Video bitrate explained simply: what it is, CBR vs VBR vs CRF, good bitrates for streaming and security cameras, what AI analytics needs, and the storage maths.
Bitrate decides whether your cameras capture usable evidence, whether storage stays within budget, whether the network copes, and whether the AI you paid for actually works. It is one of the most important decisions in any surveillance project, yet it is usually left at whatever the camera shipped with.
Most bitrate guides are written for broadcasters and game streamers. That advice is useful, but incomplete for security, which records many cameras at once, keeps footage for weeks, serves several operators, and increasingly relies on AI.
This guide explains what bitrate is, how it relates to resolution and compression, sensible ranges for streaming and for security cameras, how AI changes the decision, and the storage and bandwidth maths.
Video bitrate is how much data a video uses each second, measured in kilobits (kbps) or megabits (Mbps) per second. A 4 Mbps stream produces 4 megabits, or 500 kilobytes, every second.
Bitrate connects three things buyers care about: picture quality, storage cost and bandwidth cost. More bitrate generally means better pictures, more storage and more bandwidth. The relationship is not linear, but the direction is always the same.
Bitrate is not resolution. Resolution is the number of pixels in each frame; bitrate is how much data describes them over time. The same 1080p picture can be encoded at 1 Mbps, visibly compressed, or 10 Mbps, excellent, with very different storage and bandwidth consequences.
Nor is bitrate the compression format. The format (H.264, H.265, AV1) decides how efficiently the data is used. H.265 delivers similar quality to H.264 at a lower bitrate. Bitrate is the amount; the format is the efficiency.
Encoders control bitrate in three main ways.
Constant bitrate (CBR) uses the same amount of data every second, whatever the scene. Storage and bandwidth are predictable, but a quiet corridor at 3 AM uses as much as a busy junction at rush hour, wasting data on quiet scenes.
Variable bitrate (VBR) spends more on busy scenes and less on quiet ones, around a target average. Better quality for the same data, with less predictable moment to moment usage. It is the default for most modern surveillance and streaming.
Constant rate factor (CRF) aims for a fixed visual quality and lets the bitrate go wherever needed. Best quality per bit but least predictable, so it suits archiving and editing more than live recording.
For surveillance, VBR with a maximum cap works best. The cap protects your network and storage budget; the variable encoding inside it gives better quality per bit. CBR still has a place where the network needs perfectly predictable traffic.
Streaming platforms such as YouTube and Twitch, and tools such as OBS, publish well established guidance. It varies slightly by platform but, for H.264 at 30 frames per second, broadly runs as follows.
480p: about 1 to 2.5 Mbps. 720p: about 2.5 to 5 Mbps. 1080p: about 4.5 to 9 Mbps, more for fast motion such as gaming. 1440p: about 9 to 18 Mbps. 4K: about 18 to 25 Mbps for streaming, more for high end broadcast.
H.265 and AV1 reach similar quality at lower bitrates than H.264, though device support for AV1 is still uneven.
Adaptive streaming formats such as HLS, LL-HLS and MPEG-DASH publish several versions of the same video at different bitrates, called a bitrate ladder, and the player switches between them as bandwidth changes.
These ranges are designed for one viewer watching one stream over home internet. They make a useful reference, but security works differently.
Many cameras, all the time. A surveillance system records every camera continuously, often for weeks. Bitrate multiplies across every camera and every day. Recording a large estate at broadcast style bitrates would overwhelm almost any storage budget.
Evidence, not entertainment. The right bitrate depends on the camera's job. General awareness tolerates heavy compression. Face recognition and number plate capture need enough detail to identify. AI needs enough detail for the model to work reliably.
Many viewers. Bitrate also multiplies across operators. One 4K camera at 12 Mbps watched by one person is 12 Mbps; watched directly by five people, it can be 60 Mbps. A good video platform manages this for you.
So surveillance bitrate guidance differs from broadcast guidance, as below.
As a starting point, at 25 frames per second with VBR and moderate activity:
With H.264: 720p around 1.5 to 3.5 Mbps for general watching. 1080p around 3 to 6 Mbps for general watching, more for face recognition and more again for number plate capture. 4K around 8 to 15 Mbps for general watching, more for AI at full resolution.
