Audio and Video Glossary
Every term here is one you will meet while cutting something together, and each one says what it changes about your work rather than what a standards body calls it. Written for someone who knows what they want the finished piece to look like and has been away from production long enough that the vocabulary has gone cold.
187 terms
- Frames, time and speed
- Containers, codecs and files
- Resolution, framing and geometry
- Color, levels and exposure
- Sound, levels and loudness
- Editing craft
- Delivery and publishing
Nothing matches that. Try a shorter word, or clear the category.
Frames, time and speed
-
23.976
Also called 23.98
24 frames per second slowed by a factor of 1000/1001.
It exists because colour television had to fit a colour subcarrier into a signal built for black and white, and the fix was to slow the frame rate by a tenth of a percent. Everything downstream inherited it. If your source is 23.976 and you build a 24 timeline, audio will drift about 1.5 seconds every hour and you will not hear it until the end of a long piece.
See also: Drop frame timecode, Frame rate, Pull-down
-
Drift
Sync error that grows over the length of a piece instead of staying constant.
Constant offset means something was placed wrong and you can slide it back. Growing offset means a rate mismatch somewhere: 23.976 material on a 24 timeline, or a recorder whose clock ran slightly fast. Sliding does not fix drift. You have to fix the rate.
-
Drop frame timecode
A counting trick that keeps 29.97 timecode matching the clock. No frames are actually dropped.
At 29.97 frames per second, counting 30 frames to the second runs fast against real time by about 3.6 seconds an hour. Drop frame skips two frame numbers at the top of each minute, except every tenth minute, so the numbers stay honest. The pictures are all still there. Only the labels change.
-
Duration
How long something runs, measured from its first frame to one frame past its last.
A clip from frame 0 to frame 29 at 30fps is 30 frames long and lasts one second. Off by one errors here are the reason two clips that should butt together leave a one frame flash of black.
-
Frame
One still picture. Video is a stack of them shown fast enough to read as motion.
Everything in editing is counted in frames, not seconds, because a frame is the smallest thing you can cut on. A cut that lands one frame late is a cut that lands late, and at 30 frames per second that is a thirtieth of a second you cannot get back without re-cutting.
In the apps: Previous Frame and Next Frame in AVPrep2 step exactly one frame at a time.
See also: Frame rate, Timecode
-
Frame rate
Also called fps, frames per second
How many frames play each second.
24 is the film look and what most narrative work uses. 30 is the North American broadcast and web default. 60 is for sport, games and anything where fast motion has to stay readable. Higher is not better. It is a different feel, and it costs you file size for every frame you add.
In the apps: AVPrep3 reads the source frame rate on import and shows it in the media bin.
See also: 23.976, Frame, Interlacing, Motion blur
-
Freeze frame
Also called hold
One frame held on screen for a length of time.
Useful for landing on a face, a document or a number. Be aware that a freeze usually carries no audio, so if the sound has to continue under it you need the audio on its own track.
In the apps: AVPrep2 encodes a freeze with no audio regardless of the keep audio setting.
See also: Freeze frame, Gap
-
Gap
Also called blank, hole
Timeline space on a track with no clip in it.
A gap on a video track exports as black. A gap on an audio track exports as silence. Both are legitimate and both are accidents about half the time, so scan the timeline for them before you export.
See also: Fade to black, Ripple
-
In point
Also called mark in
The first frame you want to keep.
Setting an in point does not delete anything ahead of it. It moves the start of what you are taking from the source, and the original file is untouched.
In the apps: Set In, shortcut I, in AVPrep2.
See also: Non-destructive, Out point, Trim
-
Motion blur
The smear within a single frame caused by movement during the exposure.
It is what makes 24fps read as motion rather than a flipbook. Footage shot with too fast a shutter has none and looks stuttery no matter what you do in the edit, and no software adds it back convincingly.
See also: Frame rate, Shutter angle
-
Optical flow
Inventing in-between frames by guessing how pixels moved.
It can make a 50 percent slowdown look smooth where frame blending would look like a double exposure. It also produces melting artefacts around anything that crosses in front of something else, so check the edges of hands and hair before you trust it.
See also: Frame rate, Speed
-
Out point
Also called mark out
The last frame you want to keep.
Most tools treat out as inclusive on screen and exclusive in the maths, which is where the one frame arguments come from. Check what your app does once and then stop worrying about it.
In the apps: Set Out, shortcut O, in AVPrep2.
-
Playhead
Also called scrubber, position indicator
The line showing which frame is being displayed.
Most keyboard edits act at the playhead, not at the mouse, which is why a cut sometimes lands somewhere you were not looking.
-
Pull-down
Spreading 24 frames across 30 frames per second by repeating fields in a pattern.
The classic 3:2 pattern is why film transferred to video has a stutter you can see in slow pans. If you are working with old broadcast masters you may need to remove it before the footage will cut cleanly.
See also: 23.976, Interlacing
-
Ripple
An edit that moves everything downstream by the same amount.
Delete five seconds with ripple on and the rest of the timeline slides five seconds earlier, so nothing after it leaves a hole. Delete with ripple off and you get a gap where the clip was. Neither is correct in general. Ripple is right when the piece is a single chain of shots. Ripple is wrong when a music bed or a title underneath has to stay where it is.
In the apps: Ripple Delete and Ripple Trim in AVPrep3, held with Shift.
-
Scrub
Dragging the playhead to move through footage by hand.
Scrubbing is for finding a moment. It is not for judging a cut, because your hand speed is not playback speed and the timing will read wrong. Find it by scrubbing, then play it at speed before you decide.
See also: Playhead
-
Shutter angle
How long the sensor was open for each frame, expressed as a fraction of the frame.
180 degrees means open half the frame time, which is the film convention and the reason 1/48th of a second is the default shutter at 24fps. Going narrower gives you the crisp stuttery look used for combat scenes. Going wider smears.
See also: Frame rate, Motion blur
-
Speed
Also called time stretch, retime
Playing a clip faster or slower than it was shot.
Slowing footage below about 50 percent starts to look like a slideshow unless it was shot at a high frame rate to begin with, because the app has no frames in between to show you. Speeding up hides that problem and creates another: audio pitch. Most tools pitch shift sped up audio unless you tell them not to.
In the apps: Fit to Length in AVPrep3 works out the speed needed to land a selection in a target duration.
