Short answer
Tone mapping is the step that squeezes the brightness range of a real scene into the much smaller range a screen or print can show. Every camera does it; 'HDR' modes do more of it, usually after merging several frames to capture the range in the first place. A newer kind of HDR photo does the opposite job as well: it stores a normal picture plus a gain map, so a capable screen can show highlights brighter than white. How either looks depends on the curve chosen and on the display's headroom.
The paper that established high-dynamic-range photography put the problem in one sentence: neither computer screens nor photographic paper can show anything close to the ratio between dark and bright that exists in the real world, so cameras were never built to record it. A sunlit window and the room around it, a street lamp and the pavement under it, differ by more than a standard display can reproduce at once. Two things follow. The camera must first capture that range without clipping the top or drowning the bottom in noise. Then something must decide which parts of it the picture keeps, and how the rest is compressed. That second decision is tone mapping, and it is a rendering choice with no single correct answer.
The classic method is bracketing: photograph the scene several times at different exposures and merge the series into one radiance map in which every pixel has a usable reading. It works on a tripod and fails when anything moves between frames, leaving ghosts. Phone cameras took another route. Google's HDR+ captures a burst of frames at one exposure, chosen low enough that highlights do not clip, and merges them to pull clean detail out of the dark shadows that underexposure leaves. Because every frame has the same exposure, aligning them is more reliable. The cost is that the camera has to decide how far to underexpose before it has seen the result, which HDR+ does by matching the scene against a library of hand-tuned examples.
Underexposing to protect highlights is the same idea as 'exposing to the right' on a larger camera: keep the brightest important detail just below clipping and lift the rest later.
A global tone curve treats every pixel of a given value the same way. It is predictable, and it runs out quickly: lifting shadows enough to see into them flattens everything else. Local tone mapping lets the adjustment vary across the frame, brightening a dark doorway without washing out the sky beside it. HDR+ does this with a variant of exposure fusion. From the merged image it synthesises a short and a long exposure, then blends them region by region, favouring moderately bright pixels, before applying a global S-curve. Its authors cap the compression at a factor of eight, three stops, and warn that pushing range compression and saturation too far gives a flat, cartoon-like picture. That overcooked result is what many people mean by the 'HDR look'.
A tone curve is defined for brightness, but pictures have three channels, and how the curve is applied to them decides what happens to colour. Applied to red, green and blue separately, a compressive curve squeezes the strongest channel hardest. A bright saturated orange loses its red lead, drifts towards yellow and pales, even though no channel clipped. Applied as one factor to all three, the ratios and therefore the hue are held. The table shows both for the same light at rising exposure. The Academy's ACES documentation records the same change of approach: version 1 tone-mapped the RGB channels directly, while version 2 applies its tone scale to a lightness value alone and handles colourfulness separately so that hue survives into the display image.
Tone mapping throws highlight range away. HDR photo formats keep it for screens that can use it. The Ultra HDR format stores an ordinary standard-dynamic-range JPEG and, alongside it, a gain map recording how much brighter each pixel should be on an HDR screen. Software that does not understand the format shows the ordinary picture. Software that does combines the two and scales the boost to whatever the display can manage at that moment, which varies between devices and even on one device as its brightness changes. So an HDR photo has no single appearance. Against the 203 cd/m² reference white used in HDR production, a 1,000 cd/m² screen has a little over two stops of headroom for highlights and a 400 cd/m² laptop about one.
| Exposure | Curve on each channel | One factor for all channels |
|---|---|---|
| −1 stop | #9c6c2b · hue 71°, chroma 0.10 | #9c6023 · hue 62°, chroma 0.11 |
| +1 stop | #d5ac55 · hue 84°, chroma 0.12 | #d58533 · hue 62°, chroma 0.14 |
| +3 stops | #f2df92 · hue 96°, chroma 0.10 | #f2973b · hue 62°, chroma 0.15 |
| +5 stops | #fcf6ce · hue 100°, chroma 0.05 | #fc9d3e · hue 61°, chroma 0.15 |
| Display peak | Ratio to 203 cd/m² | Headroom (stops) |
|---|---|---|
| 203 cd/m² | 1.0× | 0.0 |
| 400 cd/m² | 2.0× | 1.0 |
| 600 cd/m² | 3.0× | 1.6 |
| 1,000 cd/m² | 4.9× | 2.3 |
| 1,600 cd/m² | 7.9× | 3.0 |
| 4,000 cd/m² | 19.7× | 4.3 |
| 10,000 cd/m² | 49.3× | 5.6 |
Why: Local tone mapping brightened the dark object and darkened the sky across too wide a border.
