ESP32 or ESP32-S3

The original ESP32-CAM is cheapest and has the most tutorials. ESP32-S3 boards add native USB, more speed for JPEG handling and vector instructions that make ESP-WHO face detection and other on-device vision much faster. For anything beyond basic streaming, the S3 is worth it.

Camera Sensor

The OV2640 (2 MP) is the classic, best supported option. The OV3660 (3 MP) gives better detail and color. The OV5640 (5 MP) adds autofocus on some modules but pushes the ESP32 hard. Lens field of view (typically 66° stock, up to 160° wide-angle) matters as much as megapixels.

PSRAM

The esp32-camera driver keeps frame buffers in PSRAM. 4 MB is enough for SVGA streaming; 8 MB lets you use larger frames, double buffering and AI models. Remember to enable PSRAM in your build settings or the camera will fail to initialize.

Programming Method

The classic ESP32-CAM needs either an ESP32-CAM-MB programmer base or a 3.3 V USB-to-TTL adapter and a jumper on GPIO0 to flash. Full-size and S3 boards with onboard USB avoid that entirely, which saves a lot of time during development.

Antenna and Power

Video needs a strong Wi-Fi link, and many ESP32-CAM boards include a U.FL connector for an external antenna. Pair any camera board with a stable 5 V, 1 A supply; brownout resets during streaming are almost always a power problem, not a firmware bug.

For someone learning electronics and cameras at the same time, Freenove’s ESP32 Camera Basic Starter Kit is the best package (about 4.6 stars from 210+ ratings). The heart of it is Freenove’s ESP32-WROVER camera board with 4 MB flash and 8 MB PSRAM, a USB code uploader, camera and microSD slot, so it plugs straight into USB like a normal dev board. Around it you get LEDs, buttons, sensors and wires, plus C and MicroPython tutorials that start with blinking an LED and build up to camera web servers. It uses the original ESP32 rather than the S3, so on-device AI is slower, but it covers the classic ESP32-CAM project list without any programmer hassle.

The XIAO ESP32S3 Sense is the smallest serious camera board you can buy (about 4.3 stars from 360+ ratings). An ESP32-S3R8 with 8 MB PSRAM sits on a 21 × 17.8 mm board with LiPo charging and a U.FL antenna connector, and the Sense expansion adds a detachable camera, a digital microphone and a microSD slot. The current Amazon listing describes an OV3660 sensor compatible with OV5640, while older stock shipped with the OV2640. It’s ideal for wearable cameras, doorbells, wildlife traps and Edge Impulse or ESP-WHO vision models, and Seeed’s wiki has ready-made camera and TinyML examples. Few GPIOs remain free, it runs warm while streaming, and only one was left in stock when we checked.

NULLLAB’s ESP32-S3 CAM Dev Kit keeps the compact ESP32-CAM idea but fixes its two biggest flaws: it has USB-C built in for flashing and power, so no programmer base or FTDI adapter is needed, and it moves to the ESP32-S3 with 8 MB PSRAM and 8 MB flash for smoother video and faster face and object recognition. It ships with an OV3660 3 MP sensor, and NULLLAB publishes pinouts and examples on GitHub. Early ratings are excellent but few (about 4.8 stars from 9 reviews), so it’s a newer, less proven option, but it’s the board we’d pick to replace a pile of classic ESP32-CAMs in a new design.

Streaming video eats Wi-Fi bandwidth, and a weak link shows up as stutter and dropped frames long before it breaks a sensor connection. Aideepen’s ESP32-S3-CAM ships with an external antenna in the box alongside an OV3660 3 MP camera on an ESP32-S3 dual-core LX7, making it the camera board to choose for garages, gardens, bird feeders or anywhere far from your router. The ESP32-S3 handles JPEG frames and vision models faster than the original ESP32. Reviews are positive but few (about 4.4 stars from 10 ratings), and only one was left in stock when we checked, so check availability.

Once you have a programmer, extra ESP32-CAM boards are cheap, and Hosyond’s 2-pack is a solid source of spares for multi-camera setups (about 4.0 stars from 110+ ratings). These are standard AI-Thinker style boards with the ESP32, a camera, a microSD slot and the flash LED, and the listing states 8 MB of PSRAM, more than the 4 MB common on classic ESP32-CAMs. Check whether the version you receive includes a programmer base; if not, pair it with the FTDI adapter below or the Aideepen kit’s CAM-MB. As with all budget ESP32-CAMs, power them from a good 5 V supply.

HiLetgo’s ESP32-CAM 2-pack keeps the classic AI-Thinker form factor but upgrades the sensor to an OV3660 3 MP camera, which gives sharper images and better color than the stock OV2640, and the listing includes a USB-C to serial CH340 base for programming. That makes it a good fit if you want to follow classic ESP32-CAM tutorials with better image quality. The esp32-camera driver supports the OV3660, but some older sketches assume an OV2640 and need minor changes. Ratings are mixed (about 3.9 stars from 160 reviews) and stock was down to one when we checked.