With H.265, the same scenes need noticeably less: 1080p around 1.5 to 4 Mbps for general work and 3 to 6 Mbps for AI; 4K around 4.5 to 10 Mbps for general work and higher for AI.
Busy scenes such as shop floors, transit hubs and road junctions need more than these baselines. Quiet scenes such as overnight corridors and server rooms can often use less without losing anything that matters.
Treat these as starting points, then test on your own footage.
AI usually needs more bitrate than human viewing, and the gap is one of the most common causes of disappointing AI in real projects.
Face recognition. Accuracy drops sharply once compression strips detail from the face. Cameras covering doorways for face recognition belong at the healthier end of the range.
Number plate recognition. Even more sensitive, especially at distance, because heavy compression smears characters into blocks. Plate cameras need more bitrate than general cameras at the same resolution.
Object detection, person tracking and motion detection. Much more tolerant: 1080p at modest H.265 bitrates usually works well.
Pose estimation and unsafe behaviour detection. In between: enough detail to see arms, legs and posture.
The lesson: set bitrate per camera according to its AI job. Face and plate cameras at the higher end, general AI cameras in the middle, cameras for simple awareness at the lower end.
As baseline guidance with H.265 at 25 frames per second and VBR:
Face recognition at doorway distance: 1080p at around 4 to 6 Mbps, or 4K at around 8 to 12 Mbps. Closer subjects can sit lower; more distant ones need more.
Number plate recognition at an entry lane: 4K at a healthy bitrate with the right lens, or a dedicated ANPR camera that puts more pixels on the plate.
Pose estimation and unsafe behaviour: 1080p at around 3 to 5 Mbps at typical distances; 4K for wide scenes where workers are further away.
Object detection, person tracking and motion detection: 1080p at around 2.5 to 4 Mbps is usually enough.
Crowd detection and counting: 1080p at around 3 to 5 Mbps; more if you need to follow individuals through the crowd.
Lighting, weather, activity and camera placement all affect what the AI really needs, so pilot on your own footage before fixing bitrates.
Bitrate converts to storage by simple arithmetic. A camera recording continuously at 4 Mbps produces 30 megabytes a minute, 1.8 gigabytes an hour and about 43 gigabytes a day. Multiply by retention days for storage per camera, then by cameras for the total.
Per camera per day: 2 Mbps is about 22 GB; 4 Mbps about 43 GB; 6 Mbps about 65 GB; 10 Mbps about 108 GB; 16 Mbps about 173 GB.
Worked example: 100 cameras, 50 at 4 Mbps, 30 at 6 Mbps and 20 at 10 Mbps, recording continuously. That is about 6.3 terabytes a day: roughly 190 terabytes for 30 days, or 570 terabytes for 90 days.
Three decisions cut storage substantially: event based recording (lower bitrate most of the time, higher when motion or AI events occur), efficient compression (H.265 rather than H.264), and tiered retention (recent footage at full quality, older footage moved to cheaper storage).
Bitrate is also bandwidth, both from cameras to the recording platform and from the platform to operators.
A 4K camera at 12 Mbps sends 12 Mbps to the recorder, constantly. A hundred such cameras send 1.2 Gbps, far more than most site internet connections can upload. Estates like that should record locally at the site, sharing selected footage centrally as needed.
Viewing multiplies across operators. The proven approach is to view routine live tiles using each camera's lighter substream and switch to the full quality main stream only when an operator opens a camera full screen.
For remote operators, phone users and multi site setups, lighter streams are almost always needed.
Visylix is an automated AI video surveillance platform built in India, designed around the realities of recording and analysing many cameras at once.
The right stream for every view. Visylix supports a main stream and a substream for every camera. It records the full quality main stream for evidence, while small live tiles on video walls of up to 36 tiles automatically use the lighter substream, and opening a camera full screen switches to full quality.
Original quality, preserved. Visylix records H.264 and H.265 exactly as the camera sends them, converting only where a browser needs H.264 for live view, so the original is preserved everywhere else. Live view arrives in under a second over WebRTC, down to under 200 ms.
Storage that works harder. Event based recording driven by motion, AI detections or operator triggers keeps storage lean, and archive tiering moves older footage to your own S3 compatible storage automatically, with integrity checks.