See also: Frame rate, Optical flow, Pitch
-
Sync
Picture and its sound landing on the same frame.
Lip sync errors are noticed at roughly 45 milliseconds of audio late and 125 milliseconds of audio early. Human hearing is far less forgiving of sound arriving before the picture than after, because in the physical world sound always arrives second.
In the apps: AVPrep3 links a video clip and its audio lane so moving one moves the other.
See also: Drift, Linked clips
-
Timecode
An address for every frame, written hours:minutes:seconds:frames.
00:01:23:14 is one minute, twenty three seconds and fourteen frames. It is how two people on a phone agree on the same moment in a file. Note the last field counts frames, not hundredths, so what comes after :29 at 30fps is the next second, not :30.
In the apps: The AVPrep3 status bar reads out position and clip length in this form.
See also: Drop frame timecode, Frame
-
Trim
Changing where a clip starts or ends without changing what plays in between.
Trimming is the most common edit there is and the one people get wrong most often, because dragging the edge of a clip in some tools stretches it instead of shortening it. Those are different operations. A trim changes the in or out point. A stretch changes the speed.
In the apps: Dragging a clip edge in AVPrep3 trims. Speed stays where you set it.
Containers, codecs and files
-
AAC
The standard compressed audio codec for video files.
At 192 kbps stereo it is transparent for almost any spoken word or music bed. Going higher is usually wasted bytes.
-
AV1
A royalty free codec more efficient than HEVC.
YouTube and Netflix stream it. Encoding it is slow and hardware support is only now widespread. It is where things are going, not where they are.
-
Bitrate
How many bits per second the file spends. The main lever on quality and size.
Double the bitrate and you roughly double the file. Quality does not double. There is a point for any given picture where more bits buy you nothing visible, and a point below which no amount of cleverness saves you. 1080p30 talking head is comfortable around 6 to 8 Mbps. The same resolution of confetti falling needs three times that.
In the apps: The AVPrep3 export dialog exposes bitrate and reports the estimated output size.
-
CBR
Also called constant bitrate
The same number of bits every second regardless of what is on screen.
Wasteful on still shots, starved on busy ones. Its one virtue is predictable file size, which matters for broadcast and for streaming to a fixed pipe.
-
Codec
The method used to squeeze the picture or sound down, and to unsqueeze it for playback.
Coder plus decoder. H.264 is a codec. MP4 is not. If a file refuses to open, the question is almost never the extension and almost always which codec is inside.
In the apps: Both apps read the codec on import and fall back to a safe path when the browser engine cannot decode it.
-
Container
Also called wrapper
The box a file is in. MP4, MOV, MKV, AVI are containers.
A container holds streams and tells a player where each one starts. It does not decide how the picture was compressed. This is why two files can both be .mp4 and one plays everywhere while the other plays nowhere: the container matched and the codec inside did not.
-
CRF
Also called constant rate factor
Encode to a quality target instead of a size target.
You pick a number, lower meaning better, and the encoder spends whatever it needs. 18 is visually lossless for most material, 23 is the usual default, 28 is visibly soft. Use CRF when quality matters and size is negotiable. Use bitrate when size is the constraint.
-
File size
Bitrate times duration, roughly, plus a little overhead.
Eight megabits per second for sixty seconds is 480 megabits, which is 60 megabytes. If a platform caps you at 8 MB you have about 1 Mbps to spend on a minute of video, and no encoder setting changes that arithmetic. You shorten the piece, drop the resolution, or accept the softness.
In the apps: AVPrep3 estimates output size from the current settings before you commit to the render.
-
Generation loss
Quality lost each time a compressed file is re-encoded.
Compression throws away detail it judges you will not miss. Re-encoding makes that judgement again, on a picture that has already lost the easy stuff, so it starts taking things you will miss. Three or four generations of H.264 is usually where it becomes obvious on faces and gradients.
See also: Intermediate, ProRes, Transcode
-
H.264
Also called AVC, x264
The default video codec of the last twenty years.
Not the most efficient any more, but it plays on everything and encodes fast. When someone says they need an mp4, this is what they mean even if they do not know it.
-
HEVC
Also called H.265
H.264's successor. Roughly half the file size for the same picture.
The catch is licensing and support. Some browsers and older televisions will not touch it. Use it when you control the playback end, not when you are handing a file to a stranger.
-
Intermediate
Also called mezzanine
A large, lightly compressed working file used between capture and delivery.
The whole idea is to take the compression hit once at the end instead of at every step. ProRes and DNxHR are the common ones.
See also: Generation loss, ProRes
-
Keyframe
Also called I-frame, GOP
A frame stored whole. The frames between it and the next one only store what changed.
You can only cut cleanly on a keyframe without re-encoding, which is why lossless trimming tools snap your cut to somewhere near but not exactly where you asked. Two seconds between keyframes is typical.
See also: Codec, Smart render, Transcode
-
Metadata
Information about the file that is not the picture or the sound.
Creation date, camera model, rotation flag, GPS. The rotation flag is the one that bites: a phone video is often stored sideways with a flag saying to turn it, and software that ignores the flag shows it sideways.
See also: Container
-
MKV
Also called Matroska
An open container that holds anything, including many audio tracks and subtitles.
Excellent for archiving and for anything with multiple languages. Poor for handing to someone else, because a lot of consumer devices refuse it.
See also: Container
-
MOV
Apple's container. Can hold almost anything, including things Windows cannot play.
A MOV from a camera is often fine. A MOV out of a Mac editing suite may contain ProRes, which Windows will not play without extra software. Same extension, completely different file.
-
MP3
The old compressed audio format. Still everywhere.
Fine as a music bed source. Poor as a working format, because every re-encode of an MP3 loses more than the last and the losses are in the high end where speech clarity lives.
See also: AAC, Generation loss
-
MP4
The most widely playable container. Usually holds H.264 video and AAC audio.
If you do not know where a file is going, make it an MP4 with H.264 and AAC. It plays on phones, on televisions, in browsers and in every editor. Nothing else is that safe.
-
ProRes
A high quality intermediate codec that stays clean through many edits.
Files are enormous, often ten to twenty times an H.264 of the same footage. That is the point. You edit and grade in ProRes so that generation loss does not accumulate, then deliver in something small.
See also: Generation loss, Intermediate, MOV
-
Remux
Moving streams into a different container without re-encoding them.
