The RAW vs JPEG dynamic range gap comes from what each file keeps. A raw file stores linear sensor data at 12 or 14 bits, so a modern APS-C or full-frame camera can hold 10 to 11.5 stops of photographic dynamic range at base ISO; a JPEG is an 8-bit, tone-curved rendering that throws much of that away. Because raw is linear, the brightest stop holds half of all the recorded values, which is why exposing as far right as you can without clipping (ETTR) gives cleaner shadows, and why you should judge clipping with care, since the camera's histogram is built from the JPEG preview, not the raw data.
- 50% of a linear raw file's levels sit in the brightest stop; ten stops down a 14-bit file has only 16.
- +1 stop of exposure gives about 1.41× the shot-noise SNR in every tone.
- Half a stop early, a stop late: how far the camera histogram missed raw clipping in RawDigger's test.
- 10.5 to 11.6 stops of photographic dynamic range for current APS-C and full-frame bodies at base ISO.
- ISO invariance means many modern sensors can be shot at base ISO and brightened later.
This is for people who already shoot raw and want to use it properly.
The numbers below come from Bill Claff's Photons to Photos measurements, two Google and MIT papers on sensor noise, and a RawDigger test that measured how far a camera's histogram drifts from its raw data.
Why does the brightest stop hold half the tonal values?
Because a sensor counts light linearly and a stop is a doubling. A 14-bit raw file has 16,384 levels; the top stop, from full scale down to half, uses levels 8,192 to 16,383, which is half of all the levels the file can store. Each stop below gets half of what remains.
That gives the table below for a 14-bit file. The deep shadows, ten stops down, have only 16 levels to describe all their gradation, and in practice most of those are buried in noise.
| Stop below clipping | Raw levels (14-bit) | Raw levels (12-bit) | Share of all levels |
|---|---|---|---|
| 1st (brightest) | 8,192 | 2,048 | 50% |
| 2nd | 4,096 | 1,024 | 25% |
| 3rd | 2,048 | 512 | 12.5% |
| 5th | 512 | 128 | 3.1% |
| 8th | 64 | 16 | 0.4% |
| 10th | 16 | 4 | 0.1% |
| 12th | 4 | 1 | 0.02% |
How a JPEG spends its levels
A JPEG works the other way. It applies a gamma-like tone curve before quantising to 8 bits (256 levels), spreading the levels roughly evenly by perceived brightness.
That is efficient for viewing but leaves little room to pull detail out of a shadow later, because the shadows were rounded to a handful of values and the highlights above the curve's shoulder were simply cut.
| Property | Raw file | Camera JPEG |
|---|---|---|
| Bit depth | 12 or 14 bits | 8 bits |
| Levels | 4,096 or 16,384 | 256 |
| Encoding | Linear sensor counts | Gamma-like tone curve applied |
| White balance, picture style | Stored as settings, changeable | Baked in |
| Highlights past the curve's shoulder | Kept up to sensor clipping | Cut |
Google's HDR+ paper gives a phone-sized version of the same point: raw pixels from its sensors were typically 10 bits against 8 bits for the processed YUV stream, but the real advantage was under 2 bits, because raw is linear and the processed stream already had a tone curve applied.
Bit depth on a spec sheet overstates the gain; linearity is the real story.
Tonal levels per stop, and what ETTR buys
What does expose to the right actually do?
ETTR gives the sensor more light without clipping the highlights you care about, then you pull the brightness down in raw processing. More photons means a higher signal-to-noise ratio everywhere in the frame, and the shadows benefit most because they started with the fewest photons.
In July 2003 Michael Reichmann, the founder of the Luminous Landscape website, was leading a photography workshop in Iceland. On a drive of about 3,000 km he talked signal-to-noise ratios with Thomas Knoll, the original author of Photoshop and Camera Raw. The fact Reichmann then set out for readers was simple: a linear 12-bit file puts 2,048 of its 4,096 levels in the brightest stop, so a dark exposure wastes most of the file.
Reichmann wrote that the idea had not occurred to him before. On 31 July 2003 he published "Expose Right": push the histogram as far right as possible without clipping, then normalise in the raw converter. He also gave the catch, slower shutter speeds, and a tripod test to see it for yourself.
