RAW, JPEG, HEIF and Tone Mode

One frame from the Z6III can weigh 27.3 MB or 2.4 MB, and both can record the same instant (pp. 117, 1012). The first is a raw file at Nikon's lossless setting, the second a full-size JPEG at the lowest grade. The extra 24.9 MB buys room to change your mind about brightness, colour and contrast after the shutter has closed.

Picture a dancer stepping into a hard amber spotlight, sequins a shade too bright and the backdrop almost black. A JPEG has already decided what that amber looks like. A raw file has kept the readings and left the decisions to you. Four items in the photo shooting menu govern that choice.

01.What a Raw File Holds

Start with one photosite, a single light-sensitive well on the sensor. During the exposure it gathers light, and afterwards the camera measures how much and turns that amount into a number. It counts light and cannot tell red from blue.

To get colour, the sensor puts a tiny red, green or blue filter over every photosite, so each one counts only the light its filter passes. The usual layout repeats a two-by-two tile of one red, two greens and one blue, called a Bayer mosaic after Bryce Bayer, the Kodak engineer who proposed it. Nikon's Reference Guide uses the word Bayer only in its video pages, where it labels both raw video formats as Bayer data (p. 197), and that video comes off the same sensor.

So the raw file, which Nikon calls NEF and names with the extension .NEF (p. 455), is a mosaic: one number per photosite, each describing one colour only. NEF files are always recorded at Large, which in the FX image area is 6,048 by 4,032 pixels (p. 119), so an FX NEF carries about 24.4 million readings. None of them is yet a colour pixel.

Demosaicing, Worked Through

Turning the mosaic into a picture is called demosaicing. For each photosite, the converter estimates the two colours its filter blocked from nearby sites that measured them. The figure below shows a patch with invented readings. The outlined blue site read 2,090 and has no red or green of its own. Its four diagonal neighbours are red sites, and a simple estimate of its red is their mean:

$$\frac{9{,}180 + 9{,}220 + 9{,}140 + 9{,}260}{4} = \frac{36{,}800}{4} = 9{,}200$$

Its four side neighbours are greens reading 5,240, 5,200, 5,280 and 5,220, with a mean of 5,235. The finished pixel is red 9,200, green 5,235 and blue 2,090, and only the blue was measured. Real converters use cleverer rules, but every pixel ends the same way: one value measured, two estimated.

The converter then applies white balance, which scales red and blue against green so a grey card comes out grey, a Picture Control's contrast, saturation and sharpening, and a tone curve that suits a screen. A JPEG has all of that done once, in the camera. A NEF stores those settings beside the readings as instructions, so white balance, exposure compensation and the Picture Control can change later while the readings stay untouched (p. 118). You can convert a NEF in the camera through Retouch in the playback i menu (pp. 118, 260) or on a computer in Nikon's free NX Studio (p. 118).

02.Bit Depth From First Principles

Each reading is stored as a whole number of fixed length, and that length is the bit depth. A bit is one digit that is either 0 or 1, so one bit holds two values. Two bits hold four: 00, 01, 10 and 11. Each extra bit doubles the count, so $n$ bits hold

$$L = 2^{n}$$

levels, where $L$ is the number of levels. Twelve bits give $2^{12} = 4{,}096$ levels and fourteen give $2^{14} = 16{,}384$. Nikon lists the Z6III's NEF stills as 14-bit (p. 998), so each reading is a whole number from 0 to 16,383. The camera's raw video formats, N-RAW and ProRes RAW, are 12-bit (p. 197).

Where the levels sit matters more than how many there are. A sensor's reading is linear: twice the light gives twice the number. One stop is a doubling or halving of light, so the brightest stop before the sensor clips takes the top half of the numbers, 8,192 to 16,383 in a 14-bit file, the next stop the quarter below, and so on. Counting down from clipping, stop number $k$ holds

$$L_k = 2^{\,n-k}$$

levels. The table below puts numbers on the pattern.

Ten stops down, a 12-bit file has four levels for a whole stop of shadow and a 14-bit file has sixteen. Those are the shadows you lift when rescuing an underexposed frame. With four levels, a smooth gradient across a dark curtain can break into visible steps, a fault called banding. With sixteen, each step is a quarter the size.

The argument has a limit. Extra levels help only where the sensor's noise, the random variation between readings, is smaller than a step. In deep shadow at high ISO the noise may span many levels, so 14 bits earn their keep mostly at low ISO. Nikon publishes no noise figures in levels, so I will not say where the crossover sits.

A finished JPEG holds 8 bits for each of red, green and blue, $2^{8} = 256$ levels per colour. That works because the conversion applies a tone curve first, giving the shadows far more of the codes than a linear spread would, in step with how the eye judges brightness. It is also why a JPEG has little left to give when you lift its shadows three stops.