Instant and lossless, because nothing is decoded. Pulling the H.264 out of a MKV and putting it in an MP4 is a remux. Changing that H.264 to HEVC is not.
-
Smart render
Also called stream copy, lossless cut
Copying untouched sections through without re-encoding them.
If you only trimmed the ends and changed nothing else, the middle can pass straight through, which is instant and lossless. It only works when the cut lands on a keyframe and the output settings match the source exactly.
See also: Generation loss, Keyframe, Remux, Transcode
-
Stream
Also called track, elementary stream
One continuous run of data inside a container: a video stream, an audio stream, a subtitle stream.
A file can carry several. A camera file with two microphones has two audio streams, and an editor that only imports the first one will silently drop half your sound.
-
Transcode
Decoding a file and re-encoding it into another codec.
Every transcode costs quality, even at high bitrates, because the second encoder is compressing the first encoder's mistakes along with the picture. Transcode once, at the end, if you can.
See also: Codec, Generation loss, Remux
-
VBR
Also called variable bitrate
Bits are spent where the picture needs them.
Better quality for the same average size. Two pass VBR looks at the whole file first, then allocates, which is why it takes twice as long and looks better than one pass at the same target.
-
WAV
Uncompressed audio. Big, lossless, universal.
Work in WAV. Deliver in something else. A 48 kHz 24 bit stereo WAV runs about 17 MB a minute, which is nothing on a modern disk and saves you from stacking compression artefacts.
See also: AAC, Bit depth, Sample rate
Resolution, framing and geometry
-
1080p
Also called Full HD
1920x1080, progressive.
Still the sane default for almost everything that is not a cinema screen. Every device plays it, it encodes fast, and on a phone nobody can tell it from 4K.
See also: 720p, Progressive, Resolution
-
4K
Also called UHD, 2160p
3840x2160 for television and web. 4096x2160 for cinema.
The two are not the same width, which is why a cinema 4K file letterboxes oddly on a television. When someone says 4K outside a cinema they mean 3840.
See also: 1080p, Resolution
-
720p
1280x720. The smaller high definition size.
Still perfectly watchable on a phone and about half the bits of 1080p. Worth choosing when a platform caps file size and the alternative is a soft, artefact ridden 1080.
See also: 1080p, File size, Resolution
-
Anamorphic
Footage squeezed horizontally at capture and stretched back on playback.
An anamorphic lens fits a wide image onto a normal sensor. If the desqueeze factor is not applied, everyone is tall and thin.
See also: Aspect ratio, Pixel aspect ratio
-
Aspect ratio
The shape of the frame, width to height.
16:9 is the television and web standard. 9:16 is vertical for phones. 1:1 is square. 2.39:1 is widescreen cinema. Mixing them in one timeline is fine as long as you decide deliberately what happens at the edges.
See also: Crop, Letterbox, Pillarbox, Vertical video
-
Banding
Visible steps in what should be a smooth gradient.
Sky, studio backdrops and slow fades are where it shows. Causes are 8 bit capture, heavy compression, and aggressive brightness correction, often all three. A very small amount of noise added back hides it, which is why some encoders have a grain option.
-
Bit depth
How many steps of brightness are recorded per channel.
8 bit gives 256 steps per channel, which is enough until you grade, at which point smooth skies band into visible stripes. 10 bit gives 1024 and survives correction. Most consumer cameras record 8 bit.
See also: Banding, Color space, Log
-
Crop
Throwing away part of the frame.
Cropping costs resolution. Crop a 1080 frame to half its width and you have 960 pixels to fill 1920, so it will be soft. Crop a 4K frame the same way and you still have 1920. This is the real reason to shoot bigger than you deliver.
See also: Resolution, Scale, Zoom
-
Deinterlace
Turning interlaced fields back into whole frames.
You either throw away half the lines and lose vertical detail, or you interpolate and lose sharpness. There is no free option. Do it once at the highest quality setting your tool offers.
See also: Interlacing
-
Downscale
Making the picture smaller.
Averaging four pixels into one removes noise and hides focus errors. Shooting 4K and delivering 1080 is a cheap way to look sharper and cleaner than the camera really was.
See also: Resolution, Scale
-
Interlacing
Also called the i in 1080i
Each frame is split into two fields, odd lines then even lines, captured a moment apart.
On a modern screen it shows as comb teeth on anything that moves. If you inherit interlaced footage you deinterlace it once, on import, and never think about it again.
See also: Deinterlace, Progressive, Pull-down
-
Letterbox
Black bars top and bottom when wide content sits in a less wide frame.
Correct when you want the whole picture and refuse to crop. Wrong when nobody asked and the result is a small picture in a big black field.
See also: Aspect ratio, Pillarbox
-
Noise
Also called grain
Random speckle, mostly from shooting in low light at high sensitivity.
Compression hates it, because random detail is exactly what compression cannot predict. Noisy footage at a fixed bitrate looks worse everywhere, not just in the noisy parts. Denoise before you encode and you get a better picture at a smaller size.
-
Overscan
The edge of the frame that old televisions cut off.
Largely historical. It is why safe areas exist at all and why broadcast still asks for them.
See also: Safe area
-
Pillarbox
Black bars left and right when tall content sits in a wide frame.
What you get putting a phone video on a television. Some people fill the bars with a blurred copy of the picture. That is a taste decision and it is not neutral.
See also: Aspect ratio, Letterbox
-
Pixel
One dot of the picture.
It has no size of its own. A pixel is large on a billboard and invisible on a phone, which is why resolution only means something once you know how far away the viewer is sitting.
See also: Resolution
-
Pixel aspect ratio
Also called PAR
Whether the pixels themselves are square.
Modern footage is square pixel and you can ignore this. Old DV and some broadcast formats are not, and if your software gets it wrong everyone looks either stretched or squashed by about ten percent.
See also: Aspect ratio, Resolution
-
Progressive
Also called the p in 1080p
Every frame is a whole picture.
What everything modern uses. The alternative, interlacing, exists only because 1930s broadcast needed to halve its bandwidth.
See also: 1080p, Interlacing
-
Resolution
How many pixels wide by how many tall.
1920x1080 is about two million pixels. 3840x2160 is about eight million, four times as many, and it needs roughly four times the bits to look equally good. Shooting 4K and delivering 1080 is a real technique, because downscaling averages away noise and softness. Delivering 4K because you shot 4K is usually just a bigger file.
