Most “best Raspberry Pi camera” lists rank modules by megapixels, which is the one number that matters least once the video leaves the board. Frigate shrinks every frame to a small detect resolution. OctoPrint and Klipper stream MJPEG at 1080p or below, and a wildlife trap lives or dies on its lens and its night image. So the right camera depends on where the stream goes next. For most people that still means the cheapest official module, the Camera Module 3, in one of its four versions.
We compare the Camera Module 3, the High Quality Camera and the AI Camera against two Arducam boards here. We also explain the part most buyers miss: a ribbon camera is not an RTSP camera, so it reaches Frigate through go2rtc. If you already run Frigate on PoE cameras, the PoE camera picks for Frigate cover the wired side. The specs, prices and streaming behavior described here were checked against Raspberry Pi’s product briefs, Arducam’s docs, the Frigate 0.18 docs and live listings on November 1, 2026.
Which Raspberry Pi camera to buy: the short answer
Buy the Camera Module 3 unless you have a reason not to. It is the only official module with phase-detection autofocus, and it costs from $25 at Raspberry Pi resellers. It also works with the stock rpicam software on every Pi with a camera connector, so you install no extra packages. The other awards solve narrower problems: a lens you can swap, inference on the sensor, or more resolution for stills. Prices move weekly, so treat each band as a reference and check the live price before you order.
| Award | Camera | Why it wins | Price band (Nov 1) |
|---|---|---|---|
| Best overall | Raspberry Pi Camera Module 3 | Autofocus, HDR, stock software, every Pi | ~$25 to $30 at resellers |
| Best on Amazon right now | Camera Module 3, Arducam bundle | The official module with an Arducam case and a 15-22 pin cable | ~$46 |
| Best for night with IR light | Raspberry Pi Camera Module 3 NoIR | No IR cut filter, so an IR illuminator lights the scene | ~$27 to $30 at resellers |
| Best for wildlife and lenses | Raspberry Pi High Quality Camera | C/CS mount, bigger 1/2.3″ sensor, telephoto option | ~$55 at resellers, ~$100 on Amazon |
| Best on-camera AI | Raspberry Pi AI Camera | Runs the detection model on the Sony IMX500 sensor | ~$75 at resellers, ~$113 on Amazon |
| Best budget autofocus | Arducam IMX519 16MP | 16MP autofocus for about the price of a Module 3 | ~$25 |
| Most resolution | Arducam 64MP Hawkeye | 64MP stills and time-lapse with autofocus | ~$60 |
Two of those rows carry a sourcing note, so read it before you click. On November 1 the official Camera Module 3 listings on Amazon (standard and Wide) showed out of stock with no seller. That is why our overall pick links to PiShop, a Raspberry Pi approved reseller, and why the Amazon pick is Arducam’s bundle of the same official module. The HQ Camera and the AI Camera are in stock on Amazon, but only from third-party sellers at close to double the reseller price. Buy those two from a reseller when one has stock.
How we picked these Pi cameras
Every spec in this guide comes from the maker’s product brief or documentation. Every price and stock status comes from a live listing at a Raspberry Pi reseller and on Amazon, read on November 1, 2026. We took the streaming claims from the Frigate restream docs, the go2rtc project README and the official Raspberry Pi camera software docs. The question we cared about most is whether a camera works with the software you will actually run.
Software support cut the list hard. A camera that needs a vendor build of libcamera can break when Raspberry Pi OS updates its own packages, so we weighted mainline support above resolution. We left out USB webcams, which Frigate already handles through its ffmpeg device source. We also left out the Global Shutter Camera, a specialist part for fast motion that had no official Amazon listing and no reseller stock when we looked.
How the cameras compare on paper
Read the sensor size and pixel size columns before the megapixel one. A bigger pixel collects more light, and light decides how much noise you see at dusk. Extra pixels mostly disappear when Frigate or a 1080p stream scales the frame down. The software column matters just as much, since “stock” means the camera works with Raspberry Pi OS packages alone.