Michael Reichmann, Luminous Landscape: Expose Right (2003)The arithmetic of one extra stop
Shot noise is the square root of the photon count (Hasinoff, Durand and Freeman, 2010: for bright pixels, SNR increases with the square root of the light collected). Expose one stop more and every tone gets twice the photons and about 1.41 times the SNR.
If you then darken the file by one stop in the raw converter, the picture looks the same brightness as the metered shot but with cleaner shadows. In the deepest shadows the gain is often larger, because the fixed read noise of the camera becomes a smaller share of a bigger signal.
Samuel Hasinoff, Frédo Durand and William Freeman put it formally in 2010: SNR rises steadily with the number of electrons collected up to saturation, which, they note, is exactly what justifies the expose-to-the-right principle. To maximise SNR in a single shot, expose as long as possible without saturating, at the lowest ISO that allows it.

When does ETTR hurt instead of help?
ETTR hurts when it clips a channel you needed, when it forces a slower shutter that blurs the subject, or when the scene's range already fills the sensor. It is a tool for low-to-moderate contrast scenes and static subjects, not a rule for every frame.
- Single-channel clipping. A red flower, a saffron flag or a sodium streetlight can clip the red channel while the luminance histogram still looks safe. Once one channel clips, the colour of that area is lost, and the converter will invent a hue.
- Motion. Adding a stop by halving shutter speed doubles motion blur. On a moving subject, a sharp but noisier frame beats a clean but blurred one.
- High-contrast scenes. If the scene already spans more stops than the camera can hold, there is no room to shift right. Protect the highlights and let the shadows go, or bracket.
- JPEG-only workflows. If you deliver the camera JPEG, ETTR just gives you a bright picture. The benefit only exists when you develop the raw file.
The opposite advice, to underexpose and lift later, gets a clear answer on Photography Stack Exchange. The most-voted reply says plainly not to, because the darkest parts are the noisiest and lifting them amplifies that.
Another highly voted reply quantifies it: a full stop of underexposure doubles the noise you see after correction, two stops quadruple it, while any overexposure starts clipping.
Underexposure is not free; it is just less obviously ruined.This post
Why does the camera histogram lie about raw clipping?
Because it is computed from the JPEG preview, after white balance, picture style, contrast and saturation are applied. White balance multiplies the red and blue channels, so a value that is safe in the raw data can show as clipped in the preview, and a saturated colour can clip in raw before the preview warns you.
A controlled test on a Canon 5D Mark II at ISO 200 compared the in-camera histogram and blinking-highlights warning with the actual raw values. With a neutral subject and standard settings, the warning appeared half a stop before the raw data actually clipped. With a red subject and low contrast, about 60,000 red pixels were already clipped with no warning, and by the time the warning appeared over 2 million were clipped, roughly a stop late. Setting UniWB (unity white balance) with maximum contrast brought warnings within about 0.3 EV of the raw data. The limit: one camera body and firmware; other models need their own test.
Read the article (free)So the histogram is useful but biased, and in different directions for neutral and saturated subjects. Three practical fixes:
- Calibrate your own camera. Shoot a grey card in steps of a third of a stop, open the files in a raw analyser such as RawDigger, and note how many stops past the first blink the raw data actually clips. You will usually find a fairly consistent margin for neutral subjects, and you can expose to it.
- Watch the RGB histogram, not luminance. The colour channels show red or blue clipping that the luminance graph averages away.
- Consider UniWB if you are serious. It sets white balance multipliers to 1, so the preview turns greenish but the histogram tracks the raw data closely. You fix white balance in the converter, where it belongs for raw anyway. More on that in white balance and colour temperature.
The top answer on the Stack Exchange histogram thread is still the right foundation: read left as shadows, right as highlights, and treat a tall wall at either edge as lost detail. ETTR just adds that the right edge should be approached deliberately.
How much dynamic range do real cameras have?
At base ISO, current APS-C bodies measure about 10.5 stops of photographic dynamic range, full-frame bodies about 10.5-11.6, and older or phone sensors much less. Above the lowest settings, each doubling of ISO costs close to one stop.
