Hector Weyl AI Camera vs AI NVR: Which Do You Need?

Choose AI in the camera when the camera can perform the required detection and your recorder can receive and use its events. Choose AI in the NVR when you want the recorder to analyse supported video streams—and its analysis capacity covers every view that needs protection.

The important part of the AI camera vs AI NVR decision is the location and capacity of each function. A recorder may accept 16 video channels while analysing perimeter events on only two. Every camera can look healthy on the monitor while several required views have no perimeter analysis assigned.

This guide uses Hector Weyl terminology, a current IntelliSight recorder specification, and a worked planning example. Product specifications were checked on 4 November 2026. The example is a design calculation, not a reported installation or a performance test.

1. What changes when AI runs in the camera?

With AI by Camera, the camera analyses its own scene and sends supported detection results to the NVR. The recorder receives the video and uses the compatible event information for functions such as event display, recording linkage or retrieval. Exactly which actions are available depends on the camera, recorder, connection method and firmware.

With AI by NVR, also called AI by Recorder, the camera sends video and the recorder performs the selected analysis. The camera does not necessarily need to generate that particular AI event itself. It still needs to deliver a supported stream that meets the recorder's requirements.

Two diagrams show analysis inside the camera or inside the recorder, with video and event results travelling to the recording system. — Figure 1. Locate the analysis first. Camera-side events and recorder-side analysis take different routes, even when their live video appears on the same screen. Original explanatory diagram.

Neither route removes the need to size recording, network capacity and storage. Moving detection into a camera does not automatically reduce the recorded video bitrate. Likewise, fitting an AI recorder does not create unlimited analysis capacity for every connected channel.

2. Read four capacities before comparing features

Start with a recorder specification rather than the largest number on its product title. These capacities answer different questions:

Capacity	What it tells you	What it does not establish
Video inputs	How many supported camera streams the recorder can accept.	How many channels can run a particular AI function.
Incoming bandwidth	The recorder's specified incoming video data allowance.	AI event compatibility or enough analysis channels.
Decoding capability	How much video can be decoded under the stated resolution and frame-rate conditions.	Equivalent AI processing capacity.
AI capacity by function and location	How many channels support a named task through camera-side or recorder-side processing.	That all listed AI functions can run together.

Then read the restrictions alongside those numbers. The Hector Weyl NVR manual separates camera-side and recorder-side AI and warns that functions can vary by model or conflict on the same channel. A specification listing perimeter protection and face detection does not, by itself, confirm simultaneous operation.

For an existing installation, write down the actual camera models and firmware before choosing an upgrade. A series name is a useful starting point; the complete SKU and its supported functions determine the design.

3. A real IntelliSight example: 16 inputs, two perimeter-analysis channels

The published specification for the HW-NVR-N6216E-L-4K IntelliSight NVR separates these limits explicitly. Its camera-side figures describe support for results from compatible cameras; they are not additional recorder-side inference channels.

Published field	Listed capacity	How to interpret it
IPC input	16 channels	The video-input count.
Incoming bandwidth	256 Mbps	A separate incoming-data specification.
Perimeter protection by camera	16 channels	Requires compatible cameras providing the supported function.
Perimeter protection by NVR	2 channels	The listed recorder-side perimeter-analysis limit.
AMD by NVR	4 channels	A different function with a different channel limit.

AMD means Active Motion Detection in Hector Weyl's terminology. Hector Weyl also uses Perimeter Sentinel for perimeter protection; the table preserves the published specification's field names. An AMD allocation is not automatically a perimeter-rule allocation. Specify whether the task is classifying motion or responding to a configured intrusion area or tripwire. The camera and recorder must support the particular task being requested.

These figures identify the planning questions, not a tested combination of every feature. Confirm the intended firmware, input conditions and simultaneous-function restrictions before final selection. Do not add the two perimeter channels to the four AMD channels and describe the result as six interchangeable AI channels.

4. Work through an eight-view perimeter requirement

Consider a proposed site with 16 cameras. The brief requires perimeter rules on eight views; the remaining views need ordinary recording. Assume the designer proposes using the NVR's own perimeter analysis for all eight priority views.

Using the listed two-channel recorder-side perimeter limit above:

Uncovered perimeter views = required views − allocated capacity = 8 − 2 = 6.

The issue is not whether all 16 cameras can appear on screen. It is that six of the eight required views still lack an analysis allocation under this proposed design.

An illustrative site has 16 camera inputs, eight views requiring perimeter rules, two recorder-side perimeter channels and six views left to resolve. — Figure 2. An illustrative project requirement compared with the selected recorder's published perimeter limit. The six-view gap is arithmetic, not a measured detection result.

One route is to use suitable AI cameras for the protected views and verify their events with the recorder. Another is to select a recorder whose documented perimeter-analysis capacity covers the requirement at the intended settings. A mixed design may also be possible, but its camera-side and recorder-side functions must be confirmed together.

Make that decision on the schedule of cameras, before purchasing. Assign each protected view an analysis location, required event and acceptance test. Leave unresolved combinations marked “to verify” rather than assuming every camera or NVR in a series behaves identically.

5. When each route makes sense

Camera-side AI is worth evaluating when several distributed views need their own supported detection. It can spread analysis across suitable cameras instead of relying on one recorder's inference allowance. Start with the relevant IntelliSight cameras or IntelliCore cameras, then check the exact SKU.

For example, the HW-IPC-G2449M-IL-AS-PV product page lists AMD and perimeter protection. That makes it a model to evaluate for those tasks. It does not establish a tested pairing with every recorder, nor does it mean all other IntelliSight models have identical features.

Recorder-side AI is worth evaluating when retaining suitable existing cameras matters and only a defined number of views need analysis. Confirm stream compatibility, resolution and frame-rate conditions, required functions and channel limits. Compare those requirements against the exact models in the IntelliSight AI NVR range.