-
Rolling shutter
Also called jello
Distortion from a sensor reading top to bottom rather than all at once.
Vertical lines lean during fast pans and propellers bend. There are corrections in software but they crop and soften, so the real fix is panning more slowly.
See also: Motion blur
-
Safe area
Also called title safe, action safe
The inner region of the frame that will survive cropping and overlays.
Title safe is about 80 percent of the frame, action safe about 90. On phone platforms the real constraint is the interface: captions, usernames and buttons sit over the bottom fifth and the right edge, so anything important there is covered.
In the apps: AVPrep3 draws title and action safe over the preview, and the platform presets adjust them.
See also: Caption, Lower third, Vertical video
-
Scale
Also called resize
Changing pixel dimensions without changing what is in frame.
Scaling down is close to free and often improves the picture. Scaling up invents pixels and always softens, whatever the marketing says about the algorithm.
-
Thumbnail
Also called poster frame
The still that represents the video before it plays.
On every platform it does more work than the first ten seconds of the edit. Pick the frame deliberately rather than letting the platform grab frame one, which is usually black.
See also: Freeze frame
-
Upscale
Making the picture bigger.
Nothing recovers detail that was never recorded. Machine learning upscalers hallucinate plausible detail, which is fine for a background plate and dangerous for anything where the detail is the point, like a document or a licence plate.
See also: Resolution, Scale
-
Vertical video
Also called 9:16, portrait
1080x1920. Taller than it is wide.
It is not a crop of a landscape shot, or it should not be. Framing for vertical means putting the subject in a tall box, and a wide shot that worked at 16:9 usually has the subject so small in 9:16 that it is useless.
In the apps: The AVPrep3 platform presets include vertical targets and show the safe area over the preview.
See also: Aspect ratio, Crop, Safe area
Color, levels and exposure
-
Alpha channel
Also called transparency
A fourth channel saying how opaque each pixel is.
Most delivery codecs do not carry it, which is why a logo exported as MP4 arrives with a black box around it. Use PNG for stills and a codec like ProRes 4444 for moving graphics.
-
Black level
How dark the darkest part of the picture actually is.
Lifted black reads as washed out or as fog. Crushed black loses detail in shadows. Broadcast usually wants true black at 16 in 8 bit video levels, and web wants 0 in full range, which is the levels problem again.
-
Chroma key
Also called green screen
Removing a background by colour.
The shoot decides whether it works. Even lighting on the screen, the subject far enough from it to avoid green spill, and a codec that did not throw away the colour detail. No software fixes a badly lit green screen.
See also: Chroma subsampling, Matte
-
Chroma subsampling
Also called 4:2:0, 4:2:2
Recording colour at lower resolution than brightness, because eyes notice brightness detail far more.
4:2:0 stores a quarter of the colour information and is what almost every delivery file uses. It is invisible on normal footage and very visible on sharp coloured text or a green screen edge, which is why keying wants 4:2:2 or better.
See also: Bit depth, Chroma key, Codec
-
Clipping
Also called blown highlights, crushed blacks
Detail lost because values went past the top or bottom of the range.
Once clipped, it is gone. No slider recovers it, because there is nothing recorded to recover. A blown window is white pixels, not a faint image of a window.
-
Color correction
Making footage accurate and making shots match each other.
Mostly white balance, exposure and contrast. The test is a face and a white object. If the face reads as that person's skin and the white thing reads white, you are close enough to start grading.
See also: Grade, Levels, White balance
-
Color space
The agreed set of colours a file is allowed to describe, and what its numbers mean.
Rec.709 is the standard for HD video and the safe answer for anything web. sRGB is its computer graphics cousin and close enough that people treat them as the same. Rec.2020 is the wide space used for HDR. Tag your file wrong and the picture is not broken, it is just the wrong colour, usually too saturated or too flat.
-
Color temperature
Also called Kelvin
How warm or cool a light source is, measured in Kelvin.
Confusingly, lower numbers are warmer. 3200K is orange candlelight territory, 5600K is noon daylight, 7000K is overcast shade and looks blue.
See also: White balance
-
Exposure
How much light reached the sensor.
Underexposed footage lifted in the edit brings up noise with it. Overexposed footage has nothing to recover, because clipped white is just white. When in doubt, protect the highlights and lift in post.
-
Gamma
The curve between the recorded number and the brightness you see.
It is not a straight line, because human vision is not linear and we can see far more steps in the dark than in the bright. Mismatch the gamma between what you graded on and what the viewer plays on and the midtones shift, usually darker on a Mac and lighter on a phone.
-
Gamut
The range of colours a space can actually describe.
Colours outside the gamut get clipped to the nearest thing inside it, which is why a vivid red neon sign turns into a flat orange blob. Wider gamut does not mean better, it means more room.
See also: Clipping, Color space
-
Grade
Also called color grading
Shaping the colour and contrast of the whole piece for effect.
Distinct from correction, which is making shots match and look accurate. Correct first so the shots agree with each other, then grade so the piece has a look. Doing it the other way round means regrading every time you fix a shot.
See also: Color correction, LUT
-
HDR
Also called high dynamic range
Video that describes a much wider range from black to peak white.
It needs a display that can show it, a colour space that can carry it, and metadata that survives every step. Get any of those wrong and HDR looks worse than plain video, not better. For web delivery it is still more trouble than it is worth unless you have a reason.
See also: Color space, Rec.709, Tone mapping
-
ISO
Also called gain, sensitivity
How much the camera amplifies the sensor signal.
Higher ISO gets you a usable picture in the dark and pays for it in noise. Most cameras have one or two native ISO values where they are cleanest, and it is worth knowing yours.
-
Levels
Also called range, limited vs full
Whether black is 0 or 16, and white is 255 or 235, in 8 bit video.
Video traditionally uses 16 to 235 and leaves headroom either side. Computers use 0 to 255. Get the tag wrong and you see it instantly: either the picture is washed out and grey, or the shadows are crushed to solid black and the highlights are blown. This single mismatch causes more "my export looks wrong" than everything else combined.
See also: Black level, Color space, Gamma
-
Log
Also called S-Log, C-Log, V-Log
A flat, low contrast recording curve that keeps more range for grading.
