you're holding.
Undo DAS + Undo ARR
Independent DAS and ARR settings for undo scrubbing, distinct from in-game handling.
3 bindings per action, press-or-hold
Up to 3 bindings per action, each distinguishing tap vs hold for dual-action mapping.
Context-scoped conflict detection
Use the same button for different operations depending on the game mode. A binding that means one thing in Survival can mean something else in Finesse without colliding.
Fully configurable bindings
Every binding is configurable — movement, hard drop, rotation, hold, soft drop. Any key for any action, on keyboard and on controller.
DAS pipelining
Advanced piece-placement technique: queue DAS charges across piece transitions to place consecutive pieces without re-holding the direction.
Cold Clear engine choices: pretrained settings for versus, cheese, stacking, C4W towers, and combo continuations.
Pretrained Cold Clear settings
Choose among General VS, Cheese Race, 9-0, 6-3, 2-7, 5-4 stacking, center 4-wide tower stacking, and combo-continuation settings.
Mode-matched bot styles
AI-assisted drills can use the pretrained style that fits the skill target, from versus pressure to opener and combo study.
C4W heuristic set
Center-4-wide training with pretrained handling for residuals, parity, density, and burns.
Piece Preview: Live vs Stable
Cold Clear's natural behavior is to constantly revise the plan as the queue updates with play. Stable freezes that — the queue locks in time so you can play out the full plan end to end without it shifting underneath you. Useful when you want to study a single committed line instead of riding live evaluation.
Bot auto-play 1× – 10×
A vibe thing. Bot plays at any speed from 1× to 10× — pick the one your eye enjoys.
Hacker-type bot mode
Any key input executes the next Cold Clear-optimal move. For muscle-memory drilling against optimal play. (Just for fun.)
Force-Calculate
When you're practicing, it's easy to drop into a flow state and place pieces without ever consulting the queue. That's a mistake. Force-Calculate stops you every N moves so you have to acknowledge what's coming and plan ahead. Think of it like a slow-feeder bowl for a dog — the friction is the point.
Board overlays for spawn axis, rotation axis, bag boundaries, and placement-order projection.
Spawn-axis column indicator
Column-5 indicator marking the piece spawn and rotation axis.
Rotation-axis column highlight
Vertical stripe overlay marking the active piece's rotation pivot column.
Rotation axis on the ghost piece
See the pivot dot before you rotate. Predict the kick.
Rotation axis on the active piece
The pivot moves with the falling piece, live.
Bag separators in the next queue
Visual separators between bags in the next queue for bag-boundary awareness.
Per-mino vs centered projection numbers
Placement-order digits rendered per mino or as a single digit at the piece centroid.
Grid-line prominence multiplier
Grid-line visibility from 0.5× to 4×. Thicker lines make it easier to subitize and count individual cells at a glance — set it to whatever your eye reads best.
Per-role visuals and skins
Pick the skin and visual for each role independently — the active piece, hold, solution projection, garbage, ghost, and on-screen visual references can each carry their own look.
Hide active piece entirely
Hide the active piece entirely, forcing placement prediction from ghost alone.
Hide the hold box
Hold still works; this just hides the box for a cleaner UI, or for modes where you don't want to look at it — visualization, subitizing, anything where the box is noise. The hide is mostly a UI convenience the system uses internally, but it's exposed if you want it.
Next queue size 0 – 6
Next-queue visibility from 0 (blind) to 6 pieces. Cold Clear respects whatever queue size you set, so its decisions stay reasonable to the information you actually see — it doesn't act as a clairvoyant.
Asymmetric generators, queue recall, subitizing, and live Cold Clear puzzle generation.
Infinite Finesse Game-Flow mode
Finesse carvings placed inside Cold Clear-guided realistic board states instead of isolated puzzle boards.
Queue Recall
The game shows you the queue, then asks you to recall it from memory. Trains the brain to treat the queue as an active participant in play, not a dormant strip you glance at.
Subitizing drills
Flat-stack recognition: instantly identify how wide a gap is so the piece combinations that fit drop into mind without counting.
T-Spin Finder + Sandbox
Every puzzle is generated at runtime, so the bank is unlimited and you never see the same one twice.
Height-Difference corpus
More than 2,500 ways to resolve a height difference. Filter by side and by available pieces to target the cases you're weakest at.
Overhang corpus
Every clean 1–5 piece overhang setup — more than 550 of them — so you can build the T-Spin into the stack instead of leaving an ugly residual after taking it.
Opener continuation nodes
Past the first bag, most openers have many bag 2 continuations, and even more bag 3s past those. Set continuation nodes in the visual reference so the whole tree is in front of you, instead of cross-referencing a separate screen the way most people practice.
Opener Quiz mode
Pick an opener you want to learn, turn on Quiz, and the game hands you a queue and asks whether it can produce the opener — checked against the dependency chart you set up (or the one Liemtris ships with). Trains the mental calculation rather than the motor execution.
Mirror mode: original / mirrored / both
Puzzle generator emits original, mirrored, or both variants to double pattern coverage from a single bank.
Forecast mode
Insert garbage and attacking patterns directly under your stack by hand so you can viscerally see what discovered attacks look like before they appear in a real game.
Piece / rotation / column filters
For Infinite Finesse. Stop drilling the placements you already know cold — filter to the piece, column, and rotation (CW, CCW, none, 180) you actually struggle with.
Per-element visual controls: skin presets plus hex-level overrides for minos, grid, board frame, and panels.
Named skins
Named visual presets: Neon Synthwave, Aurora Trace, Classic.