| Camera | Sensor | Resolution | Sensor size / pixel | Focus | Lens / field of view | Software |
|---|---|---|---|---|---|---|
| Camera Module 3 | Sony IMX708 | 11.9MP, 4608×2592 | 7.4 mm diag / 1.4 µm | PDAF autofocus, 10 cm to infinity | Fixed, 75° diagonal, f/1.8 | Stock rpicam |
| Camera Module 3 Wide | Sony IMX708 | 11.9MP | 7.4 mm diag / 1.4 µm | PDAF autofocus, 5 cm to infinity | Fixed, 120° diagonal, f/2.2 | Stock rpicam |
| High Quality Camera | Sony IMX477R | 12.3MP, 4056×3040 | 7.9 mm diag (1/2.3″) / 1.55 µm | Manual, set by the lens | C/CS mount, no lens included | Stock rpicam |
| AI Camera | Sony IMX500 | 12.3MP, on-sensor inference | 7.857 mm diag / 1.55 µm | Manual | Fixed, 78.3° diagonal | Stock, plus imx500-all |
| Arducam IMX519 | Sony IMX519 | 16MP, 4656×3496 | 1/2.53″ / 1.22 µm | Autofocus | 66° horizontal | Arducam libcamera for autofocus |
| Arducam 64MP Hawkeye | 64MP sensor | 9152×6944 | 1/1.7″ / 0.8 µm | Autofocus | 72° horizontal | Arducam libcamera and driver |
Most people misread the Hawkeye row. It has the largest sensor on paper, but its 0.8 µm pixels have about a third the area of the Module 3’s. So it trades low-light performance for detail in good light. The HQ Camera shares the biggest pixels of the group with the AI Camera and lets you put real glass in front of them, which is why it’s the wildlife pick.
Why the Camera Module 3 is the default pick
For most builds the best Raspberry Pi camera is the Camera Module 3, because it removes the two jobs that cause the most trouble: focusing and software. Its phase-detection autofocus finds focus on its own, and the IMX708 driver and tuning ship in Raspberry Pi OS, so rpicam-vid sees it the moment the ribbon is seated.

Raspberry Pi’s product brief lists video at 1080p50, 720p100 and 480p120. An HDR mode up to 3MP helps when a camera faces a bright window or a driveway in low sun. There are four versions: standard (75° diagonal, focus from 10 cm), Wide (120°, focus from 5 cm), and a NoIR build of each without the infrared cut filter. Raspberry Pi promises production until at least January 2030, which matters if you build more than one.
Pick the version by distance, not by taste. A printer enclosure puts the bed 20 to 30 cm from the lens, and there the standard field of view crops the corners, so get the Wide. The standard module suits a doorway or a hallway feeding Frigate. NoIR versions only make sense with an infrared illuminator. Without the filter, daytime colors shift pink, and at night the sensor sees nothing until you light the scene.
Who it’s for: a Frigate camera node on a Pi Zero 2 W or Pi 4, a printer camera, and anyone who wants a Pi camera without extra packages. Skip it if you need optical zoom or swappable lenses (get the HQ Camera). Skip it too if you want detection without a Hailo or a separate NVR (get the AI Camera).
PiShop had the official module in stock on November 1 at about $29 for the standard and $27.50 for the NoIR, but the Wide was sold out. On Amazon the Raspberry Pi listings had no seller, so the practical Amazon route is Arducam’s bundle of the official Camera Module 3, which adds an acrylic case and a 15-to-22-pin cable for about $46. The trade-off is price: you pay roughly $17 more than the bare module at a reseller, although the cable saves a Pi 5 or Zero owner a separate $4 order. Check the live price before you buy.
When the HQ Camera earns its higher price
Pay for the High Quality Camera when the lens matters more than the sensor. That is the case for wildlife, bird feeders and anything you watch from across a yard. It pairs a Sony IMX477R with a C/CS mount, so the field of view is whatever lens you screw on. Its 1.55 µm pixels tie the AI Camera’s for the largest in this guide.

No lens comes in the box, and buyers often forget that cost. Raspberry Pi sells two lenses. The 6 mm CS wide lens gives the HQ Camera roughly 55° by 45°. The 16 mm C telephoto narrows it to about 22° by 17°, which is what you want for a nest box or a feeder 5 m away. Adafruit listed them at about $49 and $78 on November 1, so a telephoto HQ build passes $120 before you add a Pi. The C-CS adapter is included and back focus is adjustable. You can also remove the IR cut filter, but Raspberry Pi warns that removal is permanent.
Best fit: wildlife and garden cameras, telescope and microscope builds, and anyone who wants to change the framing later without buying a new camera. Skip it if the camera points at a printer bed or a doorway, where a Module 3 does the job for half the money and focuses itself. There is no NoIR version either, so night work means removing the filter for good.
Raspberry Pi resellers list the HQ Camera at about $55, and PiShop had it in stock on November 1. The HQ Camera listing on Amazon was in stock from a third-party seller at about $100, nearly double that. Check the live price on both.
What the AI Camera does that the others cannot
A neural network runs on the AI Camera’s image sensor itself. Its Sony IMX500 carries a small accelerator with about 8 MB of model memory. So the Pi receives detections as metadata next to each frame, and its CPU stays free. Raspberry Pi says it works with any Pi that has a camera connector, so a small board can count birds or people without a HAT.