Log footage looks grey and wrong straight out of the camera. That is correct. It is holding detail in the highlights and shadows that a normal curve would have thrown away. It has to be graded, and it needs 10 bit to survive the grade without banding.
-
LUT
Also called lookup table
A table that maps every input colour to an output colour.
A technical LUT converts log to Rec.709 correctly. A creative LUT is a look someone designed. Applying a creative LUT to footage it was not built for gives you a mess, so convert first and style second.
See also: Color space, Grade, Log
-
Matte
A greyscale image saying which parts of a frame are visible.
White shows, black hides, grey is partial. Every key, mask and transparency is a matte underneath, whether the software calls it that or not.
See also: Alpha channel, Chroma key, Overlay
-
Rec.709
The colour standard for high definition video.
If you do not know what your delivery needs, this is the answer. Everything plays it correctly and nothing surprises you.
See also: Color space, Gamut
-
Saturation
How intense the colours are.
Pushing it globally is the most common amateur move and the first thing that reads as cheap, because skin goes orange before anything else gets interestingly vivid. Push specific hues instead.
See also: Vectorscope, Vibrance
-
Tone mapping
Squeezing a wide range picture into a narrower one.
How HDR gets shown on a normal screen. Done automatically it tends to look flat and grey. Done deliberately it can look fine. If you deliver HDR, check what it looks like tone mapped, because most of your audience sees that version.
-
Vectorscope
A circular graph of colour, showing hue as angle and saturation as distance from centre.
The skin tone line running toward the upper left is the useful part. If faces sit on it, the white balance is close, whatever the picture looks like on your screen.
See also: Saturation, Waveform, White balance
-
Vibrance
Saturation that leaves already saturated colours alone.
Safer than saturation for exactly the reason above: it lifts the dull parts of the frame and mostly spares faces.
See also: Saturation
-
Waveform monitor
A graph of brightness across the frame.
The only honest way to judge exposure, because your eyes adapt and your monitor lies. Read the bottom for black level and the top for clipping. Faces usually want to sit between 55 and 75 percent.
See also: Clipping, Levels, Vectorscope
-
White balance
Telling the camera what colour the light is, so white comes out white.
Daylight is blue, about 5600K. Tungsten is orange, about 3200K. Office fluorescent is green and nobody likes it. Shoot with the wrong setting and everything carries a cast, which is fixable in 10 bit and a fight in 8 bit.
See also: Color correction, Color temperature
Sound, levels and loudness
-
Bit depth (audio)
How finely each sample is measured. 16 bit for delivery, 24 bit for working.
24 bit gives you room to record quietly and lift later without hearing the noise floor come up. 16 bit is fine as a finished file and is what a CD is.
See also: Headroom, Noise floor, Sample rate
-
Channel
One independent stream of sound. Mono is one, stereo is two.
A single microphone recorded to a stereo file with silence in the right channel plays out of one speaker on half the world's devices. Check which channel your audio is really in.
-
Clipping (audio)
Signal driven past 0 dBFS, so the tops of the waveform are cut flat.
It sounds like a harsh buzz on loud syllables. Declippers guess at the missing shape and are worth trying once, but the only real fix is recording it again with the gain down.
-
Compression (audio)
Turning down the loud parts so the whole thing can be turned up.
Nothing to do with file compression. A 3:1 ratio at a -18 threshold means anything above -18 only rises a third as fast as it otherwise would. On speech it is the difference between a voice that disappears when someone turns away and one that stays present.
See also: Ducking, Dynamic range, Limiter
-
dBFS
Also called decibels full scale
The digital loudness scale. 0 is the ceiling and everything is negative.
There is nothing above 0. Go past it and the waveform is chopped flat, which sounds like crackle and cannot be undone. This is why people record peaking around -12 and leave the top alone.
See also: Clipping (audio), Headroom, Peak
-
Denoise
Also called noise reduction
Removing steady background noise by learning its shape.
Push it too far and the voice goes underwater and metallic, because the algorithm cannot tell noise from the quiet parts of speech. Take out 6 to 10 dB and live with the rest.
In the apps: Both apps expose a denoise toggle per clip.
See also: Noise floor
-
Dither
A tiny amount of deliberate noise added when reducing bit depth.
It sounds wrong until you hear the alternative, which is a gritty distortion on quiet fades. The same trick in picture is what hides banding.
See also: Banding, Bit depth (audio), Noise
-
Ducking
Also called sidechain
Automatically lowering music whenever someone speaks.
About 9 to 12 dB of duck is usually right for narration over a bed. Less and the voice fights the music. More and the music pumps distractingly. A slow release, half a second or so, stops the bed jumping back up between words.
In the apps: AVPrep3 has a ducking control on the music clip that follows the speech track.
See also: Compression (audio), Mix, Music bed
-
Dynamic range
The distance between the quietest and loudest parts.
Wide is beautiful in a quiet room and useless in a car. If your audience listens on a phone speaker in public, compress harder than feels right on headphones.
See also: Compression (audio), LUFS
-
EQ
Also called equalisation
Turning specific frequency ranges up or down.
For speech, the useful moves are a high pass around 80 Hz to remove rumble, a small cut around 200 to 400 Hz if the voice sounds boxy, and a gentle lift above 3 kHz for clarity. Boosting is easy and usually wrong. Cut first.
In the apps: AVPrep3 carries bass, mid and treble per clip.
See also: Denoise, High pass filter
-
Fade
Volume ramping up from or down to silence.
A two or three frame fade on every audio cut removes the click that a hard cut through a waveform makes. It is inaudible as a fade and very audible as a click if you skip it.
See also: Crossfade, Zero crossing
-
Gain
How much a signal is amplified, before anything else happens to it.
Distinct from volume, which is what you do at the end. Set gain at the source so the signal arrives healthy, then use volume for balance.
See also: Headroom, Noise floor
-
Headroom
The space you leave between your loudest moment and the ceiling.
Record with 12 to 18 dB of it. You cannot predict the laugh, the door slam, or the person who leans into the microphone, and headroom is the only thing that saves those moments.
See also: Clipping (audio), dBFS
-
High pass filter
Also called low cut
Removing everything below a chosen frequency.
The single most useful thing you can do to a voice recording. Traffic, air handling, handling noise and footsteps all live below 80 Hz and none of it carries speech.
See also: EQ, Noise floor
-
J cut
The next shot's sound starts before its picture.