03.Lossless and Lossy Compression

Compression stores the same information in fewer bytes. Lossless compression gives back exactly what went in, as unzipping a folder restores every file. Lossy compression gives back something close, having thrown away detail it judged least likely to be missed, and that detail is gone for good.

An uncompressed Large frame holds $6{,}048 \times 4{,}032 = 24{,}385{,}536$ readings, and at 14 bits each, divided by 8 bits per byte, that is

$$\frac{24{,}385{,}536 \times 14}{8} = 42{,}674{,}688 \text{ bytes}$$

or about 42.7 MB. Nikon lists a lossless NEF at about 27.3 MB (p. 1012). Taking a megabyte as a million bytes, which Nikon does not specify, the lossless file is about 64 per cent of the uncompressed size. The file also carries the settings and, in most raw formats, a preview picture, so treat that as rough.

How does a file shrink with nothing lost? Neighbouring photosites of one colour usually read similar values. In the figure, two greens on the top row read 5,240 and 5,260. Storing both in full costs 28 bits. Storing the first in full and the second as its difference, $+20$, needs only enough bits for a small number. Six bits cover every whole number from $-32$ to $+31$, so the pair costs $14 + 6 = 20$ bits, and the decoder adds 20 back to recover 5,260 exactly. That is one common lossless method, not necessarily Nikon's, which the Reference Guide does not describe. It also explains why foliage compresses less than clear sky.

The other two options are smaller: High efficiency★ about 17.3 MB and High efficiency about 11.4 MB (p. 1012), roughly 41 and 27 per cent of the uncompressed size. Nikon does not say how they work, only that High efficiency★ is close to lossless in quality and better than High efficiency (p. 466). If one beats the other, the weaker cannot be an exact copy, and saying High efficiency★ compares favourably with lossless, rather than calling it lossless, suggests it is not exact either. So I treat both as lossy, which is my reading of Nikon's wording, not its statement.

A JPEG is lossy by design and works on the finished picture. It splits the picture into small blocks and keeps each block's broad tone and colour carefully while recording fine texture coarsely, because the eye notices a wrong tone over an area long before a little lost grain. Nikon gives nominal ratios of about 1:4 for fine, 1:8 for normal and 1:16 for basic (p. 998), where 1:8 means one eighth of the uncompressed size. Treat those as names: an uncompressed 8-bit Large picture needs three bytes per pixel, $6{,}048 \times 4{,}032 \times 3 \approx 73.2$ million bytes, and Nikon's table lists a Large fine file at about 10.3 MB (p. 1012), nearer 1:7 than 1:4.

Each grade comes in two forms. The plain options use what Nikon calls size-priority compression and the ★ options optimal-quality compression (p. 998), which its capacity notes call optimal-compression (p. 1013). Nikon adds only that ★ makes files larger and that its capacity figures assume the plain options (p. 1013).

04.JPEG, HEIF and Tone Mode

HEIF, the High Efficiency Image File Format, is a newer container for finished pictures. It does a JPEG's job with a different compression method and room for more than 8 bits per colour. Baseline JPEG, the kind the Z6III writes (p. 998), stops at 8 bits. HEIF can hold 10 bits, $2^{10} = 1{,}024$ levels per colour, four times as many. Nikon's guide does not print the bit depth of the Z6III's HEIF stills, so take 10 bits as what the format allows, not a figure Nikon states.

Dynamic range is the span from the darkest to the brightest tone a system can hold, counted in stops. SDR, standard dynamic range, means pictures built for an ordinary screen. HDR, high dynamic range, means a screen that can go much brighter than SDR's white, so a sunlit cloud or a stage lamp can glow instead of flattening into white paper.

HLG, Hybrid Log-Gamma, is a way of encoding pictures for HDR screens that began in television broadcasting (p. 463). The hybrid is two curves joined end to end. The lower part of the range follows a curve close to SDR's, so shadows and midtones look about right on an ordinary screen. The upper part is logarithmic, giving each stop of highlight a similar share of the codes instead of halving them stop by stop, so bright tones keep their gradation on a screen that can show them.

Nikon does not say why HLG stills must be HEIF, but the standards suggest a reason. BT.2100, the colour space HLG fixes (p. 462), specifies 10 or 12 bits per colour, more than baseline JPEG's 8 can carry. Spreading a wider range across 256 levels would make each step coarse, and a smooth sky would band. A 10-bit file has four times as many steps to spend. The camera enforces the pairing: with SDR, finished pictures are JPEG with the extension .JPG, and with HLG they are HEIF with the extension .HIF (pp. 455, 462).