Metal-shader procedural backgrounds
GPU metal-shader procedural backgrounds with configurable animation speed.
Per-piece hex color override
Per-tetromino hex color override for custom or colorblind-safe palettes.
Grid-line tuning
Grid-line color, width, and glow radius.
Mino tuning
Per-mino outline, corner radius, gradient direction, outer glow, hatch and cross-hatch rendering.
Empty-cell opacity independent of filled
Empty-cell opacity configurable independently of filled-cell opacity.
Haptic profile with per-event textures
Per-event haptic texture with configurable intensity and sharpness.
Impact animation toggle
Board shake and tilt animations on wall hits and hard drops. Toggleable for competitive stability.
Board editor, custom queue, and seeded bags for ad-hoc setups and reproducible study.
Fumen-style direct cell drawing
Finger-draw directly on the board to paint garbage blocks or colored region markers. Fumen-compatible.
Colored palette for region marking
Colored block palette for zoning and annotation during study.
Saved puzzle library
Most puzzle modes let you save the ones worth revisiting, so you can come back and drill them on your own time instead of losing them to the next generated round.
Custom queue + hold editor
Tap the hold box to edit the hold piece, or tap the queue to edit the upcoming pieces. Paste an IOTLJSZ string from your clipboard, pick from the on-screen UI, or reuse a recent queue.
Seeded 7-bag for replay reproducibility
Seeded 7-bag randomizer (Fisher-Yates + LCG) produces identical piece sequences across sessions given the same seed.
Queue regeneration
Preserve the board terrain while regenerating the hold and the next queue. The underlying concept behind T-Spin Sandbox — same board, fresh pieces, new puzzle.
Line-clear toggle
Disable line-clear logic for terrain-only puzzle modes, where the terrain itself is the lesson and clearing it would erase the setup.
Per-drill stats of attempts, success rate, placement errors, and best time, tracked by piece × rotation × column.
Per-drill finesse stats
Attempts, successes, placement errors, and best time tracked per piece × rotation × column.
Puzzle attempts with daily breakdown
Puzzle attempt counts with daily aggregation and rolling-baseline comparison.
Sortable finesse stats UI
Sortable and filterable stats by piece, rotation, and success rate.
Weakest-drill detection
Next-drill selection: 50% weakest-item priority, 50% random.
Per-session vs lifetime stats
Session-scoped and lifetime-scoped statistics available as separate views.
Per-mode visual choices and Cold Clear preset selections.
Every visual + bot setting is per-mode
Each mode carries its own settings, and it matters. Survival wants the Cold Clear chewing-through-cheese preset; Sprint training wants the 6-3 quad-only preset. Finesse modes want Hold off so you place pieces in order instead of thinking about swaps, and want ghost off because the ghost gives the solution away. Switching modes loads that mode's profile automatically.
Puzzle solution preview persistence
Puzzle hint-state persistence configurable: sticky across puzzles or reset per puzzle.
Hold disableable per mode
Hold slot disableable per mode to force commitment play.
Hints toggle on puzzle modes
Per-mode toggle for hint overlays on all puzzle modes.
Custom game-mode ordering
Drag-reorder the mode grid.
Active-piece side-panel slot
Active piece rendering relocatable to a side-panel HUD tile.
JSON export/import of all settings, iCloud Drive sync, file-level visibility.
Settings export to JSON
Schema-versioned JSON export of all settings: bindings, openers, Cold Clear preset choices, and per-mode profiles.
Settings import from JSON
JSON settings import with one-tap apply.
iCloud Drive sync, visible in Files
Settings file stored in iCloud Drive, visible in the Files app for direct inspection, diff, and backup.
Reset to defaults
Restore all settings to defaults.
Guideline Fidelity
Popular Stacker Client Systems
Normal, T-Spin, and Infinite Finesse are free forever. Ready for more?
3 free modes included. Unlock all 17 with a free trial.
~$1.14/day
+ applicable tax
Start 3-Day Free TrialThen $7.99/week after trial. Cancel anytime.
~$0.83/day
+ applicable tax
Start 3-Day Free TrialThen $24.99/month after trial. Cancel anytime.
$1.92/week
+ applicable tax
Start 3-Day Free TrialThen $99.99/year after trial. Cancel anytime.
Controller Compatibility
Configurable DAS/ARR, frame-perfect timing. Tested across first-party consoles, telescopic clip-ons, and the third-party gamepads competitive stackers actually use.
PlayStation
Xbox
Nintendo
Popular picks
Console — Extended
Cloud Gaming
Telescopic — Lightning
Telescopic — USB-C
Telescopic & Standalone — Bluetooth
8BitDo — Other
GuliKit
Flydigi
BIGBIG WON
Machenike
Scuf
PowerA
Other
These models can't pair as standard iOS controllers — usually because they use proprietary wireless protocols, lack Bluetooth entirely, or are wired-only.
Console
Telescopic
Your subscription auto-renews at the plan you chose. Cancel anytime before day 4 and you won't be charged.
Yes. Cancel from your App Store settings or Stripe dashboard. No cancellation fees, no questions asked.
Free: Normal, T-Spin Finder, and Infinite Finesse forever, with no account needed. Paid: All 17 modes plus Cold Clear features, undo keybind, and iCloud sync.
Yes. Subscribe here with Sign in with Apple, then sign in with the same Apple ID in the app to unlock your subscription.
17 training modes. Cold Clear-backed analysis. Frame-precise controls. 3-day free trial; cancel anytime before day 4.