If the camera image cannot show the target clearly at the required distance or lighting level, changing the analysis location is not the first fix. Review placement, field of view, illumination and image settings before expecting different processing hardware to solve the scene.

6. Test three layers of compatibility

A successful live view proves something useful: the system has obtained a viewable stream. It does not complete the AI acceptance test. Separate the system into three layers:

Video: the stream opens and records under the intended configuration.
Events: the required detection reaches the recorder with the expected classification and channel identity.
Workflow: the event creates the intended recording, alert or searchable result for the operator.

ONVIF's Profile T overview covers advanced video streaming and associated capabilities. Its Profile M overview specifically addresses analytics metadata and events. These profiles illustrate why receiving video and using analytics results are separate integration questions.

An “ONVIF” label alone does not establish that the particular event and workflow you need work between two products. Check the applicable profile, conformant product and firmware information, supported optional functions, and the actual pairing. This guide does not assert Profile M or Profile T conformance for the example Hector Weyl products.

A three-stage acceptance chain checks recorded video, received AI events and the operator's searchable workflow separately. — Figure 3. Validate video, events and the operator's workflow separately. A working picture is the first check, not the final acceptance result. Original explanatory diagram.

7. Commission one channel before copying settings

Use a controlled, permitted test in a representative view. Keep the intended firmware and stream settings, document the chosen analysis location, and change one relevant setting at a time.

Check	Expected result	If it fails, check next
Identify the analysis source.	The intended camera-side or recorder-side function is available and assigned.	Exact SKU, firmware, channel allowance and conflicting functions.
Trigger the required rule.	A controlled crossing or intrusion produces the intended event.	Enabled rule, direction, target type, area and arming schedule.
Check event reception at the NVR.	The matching channel and event appear at the recorder.	Connection method, permissions and supported event integration.
Check recording and retrieval.	The event-linked clip is recorded, found and opened.	Recording schedule, linkage, time alignment, storage status and search filters.
Check the operator's alert.	The intended recipient receives the configured alert, if required.	Account permissions, subscription, application settings and notification path.

Keep one result for each stage. If detection works in the camera but no event reaches the recorder, investigate event integration before making the detection more sensitive. If an event appears but no clip exists, investigate recording and storage before replacing the camera.

Repeat under the lighting and target conditions that matter for the project. A daytime walk through an empty view does not validate the same rule at night, behind parked vehicles, or when several targets enter together. Record the conditions and observed outcome; avoid turning a small commissioning check into an unsupported accuracy percentage.

8. Put the AI allocation in the project brief

A useful selection sheet names each protected view and the task it must perform. Include the complete camera and NVR SKUs, firmware versions, stream configuration, connection method, AI source, concurrent functions, event actions and acceptance evidence.

Record expansion requirements by function. Four future video inputs and four future perimeter-analysis channels are different requests. Check both against the proposed hardware and configuration.

Four questions for a project brief: which view, which event, where analysis runs, and how the result will be accepted. — Figure 4. A brief becomes actionable when each protected view has a named event, an analysis location and a way to verify the result. Original planning aid.

For Hector Weyl procurement, MOQ: 100 units. Send the camera schedule and the required AI allocation when requesting a project recommendation. Confirm model selection, availability and the supported configuration with the team rather than inferring them from a product photograph or a series badge.

9. Frequently asked questions

Does an AI camera need an AI NVR? Not necessarily for the camera to perform its own supported analysis. The recorder still needs to receive and use the required events and recording functions. Verify the complete workflow rather than relying on the recorder's “AI” label.

Can an AI NVR add analytics to an ordinary IP camera? It may analyse a compatible video stream when the recorder supports the required function. Check input conditions, analysis capacity and firmware restrictions. An ordinary camera does not become capable of every AI feature just because it connects.

Does a 16-channel NVR provide 16 AI channels? Not automatically. The example HW-NVR-N6216E-L-4K lists 16 video inputs and two channels of perimeter protection by NVR. Read each AI function's separate specification.

Can camera-side and recorder-side AI run together? Some configurations may support a combination. Confirm simultaneous operation, channel allocation and any conflicting functions on the exact models and firmware. Do not assume separate table entries can all be enabled together.

Is AMD the same as perimeter protection? They are different tasks. Active Motion Detection concerns supported motion classification; perimeter rules apply configured intrusion areas or tripwires. Specify the behaviour you need and check the relevant function's channel limit.

Will AI in the camera save network bandwidth? Not automatically. The recorded video still follows its configured resolution, frame rate, codec and bitrate settings. Event metadata does not mean the camera has stopped transmitting video.

Does AI event recording guarantee IntelliSearch support? No. Confirm the exact search mode, compatible devices, firmware and required event or metadata support separately. Event filtering, image-based target retrieval and text-based retrieval are not interchangeable promises.

10. Sources and next steps

HW-NVR-N6216E-L-4K published specification — input, bandwidth and AI channel limits; checked 4 November 2026.
Hector Weyl IP Video Distribution Products 2025 Ver. 2 — PDF page 70, printed page 63, IntelliSight-L NVR comparison.
Hector Weyl Network Video Recorder User's Manual, file edition 240820 — §5.9.1, printed page 101 / PDF page 114, for processing location and model/function restrictions.
HW-IPC-G2449M-IL-AS-PV specification — example camera's listed AMD and perimeter functions.
ONVIF Profile T and ONVIF Profile M — streaming, metadata and event scope; these references do not certify an untested device pairing.

Planning a new site or upgrading existing cameras? Send Hector Weyl your camera schedule and AI requirements. Include the views that need protection, required events, existing equipment and the recording or alert outcome you expect.