You hear the new scene before you see it, which pulls the audience forward. Named for the shape the two clips make on a timeline.
-
L cut
The outgoing shot's sound continues over the next picture.
You stay with a speaker's voice while cutting to the listener's reaction. It is the most useful edit in interview work and it is what makes conversation scenes feel natural rather than like a tennis match.
-
Limiter
Compression with a ratio so steep nothing gets past the ceiling.
The last thing in the chain. Set it to catch true peaks at -1 dBTP and it protects you from the one shout you did not plan for.
See also: Clipping (audio), Compression (audio), True peak
-
LUFS
Also called loudness units full scale, LKFS
The standard measure of how loud something actually sounds to a person.
Platforms normalise to a target and turn your audio down if you exceed it, so mastering louder than the target gains you nothing and costs you dynamics. Roughly: YouTube about -14, Spotify about -14, broadcast about -23 or -24. Spoken word pieces usually sit comfortably at -16 to -14 integrated.
In the apps: The AVPrep3 platform presets carry a loudness target per destination.
See also: Dynamic range, Normalize, True peak
-
Mono
One channel, the same in both ears.
The right choice for a single speaking voice. It is louder for the same level, it never has phase problems, and nobody ever complained that narration was not in stereo.
-
Music bed
Also called underscore
Music running under narration to carry energy between points.
It should be felt and not listened to. Instrumental, no sudden hooks, and nothing with a vocal, because two voices at once means the audience gets neither.
-
Noise floor
The constant quiet hiss and hum underneath everything.
Room tone, preamp hiss, air conditioning. It becomes a problem when you lift a quiet recording, because the floor comes up with the voice. Recording closer to the source is the fix. Software noise reduction is the apology.
-
Normalize
Scaling a whole file so its loudest point, or its overall loudness, hits a target.
Peak normalising to -1 dBFS makes the loudest instant loud and tells you nothing about how the piece sounds. Loudness normalising to a LUFS target is what you actually want for delivery.
In the apps: AVPrep3 offers both peak and loudness normalisation per clip.
See also: Compression (audio), LUFS, Peak
-
Pan
Placing a sound between left and right.
Keep narration centre. Pan ambience and music wide. Hard panning anything important means half your audience on one earbud misses it.
-
Peak
The loudest single instant.
Peaks do not tell you how loud something sounds. A snare hit and a sustained voice can peak identically and sound nothing alike. Peak tells you whether you are about to clip, and nothing else.
-
Phase
Whether two copies of a sound line up or fight each other.
Two microphones on one source, slightly different distances, and the combined result is thin with a hollow middle. Flipping the polarity of one sometimes fixes it. Moving the microphone always does.
-
Pitch
How high or low a sound reads.
Speeding a clip up raises it and slowing it down lowers it, because both change how fast the waveform runs. Pitch correction fixes that and costs some quality, which is audible on music long before it is audible on speech.
See also: Sample rate, Speed
-
Plosive
The thump from P and B sounds hitting the microphone.
A pop filter prevents it. A high pass filter reduces it. Nothing removes it cleanly once recorded, so it is worth the four dollar screen.
See also: High pass filter
-
RMS
Average level over a short window. Closer to perceived loudness than peak.
Superseded for delivery by LUFS, which weights the average the way hearing does, but still a useful reading for how dense a mix is.
-
Room tone
Thirty seconds of the room making no particular sound.
Record it at every location. It is what you lay under a gap so that a cut does not fall into absolute digital silence, which sounds like the file broke.
See also: Noise floor
-
Sample rate
How many times a second the sound was measured. 48 kHz is the video standard.
44.1 kHz came from CD. 48 kHz came from video and is what every camera and every editor expects. Mixing the two in one project causes pitch and sync errors if anything resamples badly, so convert on import and stay at 48.
See also: Bit depth (audio), WAV
-
Sibilance
Also called ess
Harsh S and T sounds, usually between 5 and 8 kHz.
Made worse by close microphones and by boosting the high end for clarity. A de-esser is a compressor that only listens to that band.
See also: Compression (audio), EQ
-
Stereo
Two channels carrying a left and right image.
Right for music and ambience. Check any stereo mix in mono before delivering, because phase cancellation between the channels can make a part vanish entirely on a phone speaker.
-
True peak
Also called dBTP
The real peak of the analogue waveform between the digital samples.
A file can read -0.1 dBFS and still distort on playback, because reconstruction overshoots between samples. Delivery specs usually ask for -1 dBTP, which gives the decoder room.
See also: Clipping (audio), dBFS, LUFS
-
Waveform
The picture of loudness over time on an audio track.
You can read a lot from it: where the speech is, where the breaths are, whether it clipped. Editing dialogue by eye on the waveform and then checking by ear is far faster than hunting by ear alone.
-
Zero crossing
The point where a waveform crosses silence on its way up or down.
Cutting there means both sides start from nothing and there is no step, so no click. Snapping cuts to zero crossings is why some tools move your edit a few milliseconds.
See also: Fade
Editing craft
-
180 degree rule
Keep the camera on one side of an imaginary line between two subjects.
Cross it and the two people appear to swap sides, which reads as them turning away from each other. Break it deliberately to disorient. Break it accidentally and the scene is confusing and nobody can say why.
See also: Continuity, Eyeline
-
Autosave
The application writing a copy of your project on a timer.
Check where yours writes and how many it keeps before you need it. A rolling set of ten, two minutes apart, has saved more work than any other feature in editing software.
In the apps: AVPrep3 keeps rolling autosaves in the portable data folder beside the executable.
See also: Project file
-
B roll
Also called cutaway
Footage other than the main shot, laid over the main audio.
It does three jobs at once: it shows what is being talked about, it hides edits in the interview, and it gives the eye somewhere to go. Shoot three times more of it than you think you need.
-
Burn-in
Rendering text permanently into the picture.
Necessary where the platform will not carry a subtitle track. The cost is that the text cannot be turned off, translated, or fixed without re-exporting.
-
Caption
Also called subtitle, burned-in text
A text transcript of the speech, on screen.
Most social video is watched with the sound off, so captions are not an accessibility extra there, they are the primary channel. Two lines maximum, about 40 characters a line, and break lines at natural phrases rather than wherever the text ran out.
In the apps: AVPrep2 wraps caption lines at 42 characters.