Tone mode offers SDR and HLG, and the default is SDR (p. 446). With HLG selected the shooting display shows "HLG" (p. 462), and Nikon lists a set of costs:

• More noise, which Nikon describes as scattered bright pixels, fog and lines (p. 462).
• An ISO floor of 400, two stops above the usual ISO 100, no Hi 0.3 to Hi 1.7, and an Auto ISO Maximum sensitivity of at least ISO 800 (pp. 159, 462).
• Color space fixed at BT.2100, the HDR television standard (p. 462).
• No Active D-Lighting, Multiple exposure, HDR overlay, the high-speed frame capture modes C15 to C120 (bursts of Large JPEGs at 15 to 120 frames a second) or Custom Setting d3 Pre-Release Capture options (p. 462), and Continuous H (extended) slowed to the Continuous H rate (p. 155).
• Possibly weaker subject detection (p. 130).
• Picture Controls from Set Picture Control (HLG), limited to Standard, Monochrome and Flat (pp. 183, 999).

Viewing costs too. HLG pictures may look wrong on a computer or HDMI device without HLG support (pp. 918-919), SnapBridge may not show them (p. 463), and changing monitor or viewfinder brightness under HLG shifts their tones, with highlight fidelity falling as brightness rises (pp. 814, 816).

Tone mode reaches the raw file too. Under HLG, a RAW + JPEG/HEIF option records a NEF and a HEIF (p. 117), and Nikon notes that a JPEG made in the camera from an HLG raw comes out about 2 EV (two stops) darker than one from an SDR raw, to be corrected with exposure compensation during processing (p. 271). It does not say why. With the ISO 400 floor as well, that settles it for me: for anything I will edit from raw, Tone mode stays on SDR.

05.Image Quality Settings

Image quality governs Self-timer, Single frame, Continuous L, Continuous H and Continuous H (extended). Image quality (HSFC) governs only the high-speed frame capture modes C15 to C120 (p. 464).

Image Quality

Image quality offers thirteen choices in three families:

• RAW alone, which records a NEF and nothing else.
• Six finished-picture grades with no NEF: JPEG/HEIF fine★, JPEG/HEIF fine, JPEG/HEIF normal★, JPEG/HEIF normal, JPEG/HEIF basic★ and JPEG/HEIF basic.
• The same six grades with a NEF beside each picture, from RAW + JPEG/HEIF fine★ down to RAW + JPEG/HEIF basic.

The grades rank from highest quality to lowest as fine★, fine, normal★, normal, basic★ and basic, and JPEG/HEIF means a JPEG under SDR and a HEIF under HLG (p. 117). The default is JPEG/HEIF normal (p. 446), which keeps no raw file at all. At Large, Nikon lists plain fine at about 10.3 MB, normal 6.3 MB and basic 2.4 MB (p. 1012).

A pair shares one file name and differs only in extension, DSC_0412.NEF beside DSC_0412.JPG, or DSC_0412.HIF under HLG (p. 455). A High efficiency★ NEF plus a Large fine JPEG takes about $17.3 + 10.3 = 27.6$ MB, about one lossless NEF, and a ★ JPEG is larger still (p. 1013).

With one card in the camera, playback shows only the JPEG or HEIF of each pair, and deleting it deletes the NEF too (p. 117). With two cards under Overflow or Backup, deleting the JPEG or HEIF still deletes the NEF (p. 117). Sending the NEF to one card and the JPEG to the other is a secondary-slot setting (p. 118), covered in Card Slots and File Handling.

Image Quality (HSFC)

Image quality (HSFC) offers JPEG fine and JPEG normal, with JPEG normal as the default (p. 446). Those modes record only Large JPEGs, are unavailable under HLG (p. 157), and belong to Release Modes, Frame Rates and the Buffer. Choose JPEG fine: with no raw file behind those frames, the better grade is your only margin.

Quality Without the Menu

A control given the Image quality/size role in Custom Setting f2 changes quality with the main command dial and size with the sub-command dial (p. 712), or Image quality (HSFC) in C15 to C120 (p. 464). The default i menu carries Image quality, Image size and Tone mode (p. 83). The + RAW role adds a NEF to the next picture when a JPEG or HEIF option is selected, at the current RAW recording setting, and the old setting comes back once your finger leaves the shutter-release button, or when you press the control a second time (p. 710).

06.Image Size Settings

Image size sets the pixel dimensions of JPEG and HEIF pictures, choosing Large, Medium or Small. It never touches the NEF, which is always Large (p. 119).

Medium is not half of Large: each side is $4{,}528 / 6{,}048 \approx 0.749$ as long, so it holds $0.749^2 \approx 0.56$ of the pixels, and Small, at half each side, holds a quarter. Print size is pixels divided by dots per inch (p. 120), so a Large picture at 300 dpi is $6{,}048 / 300 \approx 20.2$ inches, or 51.2 cm, wide. Other image areas give fewer pixels at each size, which Image Area and Crop explains.

