Sensor Size > Megapixels

A larger sensor with bigger pixels (like the HQ Camera’s 1/2.3″ IMX477) gathers more light and beats higher-megapixel sensors in low light. Chase sensor quality first, resolution second.

Autofocus vs Fixed vs Lenses

Autofocus (Camera Module 3, IMX519) handles varying distances automatically. Fixed-focus is fine for a camera that never moves. C/CS or M12 lens mounts give you full manual control — telephoto, macro, wide.

Rolling vs Global Shutter

Rolling-shutter sensors distort fast-moving objects (propellers turn into bananas). For robotics, drones and machine vision, a global shutter camera captures the whole frame at once.

Cable & Board Compatibility

Full-size Pi 4/3 boards use a 15-pin CSI connector; Pi 5 and Zero use a smaller 22-pin one. Check which FFC cables ship in the box before ordering — an adapter cable costs a few dollars but stalls your weekend project.

Special Abilities

No IR filter (NoIR) enables infrared night vision with an IR light. HDR mode tames harsh scenes. The AI Camera runs neural networks on the sensor itself, leaving your Pi’s CPU free.

For real photography on a Pi, nothing touches the HQ Camera platform. The 12.3MP Sony IMX477 sits behind a proper C/CS lens mount, so you can bolt on anything from a 6mm CCTV lens to a telephoto, a microscope adapter or a telescope — with a back-focus ring and tripod mount built in. Arducam’s version uses the same IMX477 sensor and libcamera tuning as the official HQ module and includes the C-CS adapter, at around $53. Remember: lenses are sold separately, so budget another $15–50 for glass.

The Arducam 16MP IMX519 gives you more resolution than the official Camera Module 3 for less money — 4656×3496 stills, fast PDAF+CDAF autofocus, and native libcamera/Picamera2 support using the same tuning framework as official modules. This kit even includes an ABS case and cables. Video tops out at 1080p30, so it’s a stills-first choice, but for document scanning, product photography and detailed inspection shots it’s the best value on this list at about $30.

The Camera Module 3 NoIR is the same excellent 12MP autofocus module as our top pick, minus the infrared filter. Flood a scene with 850nm IR light and it sees clearly in total darkness — perfect for wildlife boxes, baby monitors, security cams and plant-growth (NDVI) experiments. Daytime colors shift toward pink/washed-out without the filter, so treat it as a dedicated night/IR camera rather than a general-purpose one. This listing had no single Buy Box price when we checked — compare seller offers on the listing before ordering.

Rolling shutters smear anything that moves fast. The IMX296 global shutter sensor in this Arducam module exposes every pixel simultaneously, freezing propellers, conveyor parts and barcodes mid-flight at up to 60fps. The 1.58MP resolution is a feature, not a bug — machine-vision pipelines want small, fast frames they can process in real time. An M12 lens is included and swappable, and the sensor matches the one in the official Raspberry Pi Global Shutter Camera, so standard libcamera workflows apply.

The 64MP Hawkeye captures 9152×6944 stills — flagship-phone territory — with autofocus and up to 10× lossless digital crop. It’s spectacular for document archiving, PCB inspection and any project where you zoom into detail after the fact. The trade-offs are real: full-res captures take seconds and want a Pi 4/5 with plenty of RAM, video is limited far below the headline resolution, and the tiny pixels need good light. Power users only — but nothing else this cheap resolves this much detail.

For about $12, the Freenove 5MP gets a camera onto your Pi with zero friction: the classic OV5647 sensor (same family as the original official Camera V1), an adjustable desk holder, a screwdriver, and — crucially — both 15-pin and 22-pin cables in the box, so it works with a Pi 5 or Zero 2 W out of the bag. Image quality is dated and focus is fixed, but for learning Picamera2, classroom sets, 3D-printer monitoring and simple time-lapses, it’s all you need.

General projects & video?

Get the Camera Module 3 — autofocus, HDR and flawless official support cover 90% of projects.

Object detection / edge AI?

The AI Camera runs neural networks on the sensor itself — real-time vision even on a Pi Zero 2 W.

Astro, macro or microscopy?

The HQ Camera (IMX477) with a C/CS lens of your choice delivers the best raw image quality here.

Wildlife or night security?

The Camera Module 3 NoIR plus an 850nm IR illuminator sees clearly in total darkness.

Robots, drones, fast motion?

The Global Shutter (IMX296) camera freezes motion without distortion — built for machine vision.

Learning or classroom sets?

The Freenove 5MP is ~$12 with both cables and a holder included — perfect to start with.

Which camera works with the Raspberry Pi 5?

All eight cameras here work with the Pi 5 — the catch is the cable. The Pi 5 (and Zero/Zero 2 W) uses a smaller 22-pin FFC connector, while older full-size Pis use the 15-pin type. Official Camera Module 3 boxes include only the 15-pin cable, so order a 22-pin “mini” camera cable alongside it. Freenove includes both cables, and most Arducam kits include a 15–22-pin cable — check the “what’s in the box” section of each listing.

Camera Module 3 vs HQ Camera — which should I buy?

Camera Module 3 for convenience, HQ for quality. The CM3 is smaller, cheaper, autofocuses itself and shoots better video (2304×1296p56 with HDR). The HQ Camera’s larger 1/2.3″ sensor plus real C/CS lenses produces clearly better stills — especially in low light or at unusual focal lengths — but you must buy a lens and focus manually. If you’re not sure you need interchangeable lenses, you don’t: buy the CM3.

Can a NoIR camera see in the dark by itself?

No. “NoIR” only means the infrared-blocking filter has been removed — the sensor still needs light, it just accepts infrared light your eyes can’t see. For night vision you must add an IR source (an 850nm LED illuminator board or IR floodlight). With one, the NoIR Camera Module 3 produces bright, detailed monochrome-looking images in total darkness; without one, it sees nothing at night.

Is a CSI camera better than a USB webcam for the Pi?

For almost every Pi project, yes. CSI cameras connect straight to the Pi’s image signal processor, giving lower latency, lower CPU usage, raw sensor access and full control from Picamera2/rpicam — things USB webcams can’t match. A webcam only wins if you need a long cable run (CSI ribbons are short) or plug-and-play use across different computers.

Do I need the AI Camera, or can I run AI with a normal camera?

You can run models like MobileNet or YOLO on any camera using the Pi’s CPU — but on a Pi 4 that often means single-digit frame rates and a hot CPU with nothing left for your application. The AI Camera’s IMX500 runs the network inside the sensor and streams results as metadata, so detection runs in real time even on a Pi Zero 2 W. If AI is the point of your project, it pays for itself; if you just want occasional snapshots analyzed, a normal camera is fine.