The chart uses Photons to Photos' "photographic dynamic range" (PDR), which counts stops from clipping down to the point where the signal-to-noise ratio, adjusted for sensor size, falls to 20. That is a stricter floor than the engineering measure (SNR of 1), so the numbers are smaller than marketing claims but comparable across formats.
See the data as a table
| Nikon D90 (APS-C, 2008) (stops) | Sony ZV-E10 (APS-C, 2021) (stops) | Sony A7 IV (full frame, 2021) (stops) | Apple iPhone 14 Pro Max (raw) (stops) | |
|---|---|---|---|---|
| ISO 100 | 9.34 | 10.48 | 11.57 | 7.89 |
| 200 | 9.2 | 9.75 | 10.64 | 7.16 |
| 400 | 8.48 | 9.2 | 10.39 | 6.25 |
| 800 | 7.52 | 8.13 | 9.31 | 6.13 |
| 1600 | 6.53 | 7.17 | 8.2 | 6.07 |
| 3200 | 5.51 | 6.15 | 7.27 | 6.06 |
| 6400 | 4.3 | 5.22 | 6.26 | 6.13 |
Source: Bill Claff, Photons to Photos: Photographic Dynamic Range vs ISO Setting (revised Nov 2026) · PDR = stops from clipping to SNR 20, normalised for sensor size. The phone curve flattening above ISO 800 is as measured; it may reflect processing before the raw file is written.
Three things stand out. The 2021 APS-C Sony holds about 1.1 stops more than the 2008 Nikon D90 at ISO 100, and stays 0.5 to 0.9 stop ahead at every higher ISO. The full-frame A7 IV adds about another stop, roughly what its larger area predicts. And the phone, even shooting raw, starts about 3.7 stops below the full-frame body, which is why phones lean on the multi-frame stacking I describe in how phone cameras compute a photo.
I shoot deliberate cinematic work on a Sony ZV-E10 and still have a Nikon D90, so this chart is personal: the ZV-E10 at ISO 1600 (7.2 stops) is within half a stop of the D90 at ISO 800 (7.5 stops). That is the kind of comparison that tells you whether an upgrade buys you anything real.
Does ISO even matter once you shoot raw?
Less than most people think, and in a specific way. For a fixed shutter speed and aperture, raising ISO does not add light; it only amplifies the signal before readout. On many modern sensors that amplification barely changes shadow noise, a property called ISO invariance, so you can shoot at a lower ISO and brighten later with little penalty, while keeping more highlight headroom.
Older sensors were different. Their read electronics added noise after the amplifier, so a higher ISO setting genuinely cleaned up the shadows for the same exposure. Hasinoff and colleagues measured how big this "high-ISO advantage" was:
| Camera | Released | Pixel pitch (µm) | High-ISO potential (dB) |
|---|---|---|---|
| Canon 5D Mark II | 2008 | 6.4 | 19.7 |
| Canon 1D Mark II | 2004 | 8.2 | 17.9 |
| Canon 1D Mark III | 2007 | 7.2 | 15.7 |
| Canon 350D | 2005 | 6.4 | 15.3 |
| Nikon D3 | 2007 | 8.5 | 11.1 |
| Nikon D300 | 2007 | 5.5 | 3.1 |
| Canon S70 (compact) | 2004 | 2.3 | 2.2 |
A large number means raising ISO helps shadows a lot (the camera is far from ISO invariant); a small number means it barely matters. The Nikon D300, at 3.1 dB, was already close to invariant in 2007. Most current sensors behave more like the D300 than the 5D Mark II, which is why "expose for the highlights at base ISO, lift the shadows in post" now works on so many bodies.
The MIT team modelled sensor noise as photon noise plus read noise before and after the ISO amplifier, using published noise data for seven cameras. They showed that, saturation aside, a higher ISO gives better SNR for the same exposure time in the darkest regions, and they turned HDR bracketing into an optimisation over exposure time and ISO. Compared with the usual fixed-ISO bracket spaced in factors of 2 or 4, their sequences gave up to 19 dB better worst-case SNR in the same capture time on some cameras. The limit: the advantage depends on the camera's noise split, so near ISO-invariant sensors gain little.