With Enable DX image sizes set to ON, DX pictures take their size from Image size (DX) while Image size governs the other image areas (p. 120), useful if you want full-size FX but smaller DX pictures. The defaults are Large, OFF and Large (p. 446), and the NEF stays Large regardless (p. 120). I leave all three alone: with a NEF in every genre, Image size touches only the companion JPEG.

07.RAW Recording

RAW recording sets how NEF files are compressed. From largest to smallest the options are Lossless compression, High efficiency★ and High efficiency (p. 466), and the default is High efficiency★ (p. 446). It applies to every NEF, including those the + RAW role adds (p. 710).

Nikon's figures come from fixed test conditions, Continuous H (extended), FX, SDR, the electronic shutter and its own 660 GB card, and vary with the scene and the type of card (pp. 1012-1013). The last two columns are my arithmetic, scaling by $512/660$ and $256/660$. The counts do not follow from file size: a lossless file is $27.3 / 17.3 \approx 1.58$ times a High efficiency★ file, yet Nikon fits $30{,}600 / 13{,}300 \approx 2.30$ times as many of the latter. Nikon does not explain the gap, so I plan from its counts.

Nikon puts the buffer, the short-term memory that holds frames before they reach the card, above 1,000 frames for all three options at ISO 100 (pp. 1012-1013), so it is no reason to prefer one. The buffer itself belongs to Release Modes, Frame Rates and the Buffer.

I keep High efficiency★ for everything except landscape. The 512 GB CFexpress card holds about 23,700 of them, more than a long day of birds will use. For landscape, where an evening may yield forty frames and shadows may be lifted several stops, lossless costs little and removes the one question I cannot answer: whether High efficiency★ discards anything. Plain High efficiency, the option Nikon ranks lowest, is about a third smaller again, and these cards do not need it.

08.Recommended Settings

This agrees with First-Day Setup, which starts with RAW + JPEG/HEIF fine★ and High efficiency★, moves to plain RAW once a raw workflow is routine, and uses lossless for landscape. The difference is that I make that move by genre rather than by date.

Wildlife and birds. With the Nikkor Z 180-600mm f/5.6-6.3 VR, frames come in long bursts and most are deleted. RAW alone at High efficiency★: every keeper gets edited, and dropping the ★ JPEG removes something Nikon says can lower the buffer (p. 1013). Outdoor dance follows the same logic.

Stage dance. With the Tamron 70-180mm f/2.8 Di III VC VXD G2, RAW at High efficiency★ and Tone mode on SDR. Stage light looks made for HLG, but more noise and weaker subject detection are the wrong trade there. At stage ISOs noise may swamp the finer 14-bit steps, so raw earns its place through the edit, above all white balance under coloured gels.

Street and portrait. RAW + JPEG/HEIF fine★ at High efficiency★. A street frame often goes to someone the same day, and a portrait subject wants a look before the edit. With the Nikkor Z 85mm f/1.8 S, sessions are short and the extra 10 MB or more a frame is nothing.

Landscape. RAW with Lossless compression. For a scene meant for an HLG-capable screen, a sunrise over water say, I would switch to HLG with RAW + JPEG/HEIF fine★ so a NEF sits beside the HEIF, and accept the ISO 400 floor on a tripod.

Video. These are photo settings, and nothing here changes gimbal dance video, which has its own settings in the video part of this guide.

09.Beginner Mistakes

Keeping the default. JPEG/HEIF normal records no raw file, so the first badly exposed frame has nothing behind it. Set RAW + JPEG/HEIF fine★ on day one.

Turning on HLG by accident. Files come out as .HIF, other people's screens may show them wrongly, and ISO will not go below 400. If "HLG" appears in the shooting display, set Tone mode back to SDR.

Shrinking Image size to save space. Image size never touches the NEF. If space is tight, RAW recording is the lever.

Deleting the JPEG and losing the raw. With one card, or with Overflow or Backup, deleting the JPEG or HEIF of a pair in the camera deletes the NEF too (p. 117). Cull on a computer, or send each format to its own card.

Judging a NEF by another editor's first view. Picture Control, Active D-Lighting and vignette settings show only in Nikon's software (p. 919). Use NX Studio to see the camera's look.

10.Related Chapters

• First-Day Setup: starting settings and a second-slot plan per genre.
• Picture Controls and Colour: the look a JPEG fixes and a NEF stores, including the HLG set.
• White Balance: the colour correction a raw file lets you redo.
• ISO and Auto ISO: sensitivity, noise and the ISO 400 floor under HLG.
• Image Area and Crop: DX, 1:1 and 16:9 and their pixel counts.
• Card Slots and File Handling: where each half of a pair is written.
• Menus and Controls: giving a button the + RAW or Image quality/size role.
• Glossary: bit depth, dynamic range, HEIF, NEF and tone mode in brief.