AI designed in, not bolted on. Visylix runs 22 AI analytics built in house, and small, distant subjects are still detected. Because bitrate and AI are linked, we match each camera's bitrate to its AI job, whether face recognition, number plate recognition or pose estimation, as part of deployment design, never left to chance.
Built for India and the world: an interface in 55 languages including 13 Indian languages, rupee pricing through Razorpay, and deployment on your own infrastructure, which keeps bandwidth inside your own network. Starter and Pro include unlimited cameras on the VMS licence, from ₹4,999 or $49 per month, with AI added per camera per month.
Planning a new deployment, or hitting a storage or bandwidth wall? Talk to the Visylix team: https://visylix.com/contact.
Video bitrate is how much data a video uses each second. Together with resolution and compression format, it decides picture quality, storage cost and bandwidth cost.
Broadcast guidance is a useful reference, but surveillance differs: many cameras, long retention, AI and several viewers.
For surveillance, use VBR with a cap. General awareness cameras need less bitrate than broadcast; AI cameras, especially for faces and plates, need more.
Storage and bandwidth multiply across the estate. Event based recording, H.265, tiered retention and substreams for live tiles keep them in check.
Set bitrate per camera, by its job. That is how you get reliable AI without breaking the budget.
Video bitrate is how much data a video uses every second, measured in kilobits (kbps) or megabits (Mbps) per second. Higher bitrate generally means better picture quality, more storage and more bandwidth. It is not the same as resolution (the number of pixels in each frame) or the compression format (how efficiently the data is used), though it works closely with both.
It depends on resolution and compression format. For 1080p at 30 frames per second using H.264, common platform guidance is roughly 4.5 to 9 Mbps, with fast moving content such as gaming needing more. 4K at 30 frames per second with H.264 commonly runs 18 to 25 Mbps. H.265 reaches similar quality at noticeably lower bitrates. YouTube and Twitch each publish their own recommended ranges.
For 1080p streaming with H.264 at 30 frames per second, around 4.5 to 6 Mbps suits calm content, 6 to 9 Mbps general content, and more for fast motion. For 1080p security cameras with H.265 at 25 frames per second, around 3 to 6 Mbps is a typical starting point, towards the upper end for cameras doing face recognition.
Match it to the camera's job. As a starting point with H.265 at 25 frames per second: 1080p cameras for general watching around 3 to 6 Mbps, more for face recognition and more again for number plate capture; 4K cameras around 4.5 to 10 Mbps for general watching and higher for AI work. Busy scenes need more; quiet scenes need less. Always test on your own footage.
Directly. Too little bitrate causes visible blockiness, smearing, banding and lost detail. The gain is not linear: raising a very low bitrate improves quality dramatically, while raising an already high one gives diminishing returns. The compression format matters too: H.265 looks better than H.264 at the same bitrate.
Adaptive bitrate streaming produces several versions of the same video at different quality levels, and the player switches between them as the viewer's bandwidth changes. HLS, LL-HLS and MPEG-DASH support it. It is standard for consumer video because it copes with changing internet conditions automatically.
Resolution is the number of pixels in each frame (1920x1080 for Full HD, 3840x2160 for 4K). Bitrate is how much data describes those pixels each second. The same 1080p picture can be squeezed into 1 Mbps (visibly compressed) or given 10 Mbps (excellent). Resolution sets the size of the picture; bitrate sets the quality of what is in it.
CBR (constant bitrate) uses the same amount of data every second, whatever is happening: predictable, but wasteful on quiet scenes. VBR (variable bitrate) spends more on busy scenes and less on quiet ones, around an average: better quality per bit, less predictable. CRF (constant rate factor) aims for a fixed picture quality and lets the bitrate float: best quality per bit, least predictable. For surveillance, VBR with a maximum cap works best.
Significantly. Face recognition needs enough detail on the face, so it needs a healthier bitrate than general watching. Number plate recognition is even more sensitive, because compression can smear characters. Object detection, person tracking and motion detection cope well with lower bitrates, and pose estimation sits in between. Set bitrate per camera according to its AI job, not one default for every camera.
It is simple arithmetic. A camera recording continuously at 4 Mbps produces about 43 GB a day; at 8 Mbps about 86 GB; at 16 Mbps about 173 GB. Multiply by your cameras and retention days and the totals grow fast. Motion or event based recording, H.265 and tiered retention all cut storage substantially, depending on how busy your scenes are.