Record your gameplay. Train a neural network. AFK while the bot farms, fights, and loots for you 24/7. Battle-tested on Genshin, WoW, FFXIV, Lost Ark, GW2, and New World.
This isn't your average auto-clicker. This is a full neural network that watches you play, learns your strats, and executes them at 30+ FPS while you sleep.
Just play your game for 10-15 min. The AI watches every input you make — keyboard combos, controller sticks, mouse flicks — and records it all at 10+ FPS. Your gameplay is the training data.
Hit one command and grab a coffee. PyTorch trains a deep neural network on your recorded plays. It auto-detects your action space, picks the right model, and tunes everything to your game's profile.
Launch the bot and go AFK. It reads the screen, predicts your next move, and executes at 30+ FPS. Built-in anti-stuck tech means no more running into walls. Farm all night, wake up to full bags.
WASD, PS4/Xbox controller (sticks, triggers, bumpers, d-pad), and full mouse (aim, delta, velocity, LMB/RMB/MMB, scroll). If you can press it, the AI can learn it and reproduce it.
A dark-themed gaming dashboard with one-click Record, Train, and Play. Looks like a Steam overlay, works like a pro tool. Game presets, model catalog, real-time logs. No terminal needed.
401 automated tests. Every game profile, every neural net, every action space — tested end-to-end with real PyTorch training. This isn't a prototype. It's production-grade.
Each title ships with a hand-tuned game profile — resolution, action space, neural net, learning rate — all dialed in so you can jump straight into training.
Three stages: Capture your plays, train the AI on your data, then unleash it in-game. Mouse support is optional — toggle it on if your game needs it.
W, S, A, D, WA, WD, SA, SD, NOKEY. One-hot directional encoding for 8-way movement.
LT, RT, Lx, Ly, Rx, Ry, D-Pad (4), Face buttons (A/B/X/Y), Bumpers, Start/Select, Sticks.
X, Y (position), dX, dY (delta), vX, vY (velocity), LMB, RMB, MMB (buttons), scroll.
From lightweight speed-runners to heavy-hitting transformer tanks — 9 architectures tested end-to-end across all 6 games. Every single one passed quality gates.
| Model | Params | Temporal | Speed | Best For | Tier |
|---|---|---|---|---|---|
| EfficientNet-LSTM | ~5M | LSTM (4 frames) | ~8ms | Best overall accuracy | Recommended |
| EfficientNet-Simple | ~5M | None | ~5ms | Single-frame prediction | Recommended |
| MobileNetV3 | ~2M | None | ~3ms | Low-end / real-time | Recommended |
| ResNet18-LSTM | ~12M | LSTM (4 frames) | ~10ms | Speed / accuracy balance | Recommended |
| EfficientNet-Transformer | ~12M | Transformer | ~15ms | Long action sequences | Advanced |
| Multi-Head Action | ~6M | Multi-head | ~8ms | Simultaneous actions | Advanced |
| Game Attention Net | ~6M | Spatial attn | ~10ms | Complex UIs | Advanced |
| InceptionV3 | ~7M | None | ~12ms | Legacy support | Legacy |
| AlexNet | ~60M | None | ~5ms | Legacy support | Legacy |
Recommendation
Use EfficientNet-LSTM for all MMORPGs. It delivers the best accuracy with temporal awareness across all 6 tested game profiles while maintaining >30 FPS inference on GPU. Use MobileNetV3 only on low-end hardware (<4GB VRAM).
No PhD required. Just follow these steps and your AI will be farming while you sleep. Windows 10/11 for gameplay; training works on any OS.
You need Python 3.10+, Git, and optionally Rust (for the Tauri desktop app).
# Check Python version python --version # Should be 3.10+ # Install Rust (optional, for Tauri desktop app) curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
Clone the repository and install all dependencies with a single command.
git clone https://github.com/ruslanmv/BOT-MMORPG-AI.git cd BOT-MMORPG-AI make install # Core + PyTorch make install-launcher # Desktop launcher (optional) make install-tauri # Tauri desktop app (optional)
Play your game normally for 10-15 minutes. The bot watches and records screen + inputs.
# Via command line make collect-data # Or use the Launcher / Tauri app GUI # Controls: T = pause/resume, Q = stop and save
One command trains the model. Auto-detects action space and hardware tier.
make train-model # GPU: ~30-60 minutes | CPU: ~2-4 hours # Default: efficientnet_lstm (recommended for all games)
Start the bot, switch to your game window, and watch the AI play using your learned strategy.
make test-model # Controls: T = pause/resume, ESC = stop completely # Built-in stuck detection + evasive maneuvers
Run the full test suite to confirm all 401 tests pass on your system.
make test # Expected: 401 passed, 1 skipped # Covers: profiles, networks, mouse, training, inference
These are the optimization tips the top players use. Follow them and your bot will perform like a speedrunner, not a noob.
Play consistently for 10-15 minutes. Repeat the same route 2-3 times. Set in-game time to noon for consistent lighting. Avoid AFK or idle moments during recording.
Lower capture resolution = faster training and inference. 480x270 captures all essential UI elements for most games. Only use 640x360 for complex UIs (WoW addons, FFXIV hotbars).
It is the proven default for all 6 tested MMORPGs. Only switch to MobileNetV3 on low-end hardware (<4GB VRAM) or if you need maximum FPS. Advanced architectures are experimental.
Select your game in the launcher or Tauri app. Presets auto-configure resolution, action space, architecture, learning rate, and temporal frames for optimal results.
Mouse recording is a beta feature, disabled by default. Enable it in the UI checkbox only if your game relies heavily on mouse aim (action combat, camera control). It is purely additive and non-destructive.
Low (<4GB): MobileNetV3, batch 8, 160x160. Medium (4-8GB): EfficientNet-Simple, batch 16, 224x224. High (>8GB): EfficientNet-LSTM, batch 32, 224x224 with 4 temporal frames.