STM32Cube AI Studio Documentation ¶
STM32Cube AI Studio is a free-of-charge GUI that converts pretrained neural network and classical ML models into optimized C code for STM32 microcontrollers. It combines model import, optimization, on-host/on-target validation, and code generation in a single tool.
On targets with the Neural-ART Accelerator NPU (e.g. STM32N6), the generated code maps operations to the NPU at operator level, falling back to the CPU only when necessary, thus maximizing hardware acceleration.
Important
STM32Cube AI Studio replaces the X-CUBE-AI plugin for STM32CubeMX, which is no longer supported from ST Edge AI Core v3.0.0 onwards. Use STM32Cube AI Studio for all new developments.
Where to start ¶
Download and install STM32Cube AI Studio, then check the prerequisites to get your environment ready.
Step-by-step first project: import a model, optimize it, and generate C code for your STM32 target.
End-to-end tutorial to deploy an AI model on the STM32N6 Neural-ART NPU and run hardware-accelerated inference.
Build and run your first AI application on mainstream STM32 (Cortex-M) MCUs, from model to on-board inference.
Latest changes, new features, fixed issues, and version history of STM32Cube AI Studio.
The Software License Agreement governing the use of STM32Cube AI Studio.
Note
The deploy and validate tutorials cover the end-to-end flow and also rely on ST Edge AI Core, so they reach a bit beyond STM32Cube AI Studio alone.
Key Features ¶
Model conversion & optimization — generates STM32-optimized C libraries from Keras, TFLite, ONNX, and scikit-learn (via ONNX) models; supports float32 and quantized formats
On-host & on-target validation — compares optimized model outputs against the original, with performance metrics and error diagnostics
Advanced memory management — external RAM/Flash, memory pools, weight compression
Neural-ART NPU support — hardware-accelerated inference on supported STM32 targets
Cross-STM32 portability — easy retargeting across STM32 families via the STM32Cube ecosystem
Project generation — produces ready-to-build IDE projects (STM32CubeIDE, IAR, Keil)
ST Edge AI Ecosystem ¶
STM32Cube AI Studio orchestrates several ST tools under the hood:
It calls ST Edge AI Core (docs) to analyze, optimize, validate, and generate C code from AI models.
It calls STM32CubeMX to configure the MCU, set up peripherals, and create IDE projects (STM32CubeIDE, Keil, IAR).
It calls STM32CubeProgrammer to flash validation firmware onto the target board.
The following tools are independent but complementary:
ST Edge AI Developer Cloud — remote benchmarking on the ST board farm.
STM32 Model Zoo and Model Zoo Services — pre-trained STM32-optimized models, plus Python scripts for training, fine-tuning, evaluation, quantization, and benchmarking before deployment.
For MEMS or STELLAR targets, use MEMS-Studio or StellarStudio, or the ST Edge AI Core CLI directly.
Migration from X-CUBE-AI ¶
Migration is optional — existing X-CUBE-AI projects continue to work. However, ST Edge AI Core v3.0.0+ requires STM32Cube AI Studio.
To migrate: create a new project in STM32Cube AI Studio, import your model, configure the target, and generate code. See FAQ and Troubleshooting for details.
Compatibility ¶
MCUs: all STM32 families supported by STM32CubeMX, except STM32F1/F2, STM32L1, STM32U0, STM32MP1/MP2.
IDEs: STM32CubeIDE, IAR, Keil.
Model formats: Keras (.h5, .keras), TFLite (.tflite), ONNX (.onnx).