What Is Video Transcoding? A Complete Beginner’s Guide

Video transcoding converts a video that is already compressed into another compressed version. The new version can use a different codec, resolution or bitrate. It is why one upload plays on a phone, a smart TV and a slow café connection. This guide explains how it works, what it costs in computing time, and where to run it.

Along the way, it puts real numbers on transcoding: file sizes, processing time, and the cost of preparing a video library. It also draws on MDN, NVIDIA and a university study of streaming costs.

What Is Video Transcoding, and What Does It Change?

Video transcoding converts compressed video into another compressed version, by decoding it and encoding it again. The new version can use a different codec, a smaller resolution, a lower bitrate, or all three.

A codec is the method that compresses the video. The bitrate is how much data each second of video uses. Resolution is the size of the picture. So change any of them, and you have transcoded the video.

Encoding compresses raw footage for the first time. Transcoding, by contrast, starts from video that is already compressed, so it always runs a decode first.

The scale explains why compression exists. For example, MDN calculates that one minute of uncompressed 1080p video at 30 frames per second needs about 14.9 GB. So a ten-minute clip would need about 149 GB.

Table comparing transcoding, transrating, transsizing and transmuxing by what changes, and whether the video is re-encoded: the first three re-encode, and transmuxing, which only changes the container, does not.

By contrast, transmuxing is the cheap one. It repackages the same compressed streams, so it is fast and loses no quality. The other three all re-encode.

How Does Video Transcoding Work?

Video transcoding works in three steps: decode the source into raw frames, adjust them, and encode them in the target format. NVIDIA’s engineers describe the same flow: decode the input stream, scale it, and re-encode it in the desired formats.

Flow diagram of video transcoding: a compressed source video is decoded back to raw frames, adjusted by resizing or filtering, encoded in the target codec, and written as one output file per version.

Why Can Video Transcoding Lose Quality?

Video transcoding can lose quality because most codecs are lossy, so each re-encode discards a little more detail. MDN explains that the more compression a codec achieves, the more detail and fidelity it loses. Therefore, transcode from the best source you have, and avoid transcoding the same file repeatedly.

Why Does Video Transcoding Use So Many CPU Cores?

Video transcoding uses many cores because a video splits into chunks that cores can process in parallel. A university study of streaming repositories notes that each group of pictures can be processed independently. Transcoding therefore commonly works at that level. In addition, MDN notes that HEVC’s block structure was designed so that software can use several cores easily.

Why Does Video Transcoding Exist?

Video transcoding exists because viewers use different devices and connections, and one file cannot suit all of them. YouTube tells broadcasters that it automatically transcodes every live stream into many output formats for exactly that reason.

The set of versions is an adaptive bitrate ladder, and the player switches between its rungs as the connection changes. MDN describes DASH, one open standard for this, as a protocol that switches bit rates based on network performance. Smaller rungs can also look better than you would expect. MDN notes that scaling 1080p down to 720p before encoding can beat squeezing 1080p into the same size.

How Much Space Does a Ladder Take?

A four-rung ladder takes about 848 MB, under a quarter of the 3,750 MB master in this example. The master is a ten-minute camera file at 50 Mbps. Transcode it into rungs of 6, 3, 1.5 and 0.8 Mbps, and the files weigh 450, 225, 112.5 and 60 MB.

Horizontal bar chart of file sizes for a ten-minute video: a 50 Mbps camera master at 3,750 MB, all four delivery rungs together at 848 MB, and the individual 1080p, 720p, 480p and 360p rungs at 450, 225, 112.5 and 60 MB.

In short, a ladder adds little storage when the master’s bitrate is far above the rungs’ bitrates.

Which Codec Should You Transcode To?

Transcode to H.264 for the widest compatibility. Add AV1 or HEVC when smaller files matter more than reaching every device. MDN’s own recommendations for the web follow the same logic: a WebM file with AV1 and Opus, or an MP4 file with H.264 and AAC.

Table comparing the H.264, HEVC and AV1 video codecs by compression, device support and licence: H.264 is the baseline and broadly compatible, HEVC can halve file size but depends on the device, and AV1 compresses up to 50 percent better and is royalty-free.

Likewise, hardware encoders show a similar pattern. In Fora Soft’s benchmark, hardware HEVC needed about 30% less bitrate than the x264 baseline. Hardware AV1 needed about 40% less. Those are house results on one machine, so treat them as direction, not promise.

