Video Formats and Codecs

Picture one minute of a dancer crossing an open-air stage. In the format I use for most of my video on the Z6III, that minute fills about 1.4 GB of card by Nikon's figures. In the camera's largest raw format, the same minute needs nearly twenty times the card, and it plays only in software made for it (p. 214). Both files hold the same dance. They differ in how much the camera throws away, how finely it counts each shade, and how the result is packed.

This chapter covers the three menu items behind those differences, which modes crop, and the format I use. Page numbers in parentheses point to Nikon's Z6III Reference Guide, edition 09, which matches firmware 2.00.

01.What a Video File Holds

Codec and Container

A parcel is a fair model of a video file. The box is the container, the file format whose ending you see on a computer: .NEV, .MOV or .MP4 on the Z6III (p. 1004). The packing inside is set by the codec, short for coder-decoder, the recipe that squeezes pictures while recording and rebuilds them for playback. One container carries pictures and sound side by side, and a player must understand both the box and the packing.

Nikon's option names state these facts in order. In H.265 10-bit (MOV), H.265 is the codec, 10-bit is the bit depth and MOV is the container. MOV is Apple's QuickTime container and NEV is Nikon's own, used only for N-RAW (p. 197). Sound is Linear PCM (pulse-code modulation, stored without compression) in NEV and MOV files and AAC (Advanced Audio Coding, a compressed format) in MP4 files (p. 1004), and Audio and Timecode covers it.

Why Compression Shrinks Files

A 4K frame holds $3840 \times 2160 = 8{,}294{,}400$ pixels. At three colour values of 10 bits each, 30 bits a pixel, thirty frames a second would need

$$8{,}294{,}400 \times 30 \times 30 \approx 7{,}465{,}000{,}000 \text{ bits per second},$$

or about 7,465 Mbps, megabits (millions of bits) per second. That figure is illustrative, since no Z6III format stores pictures this way. The format I use records 4K at 30p at about 190 Mbps (p. 207), keeping roughly one bit in $7465 / 190 \approx 39$.

Codecs save in two places. Inside a frame, a stretch of clear sky can be described as one gentle gradient, and fine detail the eye barely sees can be coarsened. Between frames the saving is larger. Film a dancer before a still backdrop and the backdrop repeats frame after frame, so a codec stores one complete frame, then only what changed in the frames that follow.

A complete frame is an intra frame, and the run of frames leaning on it is a group of pictures, or GOP. Long GOP compression, which Nikon also calls inter-frame, puts many dependent frames between complete ones and keeps files small. All-I, short for all intra, makes every frame complete, so files grow but any frame decodes alone, which makes it easier for an editing computer to cut and scrub. On the Z6III, ProRes 422 HQ is All-I and the H.265 and H.264 types are long GOP (p. 198). The RG does not describe the raw types this way.

H.265, or HEVC (High Efficiency Video Coding), is the newer standard and was designed to need fewer bits than H.264, or AVC (Advanced Video Coding). Yet the Z6III gives H.265 8-bit about 80 Mbps at 1920 by 1080 and 60p against about 50 for H.264 (p. 208), which suggests Nikon spends the efficiency on quality rather than size.

Bit Depth and Tonal Steps

Picture a slice of evening sky brightening smoothly across the frame. A file stores each colour channel as a whole number, and $n$ bits give $2^n$ possible levels. Eight bits give $2^8 = 256$, numbered 0 to 255. Suppose the slice spans only eight of them, 160 to 167. Every point of the smooth gradient is rounded to one of those eight, so the sky is stored as eight flat bands. Ten bits give $2^{10} = 1{,}024$ levels, four times as many, so the same slice spans $8 \times 4 = 32$ values, 640 to 671, and is stored as 32 narrower bands.

The bands matter once contrast is added later. Shoot the slice with the Flat Picture Control, which records low contrast on purpose, then give it four times the contrast in a finished video shown at 8 bits. The slice should now cover $8 \times 4 = 32$ display levels. An 8-bit recording has eight values for those 32 levels, so each step jumps $32 / 8 = 4$ levels and the sky shows stripes, a fault called banding. A 10-bit recording has one value per level, and the sky stays smooth.

Each extra bit doubles the levels, so 12 bits give $2^{12} = 4{,}096$. Across three channels the gap multiplies: $256^3 = 16{,}777{,}216$ colours against $1024^3 = 1{,}073{,}741{,}824$. The Z6III records 8 bits in its H.264 type and the 8-bit H.265 type, 10 bits in ProRes 422 HQ and the 10-bit H.265 type, and 12 bits in both raw types (p. 1004).