For Frigate users, none of that reaches Frigate. Detections come out through rpicam-apps post-processing and Picamera2, with the models in the imx500-all package. Frigate’s docs never mention the IMX500, and Frigate runs its own detector on the video it receives. Fed into Frigate, the AI Camera is a 12.3MP camera with a fixed lens at roughly three times the price of a Module 3. Our Raspberry Pi 5 AI accelerator guide compares it with the Hailo HATs, which Frigate does support.
Use it for: standalone counters, wildlife triggers that save a clip only when an animal is in frame, and Picamera2 projects where you write the logic yourself. Skip it if the camera feeds Frigate, because you pay for an accelerator Frigate ignores. Raspberry Pi lists it at $70 with production until at least January 2028. PiShop had it at about $75, while the AI Camera on Amazon was about $113 from a third-party seller with low stock, so check the live price first.
Why the Arducam IMX519 is the budget autofocus option
For about $25 on Amazon, the Arducam IMX519 gives you a 16MP autofocus camera, which undercuts the Camera Module 3 bundle there and roughly matches the official module’s reseller price. Arducam’s docs list a 1/2.53″ Sony sensor at 4656×3496, a 66° horizontal field of view and support on the Pi 5 under Trixie and Bookworm.
Those savings come with a software trade. Raspberry Pi’s kernel carries an imx519 overlay, but Arducam’s own docs say autofocus control needs Arducam’s build of libcamera and rpicam-apps, installed with its script. That adds a second package source to your Pi. We rank it below the Module 3 for that reason, even though it costs less. A Raspberry Pi OS update can replace those packages, and you will be the one who notices when focus stops working.
Good for: tinkerers who want more pixels on a tight budget and do not mind maintaining a vendor package. Skip it if the Pi runs unattended, such as a Frigate node in a loft, because that is exactly where a broken focus after an update goes unnoticed for weeks. It was about $25 in stock from Arducam’s store on Amazon on November 1; check the IMX519’s live price on Amazon.
Where the Arducam 64MP Hawkeye fits
Treat the Hawkeye as a stills and time-lapse camera, not a video camera for Frigate. Its 64MP sensor (9152×6944 on a 1/1.7″ chip) records far more detail than any stream can carry. That pays off for plant time-lapses, document capture and close-up stills in daylight.
Its trade-offs stack up quickly. Small 0.8 µm pixels make it the weakest low-light camera here. Arducam’s docs also require its own install script, which adds a custom libcamera, rpicam-apps and a kernel driver before the arducam-64mp overlay works.
Buy it for: high-resolution stills, time-lapse rigs and scanning projects in good light. Skip it if the camera’s job is video, night monitoring or a printer feed, where a Module 3 does better for less. It was about $60 in stock on Amazon on November 1; check the Hawkeye’s live price on Amazon.
How a CSI camera reaches Frigate through go2rtc
Frigate expects network streams, and a ribbon camera has no network stream until something makes one. The Pi camera speaks MIPI CSI-2 to the board, libcamera turns that into frames, and rpicam-vid encodes them to H.264. go2rtc ships inside Frigate and also runs on its own. Its exec source starts a command and restreams whatever that command writes to stdout as RTSP.
In order, the path has three moves:
- Capture: the camera sends raw frames over the ribbon, and libcamera applies the sensor’s tuning, autofocus and exposure.
- Encode and restream: go2rtc launches
rpicam-vidthrough an exec source and reads its H.264 output. It serves that as RTSP on port 8554, and go2rtc’s README lists WebRTC for browsers on 8555. - Detect and record: Frigate pulls that RTSP stream like any IP camera and decodes it. Frames then go to its detector, a Hailo on a Pi 5 or an Intel iGPU on a mini PC.
Drawn out, the same path also shows the live view branch that Home Assistant and browsers use:

Two details in that path decide which Pi you put behind the camera. Frigate’s restream docs say exec sources in Frigate’s bundled go2rtc are disabled by default. They only run when the GO2RTC_ALLOW_ARBITRARY_EXEC environment variable is true in Frigate’s container, so a config copied from an older forum post will not start the camera until you set it. That gate is Frigate’s: it applies when the exec stream lives in Frigate’s own go2rtc block, not when a separate go2rtc on the camera Pi does the exec. The second detail is the encoder. A Pi 5 has no hardware H.264 encoder, so rpicam-vid encodes in software through libav on its CPU. A Pi 4 or Pi Zero 2 W still encodes in hardware, and that makes a cheap Zero 2 W a good dedicated camera node.