See also: Accessibility, Legibility, SRT
-
Close-up
Head and shoulders or tighter.
Where emotion lives. Also where poor focus, bad lighting and every skin detail become obvious, so it is the least forgiving frame you can choose.
See also: Focus, Medium shot
-
Continuity
Things staying consistent across a cut: position, direction, light, props.
A glass that is full then empty pulls the audience out of the piece for a second, and that second is gone. In documentary and interview work the common break is a cut where the subject's head jumps position, which is what B roll is for.
-
Crossfade
Also called dissolve
One shot fading out as the next fades in, overlapping.
Half a second is the usual length. Long dissolves read as dreamlike or as filler. The thing people forget is that a crossfade needs extra frames on both clips to blend, so a shot trimmed to the exact frame you want has nothing left to give the transition.
In the apps: The AVPrep3 crossfade takes frames from either side of the join, which is why it needs both clips to be longer than half the transition.
See also: Crossfade, Handles, Transition
-
Cut
An instant change from one shot to the next.
The default and almost always the right answer. A cut is invisible when the audience is following something across it: a movement, a question, a line of thought. It is visible and jarring when nothing carries across, which is why bad cutting feels random rather than fast.
See also: Continuity, J cut, Transition
-
Depth of field
How much of the scene in front of and behind the subject is acceptably sharp.
Wider aperture, longer lens and closer subject all make it shallower. Phone cameras have very deep depth of field, which is why their portrait modes fake the blur.
-
Dropped frames
Frames the playback engine could not deliver in time, so it skipped them.
A preview problem, not a file problem. It means your machine cannot keep up with what you are asking it to composite live. Lower the preview resolution or use proxies. The export is unaffected.
-
Establishing shot
A wide shot that tells the audience where they are.
Without one, a tight shot of two people talking could be anywhere, and the audience spends the first ten seconds working out the room instead of listening.
See also: Continuity, Wide shot
-
Eyeline
Where the subject is looking, and whether it matches across shots.
In an interview, keep the subject looking just off the lens on the same side in every setup, or cutting between them reads as two separate conversations.
See also: 180 degree rule, Continuity
-
Fade to black
Picture ramping down to black.
Punctuation. It ends a section the way a paragraph break ends a paragraph, and using it between every shot is like ending every sentence with a full stop and a blank page.
See also: Fade to black, Transition
-
Focus
Which distance from the camera is sharp.
Shallow focus separates a subject from a busy background and punishes any movement toward or away from the camera. Nothing in software recovers a soft shot, so it is the one thing worth checking on a monitor at the time.
See also: Close-up, Depth of field
-
Handles
Extra frames kept beyond your in and out points.
A second at each end. They cost nothing and they are the difference between being able to adjust a cut later and having to go back to the source. Every transition eats them.
-
Headroom (framing)
The space between the top of the head and the top of the frame.
Too much and the subject sinks. Too little and they look cramped. The eyes want to sit near the upper third line, which usually gives you the right amount without measuring.
See also: Lead room, Rule of thirds
-
Jump cut
Cutting two nearly identical shots together so the subject appears to jump.
An error in narrative work and a style in vlogging. If you did not choose it, cover it with B roll or a cutaway. If you did choose it, commit and do it often enough that it reads as intentional.
See also: B roll, Continuity
-
Lead room
Also called looking room, nose room
Space in front of where the subject is looking or moving.
Without it the subject looks like they are about to walk into the edge of the frame. With it, the composition breathes.
See also: Eyeline, Rule of thirds
-
Linked clips
A video clip and its audio moving together as one.
Keeps sync while you rearrange. Unlink deliberately when you want a J or L cut, and relink when you are done, or something will slide out of sync later and you will not notice until export.
-
Lower third
A name and title graphic in the bottom portion of the frame.
On for four to six seconds, long enough to read twice. Keep it out of the bottom fifth on phone platforms, where the interface covers it.
-
Medium shot
Roughly waist up.
The workhorse of interviews and presenting. Close enough to read expression, wide enough to show gesture.
-
Mix
Balancing everything that makes sound so the piece is intelligible.
Order of operations: get speech clear first, then set music under it, then place effects. Doing it in any other order means redoing the speech when the music arrives.
-
Montage
A sequence of shots compressing time or making an argument by accumulation.
It works on rhythm rather than continuity, which is why a music bed usually carries it. Cut on the beat and it flows. Cut off the beat for the whole montage and it feels wrong even to people who cannot name why.
-
Non-destructive
Editing that never changes the original files.
Your project file holds decisions, not media. This is why a project folder is small and why moving the source files breaks it. It is also why you can always go back.
See also: Offline media, Project file
-
Offline media
Also called missing media
Source files the project can no longer find.
Caused by moving, renaming or unplugging. Keep media in a folder next to the project and move the whole folder as one thing.
See also: Non-destructive, Project file
-
Overlay
Any element composited above the main picture.
Logos, lower thirds, watermarks, graphics. Whatever sits on top has to survive compression and has to respect the safe area.
See also: Alpha channel, Safe area, Watermark
-
Pacing
Also called rhythm
How long shots are held and how that length changes.
Uniform shot length is the most common fault in a first edit. Vary it. Hold the important thing longer than feels comfortable, and cut away from the unimportant thing sooner than feels polite.
-
Preview
Playing the timeline back without writing a file.
Preview quality is a trade. Half resolution plays smoothly and lies to you about sharpness. Full resolution tells the truth and may stutter. Judge timing at half and judge detail at full.
In the apps: AVPrep3 composites preview in the renderer and never starts an encoder during playback.
See also: Dropped frames, Proxy, Render
-
Project file
The file holding your edit decisions, references and settings.
It is not your video. Losing it while keeping the sources means redoing the work. Back it up more often than you back up footage, because it is tiny and irreplaceable.
See also: Autosave, Non-destructive, Offline media
-
Proxy
A small low quality copy used for editing, swapped for the original at export.
How people cut 4K on laptops. The risk is forgetting to switch back, and exporting the proxies.
See also: Intermediate, Preview, Transcode
-
Render
Computing the finished frames and writing them to a file.
Distinct from preview, which computes frames on the fly and throws them away. Render times scale with resolution, effects and codec, and a render that takes longer than the piece itself is normal for anything with much processing on it.
-
Rule of thirds
Placing the subject a third of the way in rather than dead centre.
