ESPRESSIF · ESP-VISION OPEN-SOURCE FRAMEWORK

Real-time vision AI.
In a few lines of Python.
On ESP32.

Capture, processing, inference, and display — all on the device.

Flash It Now Open Web IDE ↗

01LOW-CODE

Less code. More power.

Camera, image processing, on-device inference, and hardware peripherals are all accessible through Python APIs.

OUTPUTESP32-P4

Per-frame inference with a quantized ESP-DL modelVision AI

main.py17 lines

import espdl
import sensor
import time

sensor.reset()
sensor.set_pixformat(sensor.RGB565)
sensor.set_framesize(sensor.QVGA)
sensor.skip_frames(time=1000)

det = espdl.ESPDet("/sdcard/hand_det.espdl", score=0.5, nms=0.7)
while True:
    img = sensor.snapshot()
    for x, y, w, h, score, category in det.detect(img):
        img.draw_rectangle(x, y, w, h, color=(255, 0, 0), thickness=2)
        img.draw_string(x, max(0, y - 12), "%.2f:%d" % (score, category))
    img.flush()
    time.sleep_ms(20)

02CAPABILITIES

Complexity. Handled.

Capture, processing, inference, and control all run on the device.

API01

Low-Code Python API

Unified sensor · image · display · espdl APIs — real-time results in a few lines of Python. Flash online, connect the Web IDE, no toolchain to set up.

AI02

On-Device ESP-DL Inference

Object detection, pose estimation, and image classification; load a quantized .espdl model in one line — real-time, offline, on device — or bring your own PyTorch / TensorFlow model.

VLM03

Cloud Vision LLMs

Stream straight to OpenAI-compatible vision APIs and tap multimodal models like GPT-4o — complex scene understanding with no local compute.

IMG04

Image Processing

Drawing, filtering, color tracking, feature detection, QR codes, barcodes, and AprilTags.

CDC05

Hardware Codec

H.264 / MJPEG / RTSP and USB CDC live preview, saturating the on-chip multimedia accelerators.

IO06

Rich Peripheral Support

Cameras, displays, SPI, I2C, UART, SD cards and more work out of the box, with a MicroPython machine-compatible API.

03BROWSER-BASED

Zero setup. Just go.

No toolchain installation is required; flashing, writing, and running all happen in the browser.

BROWSER

One-Click Flashing

Connect your board straight from the browser and flash the official firmware.

Online Flashing · Web Serial○ No device connected

Loading firmware manifest…

IDE / VS CODE

Write & Run

Write scripts in the Web IDE or VS Code and watch capture and inference results in real time.

Open Web IDE Download VS Code Extension

04CONNECT AI

Describe it. AI builds it.

Connect the ESP-VISION MCP to an AI assistant to build edge vision applications through conversation.

Server URL

https://mcp.vision.espressif.com

Auto Install

Click the button below to add this server to Cursor automatically.

Add to Cursor →

Manual config (mcp.json)

{
  "mcpServers": {
    "esp-vision-mcp": {
      "url": "https://mcp.vision.espressif.com"
    }
  }
}

Auto Install

Click the button below to add this server to VS Code automatically.

Install to VS Code →

VS Code config

{
  "mcp": {
    "servers": {
      "esp-vision-mcp": {
        "type": "http",
        "url": "https://mcp.vision.espressif.com"
      }
    }
  }
}

05MODELS

Any model. Just one line.

ESP-DL and TFLite Micro models are loaded from device storage in a single line of code.

ESP-DL7 models

Model	Task	Input	Dataset	Size
ESPDet Pico Cat ESPDet Pico · `espdl.ESPDet` Detects cats in camera images. cat	Object Detection	224×224 RGB565	Cat	487 KB
ESPDet Pico Cat & Dog ESPDet Pico · `espdl.ESPDet` Detects cats and dogs in camera images. catdog	Object Detection	224×224 RGB565	Cat & Dog↗	561 KB
ESPDet Pico Dog ESPDet Pico · `espdl.ESPDet` Detects dogs in camera images. dog	Object Detection	224×224 RGB565	Dog	486 KB
ESPDet Pico Face ESPDet Pico · `espdl.ESPDet` Detects human faces in camera images. face	Object Detection	224×224 RGB565	Face	484 KB
ESPDet Pico Hand ESPDet Pico · `espdl.ESPDet` Detects human hands in camera images. hand	Object Detection	224×224 RGB565	Hand	486 KB
YOLO11n COCO YOLO11n · `espdl.YOLO11` Detects the 80 COCO object classes in camera images.	Object Detection	160×160 RGB565	COCO	2.7 MB
YOLO11n-Pose COCO YOLO11n-Pose · `espdl.YOLO11nPose` Detects people and estimates 17 COCO body keypoints (human pose) in camera images. person	Pose Estimation	160×160 RGB565	COCO	3.0 MB

TFLite Micro2 models

Model	Task	Input	Dataset	Size
Person Detection MobileNet · `tflite.Model` TensorFlow Lite Micro person-detection model: classifies whether a person is present in a 96x96 grayscale camera frame. no personperson	Image Classification	96×96 GRAYSCALE	Visual Wake Words	294 KB
Sine MLP · `tflite.Model` TensorFlow Lite Micro "hello world" model: approximates sin(x) for x in [0, 2*pi] from a single scalar input.	Regression	1 FLOAT32	Synthetic	2 KB

Add a Custom Model ↗

06BOARDS

One framework. Every board.

Camera, display, and storage work out of the box across the ESP32-S31, ESP32-P4, and ESP32-S3 series.

Board	Chip	ESP-VISION Support
ESP32-P4X-EYE↗	ESP32-P4	Supported sensor · image · display · espdl · tflite · imageio · h264 · rtsp · barcode
ESP32-P4X-Function-EV-Board↗	ESP32-P4	Supported sensor · image · display · espdl · tflite · imageio · h264 · rtsp · barcode
ESP32-S3-EYE↗	ESP32-S3	Supported sensor · image · display · espdl · tflite · imageio
ESP32-S31-Korvo↗	ESP32-S31	SupportedESP-IDF master only sensor · image · display · espdl · tflite · imageio