If you want the ESP32-S3 camera board and a complete electronics course in one box, Freenove’s ESP32-S3 Ultimate Starter Kit is the best-rated option (about 4.6 stars from 140 ratings). It includes the same ESP32-S3 camera board as our ESP32-S3 upgrade pick, with 8 MB flash, 8 MB PSRAM, a USB code uploader, native USB-OTG, the camera and a microSD slot with a 1 GB card. Around it, Freenove adds 243 components, modules and sensors and an 828-page tutorial with 121 projects in MicroPython, C and Processing, starting with basic I/O and building to camera web servers. It costs more than the board alone, so skip it if you already own a parts kit.

For multi-camera setups such as a home security system, a greenhouse or several 3D printers, FORIOT’s ESP32-S3-CAM 3-pack gives you three modern camera boards at a sensible per-board cost (about 4.1 stars from 44 ratings). Each board uses an ESP32-S3-WROOM N16R8 module, so you get 16 MB of flash and 8 MB of PSRAM, the most memory in this guide, plus an OV3660 camera and two USB-C ports: one for USB-OTG and one for TTL serial programming and logs, so no programmer base is needed. The listing calls the OV3660 a 2-megapixel sensor, while the OV3660 is normally rated at 3 MP, so expect specs to vary slightly by batch. Test all three boards on arrival.

The stock ESP32-CAM lens has a fairly narrow view, which is fine for a doorbell but poor for a room, a 3D printer bed or a driveway. These OV2640 modules with a 160° wide-angle lens swap straight onto the ESP32-CAM’s camera connector and capture a much wider scene with the same driver and code. The 2-pack also gives you a spare for when the fragile ribbon cable inevitably gets creased. Expect some fish-eye distortion at the edges. Ratings are decent (about 4.1 stars from 66 reviews), and stock was low when we checked.

CAM-MB base boards are convenient, but they occasionally fail or don’t fit boards from other sellers. A good USB-to-TTL adapter flashes any ESP32-CAM and doubles as a serial console for every other project. DSD TECH’s SH-U09C2 uses a genuine FTDI FT232RL (not a clone), switches between 5 V, 3.3 V and 1.8 V logic with a jumper, and comes in a clear protective case, and it has about 4.8 stars from 930+ ratings. For the ESP32-CAM, set it to 3.3 V logic, connect TX to U0R and RX to U0T, power the camera from 5 V, and tie GPIO0 to GND while you flash.

Following classic ESP32-CAM tutorials?

Buy the Aideepen ESP32-CAM + MB 2-pack, and keep the FTDI adapter as a backup programmer.

Face detection or vision AI?

Go ESP32-S3: the Freenove S3 camera board for development, the XIAO ESP32S3 Sense for tiny builds.

Complete beginner?

The Freenove ESP32 Camera Basic Starter Kit for the classic ESP32, or the Freenove ESP32-S3 Ultimate Starter Kit for 121 projects on the S3.

Camera far from the router?

The Aideepen ESP32-S3-CAM, which ships with an external antenna in the box.

Tired of programmer boards?

The NULLLAB ESP32-S3 CAM has USB-C built in at ESP32-CAM size.

Need several cameras?

The FORIOT ESP32-S3-CAM 3-pack gives you three N16R8 camera boards with dual USB-C.

Monitoring a room or 3D printer?

Swap in the 160° OV2640 lens for a much wider field of view.

Why does my ESP32-CAM keep rebooting with a brownout error?

Almost always power. The camera, Wi-Fi radio and flash LED draw current peaks the board’s regulator can’t handle from a weak source. Use a 5 V supply rated for at least 500 mA (1 A is safer), keep the USB cable short and thick, and add a 470 to 1000 µF capacitor across 5 V and GND if needed. Avoid disabling the brownout detector as a fix; it only hides the real problem.

How do I program an ESP32-CAM without a CAM-MB board?

Use a 3.3 V USB-to-TTL adapter such as the FTDI pick above. Connect adapter TX to U0R, RX to U0T, GND to GND, and power the ESP32-CAM from 5 V. Tie GPIO0 to GND, press reset, upload, then remove the GPIO0 jumper and reset again to run your code.

What resolution and frame rate can I expect?

For live MJPEG streaming over Wi-Fi, VGA (640 × 480) or SVGA (800 × 600) at reasonable frame rates is realistic on a good connection. Higher resolutions like UXGA work for still photos but stream slowly. ESP32-S3 boards with 8 MB PSRAM and a strong antenna get noticeably better results. Our ESP32-CAM ESP-IDF live streaming tutorial shows how to set frame size and JPEG quality.

Which microSD cards work with the ESP32-CAM?

Use SDHC cards (4 to 32 GB) formatted FAT32. Many mount failures come from large exFAT cards. Note that the SD card shares pins with the flash LED (GPIO4), so in 4-bit SD mode the LED may glow while writing; 1-bit mode avoids that. See our take photos and save to SD card tutorial.

Should I buy an ESP32-CAM or an ESP32-S3 camera board?

Buy the classic ESP32-CAM if cost matters most and you’re following existing tutorials. Choose an ESP32-S3 board for native USB, faster processing, better face detection with ESP-WHO, and room for AI models. Our ESP32-S3 buying guide covers more S3 boards.