Fix: Reduce the local contrast or 'structure' setting, or blend exposures by hand with a soft mask along the edge.
Why: Range compression and saturation were both pushed too far.
Fix: Accept some real shadow, lower the shadow lift, and take saturation back below where it started.
Why: A bracketed series was merged although something moved between frames.
Fix: Use the camera's single-exposure burst mode, a faster bracket, or the merge tool's deghosting option.
Why: The app or service stripped the gain map or cannot read it, so only the standard picture is shown.
Fix: Judge and edit the standard rendition as well, since that is what most viewers will see.
Each statement is labelled by kind — established fact, a standard’s requirement, observed market data, a convention, or Colourwise’s own interpretation or analysis — with the strength of the evidence behind it.
FactStrong evidence
Debevec and Malik's 1997 method combines a series of photographs taken at different exposure settings into a single high-dynamic-range radiance map, on the grounds that neither screens nor paper can display the dynamic range present in real scenes.
Source: Recovering High Dynamic Range Radiance Maps from Photographs (Debevec and Malik, SIGGRAPH 97)
FactStrong evidence
Google's HDR+ deliberately underexposes a constant-exposure burst to avoid clipping highlights and limits its dynamic range compression to a factor of 8, noting that excessive range compression and saturation produce a flat, cartoon-like rendition.
Caveat: The pipeline as published in 2016; later phones and other makers differ.
FactStrong evidence
HDR+ performs local tone mapping with a variant of exposure fusion, blending a short and a long synthetic exposure derived from the merged image with weights that favour moderately bright pixels.
FactStrong evidence
ACES 1 applied tone mapping directly to RGB channels, whereas the ACES 2 rendering transform applies its tone scale to a lightness correlate only and adjusts colourfulness separately so that the hue of the original value is maintained.
Source: ACES Documentation: system overview, encodings and output transforms
FactStrong evidence
The Ultra HDR image format stores a standard-dynamic-range JPEG with a logarithmic gain map; readers that do not support it display the SDR image, and readers that do combine the two and adapt to the ratio between the display's HDR and SDR white levels.
Caveat: One published format. Other gain-map encodings exist, including ISO 21496-1, which the same document covers for compatibility; the ISO text itself was not read.
Source: Ultra HDR Image Format v1.1
StandardStrong evidence
ITU-R Report BT.2408 recommends 203 cd/m² as the reference level for diffuse white in HDR production.
Caveat: A report giving operational guidance for television, applied here to stills by analogy.
Source: Report ITU-R BT.2408: Guidance for operational practices in HDR television production
Colourwise analysisStrong evidence
Measured from a 203 cd/m² reference white, a display peaking at 400 cd/m² offers about 1.0 stop of highlight headroom, one at 1,000 cd/m² about 2.3 stops, and the 10,000 cd/m² ceiling of the PQ signal about 5.6 stops.
Based on: Colourwise calculation: log₂ of peak luminance divided by the 203 cd/m² reference white. See the headroom table.
Caveat: Real devices set their own SDR white level and vary their headroom with brightness setting, battery and heat.
Source: Report ITU-R BT.2408: Guidance for operational practices in HDR television production; ITU-R Recommendations BT.709, BT.2020 and BT.2100
Colourwise analysisModerate evidence
A compressive curve applied to each channel separately shifts a saturated orange progressively towards yellow as exposure rises and, at high exposure, strips most of its chroma, while the same curve applied as one factor to all three channels leaves hue unchanged.
Based on: Colourwise calculation with an illustrative curve y = x ÷ (1 + x) on linear RGB of 1 : 0.35 : 0.05, encoded to sRGB and read as OKLCH hue and chroma. See the table. The curve is chosen for illustration and is not any product's.
Source: International Electrotechnical Commission (IEC) webstore and catalogue; ACES Documentation: system overview, encodings and output transforms
Reviewed 1 October 2026. Colourwise summarises its sources in its own words and does not reproduce standards text or proprietary colour data. Spotted an error? Tell us.