Colour Sampling

Most video stores YCbCr rather than red, green and blue. Y is each pixel's brightness, and Cb and Cr say how far its colour leans towards blue or red. The eye resolves brightness more finely than colour, so formats share colour between neighbours. Take a two-by-two block of four pixels. In 4:4:4 each pixel keeps its colour, so the block stores 4 brightness and 8 colour values, 12 in all. In 4:2:2 side-by-side pairs share colour, giving $4 + 4 = 8$. In 4:2:0 all four share, giving $4 + 2 = 6$, half of 4:4:4.

The loss rarely shows, except when colour is pushed hard, as when lifting a subject off a green screen. ProRes 422 HQ records 4:2:2 and the H.265 and H.264 types record 4:2:0 (p. 198). The raw types keep the sensor's Bayer data instead, one reading per photosite under its colour filter, for conversion later (p. 197).

02.RAW Video and Its Proxy

A NEF photograph holds the sensor's readings before the camera turns them into a picture, and raw video does the same for each frame of a clip. The Z6III offers Nikon's N-RAW 12-bit (NEV) and Apple's ProRes RAW HQ 12-bit (MOV), and Nikon expects both to be processed and edited on a high-performance computer of the kind video professionals use (p. 197). Few devices play such files, so the camera writes a proxy alongside, a light H.264 8-bit MP4 at 1920 by 1080 for review on the camera (p. 197). The proxy shares the raw clip's name with an .MP4 ending (p. 455).

Raw takes things away. The ISO settings Hi 0.3 to Hi 2.0 vanish. Electronic VR, the camera's digital stabilisation for video (VR means vibration reduction), is unavailable, and so are Skin softening, High ISO NR, Active D-Lighting and Diffraction compensation. Digital flicker reduction is off, clips cannot be edited in the camera, and HDMI output is limited to 1920 by 1080 (p. 214). For raw, the frame size, not the Image area item, picks the FX or DX area (p. 210). Raw will not record to SD cards or to XQD cards of 32 GB or less (p. 212), and Nikon keeps a list of cards it recommends for it (p. 1011), discussed in Card Slots and File Handling.

I do not record raw video. Raw repays its size in a colour grade, but I post footage as shot, and my Lexar cards are not on Nikon's raw list. For a reader with both, raw keeps the most the sensor can give.

03.Video Quality (N-RAW)

This item offers High quality, the default (p. 578), and Normal (p. 586), and it governs N-RAW alone (p. 197). Normal halves the stream. At FX 6048 by 3402 and 30p, High quality averages about 1,870 Mbps and Normal about 940 (pp. 205-206), and $940 / 1870 \approx 0.50$, with the other rows halving likewise. Nikon does not say what Normal costs in the picture. With card space to spare, keep High quality.

04.Video File Type

This item picks the codec, bit depth and container in one step, and its default is H.265 8-bit (MOV) (p. 578). The table sets out the six options (pp. 197-199).

The tone mode is chosen by highlighting one of the four types that allow it and pressing right on the multi selector (p. 199), and I found no default tone mode printed. SDR, standard dynamic range, is the ordinary look. HLG, Hybrid Log Gamma, is a high dynamic range (HDR) format for screens with brighter highlights, and N-Log is Nikon's flat log recording, made to be graded, meaning colour and contrast are set later on a computer. N-Log, HLG and Exposure Tools explains each.

ProRes RAW HQ 12-bit (MOV) lacks N-RAW's fastest modes, as the raw table below shows (p. 200), and the RG prints no bit rates for it (p. 205).

ProRes 422 HQ 10-bit (MOV) is Apple's editing format, which Nikon aims at post-production, the work after shooting (p. 198). All-I and 4:2:2 make it easy to cut and tolerant of colour work, at a price. Apple's ProRes white paper targets 884 Mbps for it at 3840 by 2160 and 30p, against about 190 Mbps for H.265 10-bit (p. 207), so it needs $884 / 190 \approx 4.7$ times the card per minute. Apple's figure is a codec target, not a Z6III measurement. On cards of 32 GB or less a ProRes 422 HQ clip ends at 4 GB (p. 212).