So the usual architecture splits the work. A Pi Zero 2 W or Pi 4 sits next to the camera and runs go2rtc, and Frigate runs on a box with a proper detector. Our Frigate NVR appliance build is that box on an N150 mini PC, and the Frigate Docker install guide covers the software. If you want the detector on a Pi 5 instead, the Hailo vs Coral comparison explains why Frigate’s docs now point at Hailo.
What OctoPrint and Klipper expect from a Pi camera
Printer software wants an MJPEG stream at 1080p or below, so resolution beyond that is wasted and autofocus is the feature that matters. A camera 20 cm from a print bed has a shallow depth of field, and a fixed-focus module set for a room will show the nozzle as a blur.
On the OctoPrint side, OctoPi 1.1.0 moved to a Bookworm base and added Pi 5 support. Its release notes say the Camera Module v2 does not work on the plain stable image but does on “OctoPi with New Camera Stack”, and we would start any libcamera camera, Module 3 included, on that image. On the Klipper side, Mainsail’s crowsnest points Pi camera users to its spyglass backend, a Picamera2 MJPEG server whose autofocus defaults to continuous. The crowsnest docs say to keep 16:9 sensors like the Camera Module 3 at or below 1920×1080 on any Pi, because higher resolutions can crash spyglass. Spyglass always serves MJPEG, and on a Pi 5 it keeps WebRTC off unless you pass --force-webrtc, because software H.264 is CPU-heavy there and can disturb Klipper’s timing.
Put together, we would buy the Camera Module 3 Wide for a printer. Its 120° view covers a full bed at close range, focus starts at 5 cm, and spyglass keeps it focused as the print grows. The HQ Camera works too, but you set focus by hand on the lens, and a fixed focus plane can’t follow a print as it grows taller.
What to check before you buy a Pi camera
A handful of details decide whether a Pi camera works for your build, and no listing puts them in its title. Check these against your board and your use before you order:
- The cable for your board: Raspberry Pi cameras ship with a standard 15-pin cable (retailers list the AI Camera with both cables). The Pi 5 and the Zero boards use the mini 22-pin connector, so they need the Standard-Mini cable, about $4 from a reseller.
- Mainline or vendor software: the official modules work with stock Raspberry Pi OS packages, whereas the IMX519’s autofocus and the Hawkeye need Arducam’s own libcamera build.
- Pixel size over megapixels: 1.4 to 1.55 µm pixels (Module 3, HQ) handle dusk far better than 0.8 µm ones, and Frigate and printer streams downscale the extra pixels anyway.
- NoIR means you supply the light: a NoIR camera sees infrared only when an IR illuminator lights the scene. Its daytime colors also look off, so use it for night-only or IR-lit spots.
- The encoder on the Pi behind it: a Pi 4 or Zero 2 W encodes H.264 in hardware, while a Pi 5 spends CPU on it. That matters when the same Pi also runs Frigate or Klipper.
- Ribbon length: the official Module 3 and HQ ribbons are 200 mm, so plan where the Pi sits, or buy a longer cable before you mount anything outdoors.
Of these, the cable catches the most Pi 5 buyers, because the camera arrives with a cable that does not fit the board. If you are also choosing the board, our Raspberry Pi 5 kit guide covers which kits include what, and if the camera mounts on the board, pick a Pi 5 case with a camera slot so the ribbon has a way out. A camera node that records locally should also run on a high-endurance microSD card rather than the one in the box.
Common misconceptions about Raspberry Pi cameras
A few beliefs keep sending people to the wrong camera, and each one has a mechanism behind it that’s worth understanding.
“More megapixels means better Frigate detection”
Frigate resizes every frame to its detect resolution before the model sees it. Contrast and noise survive that resize, and both come from pixel size and light. So a Module 3 beats a Hawkeye at dusk, even though the Hawkeye has five times the pixels.
“The AI Camera makes Frigate faster”
Frigate runs its own detector on the decoded stream and doesn’t read the IMX500’s metadata, so the on-sensor model does nothing for it. If Frigate detection is slow, the fix is a Hailo, a Coral or an iGPU on the Frigate host, not a different camera.
“A Pi camera is just a webcam”
A USB webcam hands the system a ready-made video device that ffmpeg can open. A CSI camera hands libcamera raw sensor data, which needs tuning, autofocus and encoding before it is video at all. That is why the path into Frigate runs through rpicam-vid and go2rtc rather than a device path. In our view the extra step is worth it, because it buys the better sensors, autofocus and HDR that webcams at this price lack.