What Is the Difference Between a Codec and a Container?

A codec compresses the video, while a container is the file that holds the compressed video and audio. For example, MDN’s codec table shows which fit together: H.264 goes in MP4, AV1 goes in MP4 or WebM, and HEVC goes in MP4. A “.mp4” file can therefore hold different codecs, so the extension alone does not say what is inside.

Table of three video containers, MP4, WebM and Ogg, with their file types and the video codecs each can hold: MP4 holds H.264, HEVC, AV1 and VP9, WebM holds AV1, VP8 and VP9, and Ogg holds Theora, VP8 and VP9.

How Long Does Video Transcoding Take?

In one published benchmark, video transcoding took from about 5% to 250% of a clip’s length, depending on the encoder and its preset. The university study calls the process compute-intensive and time-consuming, and the numbers show why.

In Fora Soft’s benchmark, NVIDIA’s hardware H.264 encoder, NVENC, ran at 600 frames per second at 1080p. Meanwhile, the x264 software encoder ran at 240 at its veryfast preset, 95 at medium and 12 at veryslow. A ten-minute clip at 30 frames per second has 18,000 frames, so you can divide by those speeds.

Horizontal bar chart of the time to transcode a ten-minute 1080p clip: x264 veryslow takes 25 minutes, x264 medium 3 minutes 9 seconds, x264 veryfast 1 minute 15 seconds, and NVENC H.264 on a GPU 30 seconds.

Slower presets compress better, so the fastest option is not always the best one. Even so, a longer wait only makes sense when many people will watch the result. That leads to the next question.

Should You Transcode Every Video in Advance?

Not always. A university study found that preparing every version in advance is inefficient, because most videos are watched rarely. In other words, the authors describe a long tail: a few videos collect most views, and a huge share are barely watched.

Their paper also reports that large platforms prepare dozens of versions per title. It cites Netflix preparing 40 to 50 formats of one video, from a 2012 talk, so the figure is dated. Storing all of them costs money. Transcoding on request costs money too, because cloud compute is generally pricier than storage.

Table comparing three strategies for preparing a video library: transcoding everything in advance, transcoding on demand, and partial pre-transcoding, by storage cost, compute cost, and the type of library each suits best.

As a result, the authors proposed the third option. The start of a video is watched more than the rest. So they keep the opening segments ready and transcode the remainder on demand. In their simulation of 50,000 videos, it cut costs by up to 70% against transcoding on every request. Against transcoding everything in advance, the saving reached up to 60%. The prices came from Amazon’s rates at the time.

Where Does Video Transcoding Run?

Video transcoding runs on a CPU, a GPU encoder or a cloud service. The right place depends on volume, quality and control, and each one trades something away.

• A CPU runs software encoders such as x264 and x265. Slow presets give the best compression, but they cost time.
• A GPU encoder such as NVENC is much faster, so it frees the CPU for other work.
• A cloud service usually bills for the video it processes and hides the hardware, so you control little.
• A rented dedicated server costs a flat monthly fee and gives you root access, so you pick the encoder.

Still, choosing between those four is a hardware question more than a video one. It deserves its own guide.

How Do You Transcode a Video Yourself?

You can transcode a video yourself with one FFmpeg command, on any computer or server. This example turns a MOV file into an MP4 that plays almost everywhere:

ffmpeg -i input.mov -c:v libx264 -preset medium -crf 23 -c:a aac output.mp4

In short, the command reads the source, re-encodes the video with the x264 encoder, and writes an MP4. The preset trades speed for compression. The CRF value sets quality, so lower numbers give better quality and bigger files.

FFmpeg also prints a speed figure when it finishes, where 1.0x means real time. That single number tells you how fast your hardware transcodes.

Video Transcoding: The Short Version

Video transcoding converts compressed video into other compressed versions. Every viewer then gets a file that fits their device and connection. It costs computing time, so match the encoder to the job.

• Use H.264 for compatibility, and add AV1 or HEVC when smaller files matter more.
• Expect minutes of processing per version: slow CPU presets compress best, and a GPU encoder is the fast option.
• Prepare in advance what many people watch, and transcode the long tail on demand.

Frequently Asked Questions

What is video transcoding?

What is the difference between video encoding and transcoding?

Does video transcoding reduce quality?

How long does video transcoding take?

Is video transcoding the same as compression?

What is the best codec for video transcoding?