H.265 10-bit (MOV) is also meant for editing on a capable computer (p. 198), and it is the only type with all three tone modes (p. 199). Long GOP and 4:2:0 keep it near 190 Mbps at 4K 30p (p. 207). The catch is on p. 288: the camera's editing options and Playback speed apply only to clips recorded as H.264 8-bit (MP4) or H.265 8-bit (MOV). On the camera, then, a 10-bit clip cannot be trimmed, have frames saved, or play at a slowed Playback speed. The playback i menu says a chosen speed governs all video playback (pp. 240-241), but Nikon is silent on whether a speed set on an 8-bit clip carries to a 10-bit one.

H.265 8-bit (MOV), the default, records SDR only (p. 199), and Nikon praises its compression (p. 198). At about 150 Mbps for 4K 30p (p. 207) it makes the smallest 4K files on offer. H.264 8-bit (MP4), which Nikon calls widely supported (p. 198), is limited to 1920 by 1080 (pp. 201-202) and writes no timecodes, the running clock stamps that line up cameras and recorders (p. 617). It is the only slow-motion type.

05.Frame Size and Frame Rate

Frame size is the picture's width and height in pixels, and frame rate is the number of pictures a second, counted in fps, frames per second. In an option such as 3840 by 2160 at 30p, the default (p. 578), the p means progressive: each frame is one complete picture. The rounded names hide fractions. 24p is 23.976 fps, 30p is 29.97, 60p is 59.94, 120p is 119.88 and 240p is 239.76, while 25p, 50p, 100p and 200p are exact (p. 202). Video from First Principles explains why these rates exist.

5.4K is 5376 by 3024 and 4K UHD (Ultra High Definition) is 3840 by 2160 (p. 1004), and 1920 by 1080 is often called Full HD. Every frame is 16:9 (p. 210). The list changes with the file type (p. 200), so it splits into two tables.

The H.265 column covers both bit depths (pp. 201-202). Electronic VR is locked OFF for every 5.4K option, 4K at 120p and 100p, 1920 by 1080 from 100p to 240p, and slow motion (pp. 201-202), and Video Autofocus and Stabilisation covers it.

Slow Motion

Slow motion reads the sensor fast and records at a normal rate. The 30p $\times 4$ option reads at 120p and records 30p, so playback is $120 / 30 = 4$ times slower and about 10 seconds of action fills about 40 (p. 203). The 25p $\times 4$ option reads at 100p and records 25p, and the 24p $\times 5$ option reads at 120p and records 24p, $120 / 24 = 5$ times slower. All three are 1920 by 1080 in H.264 (p. 202), at about 30 Mbps (p. 208).

A slow-motion clip has no sound, recording stops after about three minutes, and the g11 custom setting, Extended shutter speeds (S/M), is switched off (p. 203). Video flicker reduction, Electronic VR and timecode output are off, as is Hi-Res Zoom in the FX image area, and an HDMI recorder receives normal-speed footage (p. 204).

06.Full Width or Crop

FX is the full-frame area and DX a smaller central one, about 35.9 by 20.2 mm against 23.7 by 13.3 mm in video (p. 211). Since $35.9 / 23.7 \approx 1.5$, a 180 mm lens frames in DX roughly as a $180 \times 1.5 = 270$ mm lens does in FX. Image Area and Crop covers the crop factor and the video menu's own Image area item, FX by default (p. 578). Some modes fix their own area and the rest follow that item (pp. 200, 210).

Nikon gives the 240p and 200p modes about 95 percent of the FX angle of view and does not say whether the Image area item affects them (p. 210). The slow-motion row is my inference: p. 210 fixes nothing for it, its cautions treat the FX image area as a possible setting (p. 204), and Hi-Res Zoom lists all three slow-motion options under the DX image area (p. 217).

The full widths come from the stills sensor, 6048 pixels across in FX and 3984 in DX (p. 998). Trimmed to 16:9, $6048 \times 9 / 16 = 3402$ and $3984 \times 9 / 16 \approx 2241$, listed as 2240, which are the raw frames (p. 200). The FX 4032 frame and 5.4K span the same width with fewer pixels, $6048 / 4032 = 1.5$ and $6048 / 5376 = 1.125$ sensor pixels per recorded pixel, by a method Nikon does not describe.

A DX lens forces the DX area, removes the FX raw options, turns 5.4K into 3840 by 2160, and drops 1920 by 1080 at 240p or 200p to 120p or 100p (pp. 201, 210). Electronic VR shrinks any crop by an amount the RG does not state (p. 211), and Nikon's launch release puts the view at that of a lens about 1.25 times longer. What a Partially Stacked Sensor Changes discusses whether readout speed explains the DX crop at 4K 120p, a reason Nikon does not give.