Read the paper (free)
Is underexposing and stacking smarter than ETTR?
For high-contrast scenes, yes, and it is what phones do. Instead of one frame pushed right, capture several frames exposed to protect the highlights, align them and average them; noise in the shadows falls with the square root of the frame count while highlights stay safe.
Google's HDR+ deliberately underexposes every frame so fewer pixels saturate, captures 2-8 frames of identical exposure, and merges them to reduce noise in the shadows before tone mapping. It caps the deliberate underexposure at 8× (3 stops) because pushing further makes the result look cartoonish, and its exposure schedule holds 8 ms at minimum gain before raising gain to 4× and then both up to 100 ms and 96×. The limit: it needs alignment and robust merging, so fast-moving subjects get fewer usable frames.
Read the paper (free)On a camera you can do the manual version: a short burst on a tripod, merged in your raw converter or a stacking tool. Exposure bracketing gives a wider range; equal-exposure stacking gives cleaner shadows with no tone-mapping seams. The fundamentals of how aperture, shutter and ISO trade stops are in the exposure triangle explained.
What should a raw workflow look like, step by step?
- Meter for the highlight that matters. Spot-meter a bright detail you want to keep, or watch the RGB histogram and blinkies.
- Shift right by your calibrated margin. If your tests show raw clipping about 0.5 EV after the first blink on neutral subjects, use that margin. Use none on saturated colours.
- Stay at base ISO if the sensor is close to invariant and the shutter speed allows it; raise ISO only to protect shutter speed.
- In the converter, set exposure first, then the curve. Bring the midtones to where they should be, then shape highlights and shadows with a tone curve rather than global clarity.
- Check channels at the end. Look at the red and blue channels in saturated areas before export; clipped channels show as flat, posterised patches.
For touring and travel shooting, where the light changes constantly and you get one chance, the habits differ again; I cover them in the phone photography guide, and my camera kit is listed on the gear page.

What's new
Questions people ask
What are real examples of RAW's advantage over JPEG?
Photography Stack Exchange's long-running thread has good ones: a wedding first-dance shot that was unusable as a JPEG but recoverable from raw, a sunset whose sky and foreground detail came back in Lightroom, and a concert frame at ISO 1600 pushed hard without falling apart. The top answer makes the wider point: raw's value is editing latitude, because the final print or screen holds far less range than the file.
Should I underexpose to protect highlights?
Only as much as the highlights need. A full stop of underexposure roughly doubles visible noise after correction; any overexposure past clipping loses detail permanently. Expose as far right as the important highlights allow, and on ISO-invariant sensors stay at base ISO.
Is the histogram on my camera the raw histogram?
No. It is computed from the JPEG preview with your white balance and picture style applied. In RawDigger's test on a Canon 5D Mark II, the warning came half a stop early on neutral subjects and about a stop late on red ones.
Does 14-bit raw give more dynamic range than 12-bit?
Only if the sensor's noise floor is low enough to use the extra levels. Bit depth sets how finely the range is divided, not how wide it is; the measured PDR curves are the better guide.
Sources
- Bill Claff, Photons to Photos: Photographic Dynamic Range vs ISO Setting (revised Nov 2026)
- Photons to Photos: Engineering and Photographic Dynamic Range (definition of PDR)
- Hasinoff, Durand, Freeman: Noise-Optimal Capture for High Dynamic Range Photography, CVPR (2010)
- Hasinoff et al.: Burst photography for high dynamic range and low-light imaging on mobile cameras, SIGGRAPH Asia (2016)
- RawDigger: Beware the Histogram (2012)
- Adobe: Digital Negative (DNG)
- RedShark News: Smartphone camera news from MWC Barcelona 2026 (2026)
- YM Cinema: Sony LYTIA L910 LOFIC sensor (2026)
- Photography Stack Exchange: Good examples of RAW's advantages over JPEG?
- Photography Stack Exchange: How and why do you use an image histogram?
- Photography Stack Exchange: Should I shoot underexposed photos?