A guideline, not a law. It works because it leaves room for the subject to look or move into. Centre framing is strong and deliberate when the subject is addressing the camera.
In the apps: AVPrep3 can overlay a thirds grid on the preview.
See also: Headroom (framing), Lead room, Safe area
-
Split
Also called razor, cut tool
Dividing one clip into two at the playhead.
Both halves still reference the same source file. Nothing is duplicated and nothing is deleted, which is why splitting and then deleting one half is the normal way to remove a middle section.
In the apps: AVPrep3 has a razor tool. AVPrep2 does not, and gets the same result with two in and out pairs.
-
SRT
A plain text subtitle file: a number, a timecode range, the text, a blank line.
Readable in any text editor and accepted by every platform. Keep captions as an SRT rather than burning them in when you can, because the viewer can then turn them off and the platform can translate them.
-
Timeline
The horizontal workspace where clips are arranged in time.
The thing that makes an editor an editor rather than a converter. Position on the timeline is when something happens, and which track it is on decides what covers what.
See also: Non-destructive, Playhead, Track
-
Title
Text on screen.
Legibility beats style every time. High contrast, a weight heavy enough to survive compression, and a size that works on a phone held at arm's length. If you cannot read it on your own phone, nobody can.
In the apps: AVPrep3 puts a title on its own track so it can sit over any shot.
See also: Caption, Legibility, Lower third
-
Track
Also called lane
A horizontal row on the timeline holding clips.
Video tracks stack, and higher tracks cover lower ones. Audio tracks sum, so everything on them plays at once. That difference is the whole logic of a timeline.
-
Transition
Anything other than a straight cut between two shots.
Every transition says something. A dissolve says time passed. A fade to black says a section ended. A wipe says this is a different kind of thing entirely, and in most serious work it says the editor had a toolbar. Use a cut unless you can say what the transition means.
In the apps: AVPrep3 keeps clips edge to edge and straddles the join, so adding a crossfade does not move anything.
See also: Crossfade, Cut, Fade to black
-
Watermark
A persistent mark identifying the owner.
Small, low opacity, in a corner that no interface element covers. If it distracts from the content it is too big. If it is easy to crop off it is in the wrong corner.
In the apps: Both apps can pin a watermark so it stays put across the whole timeline.
-
Wide shot
Also called long shot
Subject small in frame, surroundings visible.
Shows relationships and space. Poor for emotion, because you cannot read a face at that size, and poor on phones for the same reason.
See also: Close-up, Establishing shot, Medium shot
-
Zoom
Also called Ken Burns, push in
Scaling into the frame over time.
On a still photograph it is the difference between a slideshow and a sequence. Keep the move slow and constant, start and end on rest, and never zoom past the point where the source runs out of pixels.
See also: Crop, Freeze frame, Scale
Delivery and publishing
-
Accessibility
Making the piece usable by people who cannot hear it or cannot see it well.
Captions for speech, adequate contrast on text, and a transcript alongside. It is a legal requirement in a lot of contexts and a practical one everywhere, since most viewers have the sound off.
See also: Caption, Legibility, SRT
-
Archive master
The highest quality version you keep after delivery.
Keep the project file, the sources, and one high quality flat export. In five years the platform version will be gone and the sources may not open, but the flat master will.
See also: Generation loss, Intermediate, Project file
-
Aspect fit vs fill
Whether the whole picture is shown with bars, or the frame is filled and the edges lost.
Fit keeps everything and adds black. Fill keeps no black and crops. Pick per piece. Filling a 16:9 interview into 9:16 usually cuts the subject's head in half.
See also: Crop, Letterbox, Vertical video
-
Aspect of the upload
Also called what the platform does to it
Every service re-encodes, normalises loudness and may crop.
Assume your file is a suggestion. Give it clean input, keep important things inside the safe area, and master to the platform's loudness target so its normaliser does the least.
See also: LUFS, Platform specs, Safe area
-
Checksum
Also called hash, SHA-256
A short fingerprint of a file's contents.
Two files with the same checksum are byte for byte identical. It is how you prove a download arrived intact and how you prove an archive has not changed.
See also: Archive master
-
Export
Also called output, publish
Writing the finished timeline to a file.
The settings here decide everything the viewer sees, and the defaults in most software are a compromise nobody chose. Know your destination before you open the dialog.
In the apps: AVPrep3 exports in three passes: normalise each clip, composite, then mix and encode.
-
Faststart
Also called web optimised, moov atom
Moving the file's index to the front so it can start playing before it has downloaded.
Without it, a browser has to fetch the whole file before showing frame one. It is a one checkbox difference and it is the reason some self hosted videos seem to take forever to start.
-
Hardware encoding
Also called NVENC, QuickSync, AMF
Using a dedicated chip on the graphics card to encode.
Many times faster and slightly worse at the same bitrate. Use it for drafts and for long recordings. Use software encoding for the final file if quality per byte matters.
-
Legibility
Whether text on screen can actually be read.
Test on the smallest screen anyone will use. Compression eats thin strokes and low contrast, so what looks elegant in the editor can turn to mush at 2 Mbps.
-
Platform specs
What a given service actually accepts and what it does to your file.
Every platform re-encodes what you upload. Your job is to give it the cleanest input you can within its limits, not to match its output bitrate. Upload higher quality than the platform delivers and let its encoder do the reduction.
-
Portable application
Software that runs from a folder without installing.
Everything it needs, including its settings, lives beside the executable. Copy the folder to another machine and it runs with your presets intact. Nothing is written to the system.
In the apps: AVPrep2 and AVPrep3 both run this way, with a data folder next to the executable.
See also: Archive master
-
Preset
A saved bundle of export settings for one destination.
The point is repeatability. If the same settings produce a file that works on a platform once, saving them means you never rediscover it under deadline.
In the apps: Both apps ship platform presets and let you save your own.
See also: Export, Platform specs
-
Streaming
Delivering video in pieces as it plays rather than as one download.
Adaptive streaming switches quality as bandwidth changes, which is why a video goes blurry then sharpens. It needs the file prepared as segments at several bitrates, which is a step beyond exporting an MP4.
See also: Bitrate, Faststart, Platform specs
-
Two pass
Encoding once to analyse, then again to allocate bits where they are needed.
Worth the doubled time when you have a hard size limit. Pointless when you are encoding to a quality target, because CRF already spends bits where they matter.