Oversampling

Oversampling means reading more pixels than the file holds and combining them down. The RG never uses the word, but Nikon's launch announcement of 17 June 2024 says the camera can make 4K UHD footage by oversampling from 6K, and a footnote limits that to 4K at 60p, 30p, 25p and 24p in the FX area, without mentioning 50p. If the 6K readout is the 6048 by 3402 raw grid, which is my assumption, each 4K frame draws on

$$\frac{6048 \times 3402}{3840 \times 2160} = \frac{20{,}575{,}296}{8{,}294{,}400} \approx 2.48$$

readings per recorded pixel. Combining several readings generally averages away noise and keeps detail cleaner than skipping rows. DX is 3,984 pixels wide, so DX 4K has at most $3984 / 3840 \approx 1.04$ readings across, which suggests little or no oversampling there. That too is my inference.

07.How Big a Minute Is

Bit rate becomes card space in two steps: multiply by the seconds, then divide by 8 bits per byte. For H.265 10-bit at 4K 30p, about 190 Mbps (p. 207),

$$\frac{190 \times 60}{8} = 1{,}425 \text{ MB} \approx 1.4 \text{ GB},$$

taking 1 GB as 1,000 MB, and real files vary around Nikon's approximate rate. Every other format scales with its bit rate, so FX 6048 N-RAW at 60p in High quality, about 3,730 Mbps (p. 205), fills $3730 / 190 \approx 19.6$ times the card each minute, and Cards and Data Rates for Video works through each one.

08.My Everyday Format

I record H.265 10-bit (MOV) in SDR at 3840 by 2160 and 30p, or 60p for passages that may be slowed later, since $60 / 30 = 2$ leaves enough frames for half speed. Three reasons decide it.

Ten bits store four times as many tonal steps in each channel, so a smooth gradient such as a dusk sky or a lit backdrop is recorded in finer steps with less banding, even in a clip posted as shot. They also leave headroom for contrast added later, though I post clips as shot. The files stay small, about a fifth of Apple's ProRes 422 HQ target, $190 / 884 \approx 0.21$, and only $190 / 150 \approx 1.27$ times the 8-bit default (p. 207). And since this type alone offers SDR, HLG and N-Log (p. 199), a later move to log or HDR changes the tone mode, not the file type, though each brings its own ISO floors and drops Set Picture Control (pp. 215-216).

The cost is p. 288: on the camera a 10-bit clip cannot be trimmed, have frames saved, or play at a slowed Playback speed. I accept it, because those jobs belong on a computer and I post clips as shot. 4K at 30p and 60p in FX also sits within the modes Nikon calls oversampled.

09.Recommended Settings

Stage dance. 4K at 30p in FX and SDR, with the Picture Control as Video from First Principles sets it.

Outdoor and gimbal dance. 4K at 60p for passages that may be slowed, 30p for the rest.

Wildlife and birds. DX for reach with the 180-600, accepting that DX 4K probably gets little oversampling.

10.Beginner Mistakes

Looking for HLG under an 8-bit type. HLG appears only under H.265 10-bit (MOV) (p. 199). Change the file type first.

Expecting sound or 4K in slow motion. Slow motion is silent 1920 by 1080 (pp. 202-203). Shoot 120p in H.265 if an editor will slow it.

Wondering why 4K at 120p looks tighter. It is fixed to DX (p. 210). Widen the lens, or drop to 60p for full width.

Expecting to trim or slow a 10-bit clip on the camera. Only 8-bit clips take the editing options and Playback speed (p. 288). Use a computer for 10-bit clips.

Leaving the 8-bit default for footage meant to be adjusted. H.265 8-bit (MOV) (p. 578) can band when Flat footage gains contrast. Switch to 10-bit before the shoot.

11.Related Chapters

• Video from First Principles: frame rates, shutter speed and exposure for video.
• N-Log, HLG and Exposure Tools: what each tone mode does.
• Cards and Data Rates for Video: every bit rate, card speed and recording time.
• Video Autofocus and Stabilisation: Electronic VR and its crop.
• RAW, JPEG, HEIF and Tone Mode: bit depth and the Bayer mosaic for stills.
• Image Area and Crop: FX, DX and the crop factor.
• Picture Controls and Colour: Standard and Flat.
• What a Partially Stacked Sensor Changes: readout speed behind the fast modes.
• Card Slots and File Handling: Nikon's raw-video card list.