Developing mountain photos from RAW
A RAW file is not a “finished but dull” photograph that simply needs to be made more spectacular. It is a set of capture data and metadata that the processing engine must interpret to produce a visible image: demosaicing, colour rendering, tonal curve, white balance and other operations already take place before you even begin your creative choices.
This distinction changes the entire method. Processing is not about “recovering the maximum” or systematically opening every shadow. It is about deciding which recorded information should remain visible, which hierarchy of light should be preserved and which final output the image must support — screen, web or print.
This guide deliberately remains software-independent. Names differ between Lightroom, Camera Raw, Capture One, DxO PhotoLab, darktable or RawTherapee; the useful principles remain the same: build a starting rendering, adjust tone and colour, localize corrections, deal with optics, noise and sharpness, then prepare an output suited to its destination.
Profile, demosaicing and the starting curve explain why two engines can show two different images from the same file.
An adjustment can reveal recorded information; it cannot turn a completely clipped or blurred area into faithful data.
Opening every shadow and holding back every highlight can remove precisely the relief that made the scene work.
Noise, sharpening, dimensions, colour space and compression must also be judged according to the final use.
RAW: data to interpret, not a hidden final image
On most colour cameras, the sensor does not directly measure a complete red, green and blue value for every final pixel. A colour filter array means that each photosite mainly records one component. The RAW engine must then demosaic these samples, estimating the missing components from neighbouring data to construct usable RGB pixels.
Before display, the engine must also give those values a colourimetric meaning, apply white balance and transform a largely linear capture response into a tonal rendering compatible with a screen or output file. That is why “displaying RAW without processing” is misleading: as soon as a readable colour photograph appears on screen, rendering choices have already been made by the engine.
| Point | RAW | In-camera JPEG |
|---|---|---|
| Data | Preserves capture data in the camera’s RAW format, with final processing left to the RAW engine. | An image already demosaiced, rendered and compressed, with sharpening and noise processing defined by the pipeline and camera settings. |
| White balance | Remains largely reinterpretable because the engine can modify channel gains from the RAW data. | The colour rendering has already been transformed; major corrections have less latitude. |
| Tone and colour | Profile, curve and rendering can be chosen during processing. | A large part of these decisions is already baked into the file. |
| Compression / bit depth | Depends on format and camera; many RAW files retain more tonal data than an 8-bit JPEG. | Commonly an 8-bit format with lossy compression. |
| Use | Source for processing and archiving. | Lightweight, widely compatible file ready to share. |
Why the same RAW does not look the same everywhere
Two software packages may use different demosaicing algorithms, different camera profiles, a different tonal curve, different default noise reduction or even a different processing engine depending on the version. They can therefore produce different starting points without either one having “falsified” the file.
The preview displayed by the camera adds another source of difference: it is generally a JPEG rendering built with the camera’s image settings. RAW software may read those metadata or partially ignore them and apply its own rendering. Comparing the camera screen with the imported image as though they were two neutral readings of the same object therefore often creates a false alarm.
Non-destructive does not mean without rendering
In a non-destructive RAW workflow, the original remains intact. The software stores instructions — in a catalogue, database, sidecar or another mechanism — then regenerates the preview from the RAW and those instructions. The actual pixel file intended for the web, print or another editor is created only when exporting or rendering to JPEG, TIFF, PSD or another output format.
Profile and white balance: choose a base before correcting
Start by choosing a coherent starting rendering before trying to “repair” the image with ten adjustments. A more contrasty profile can make shadows look blocked; a more linear profile may look flat but leave more room for your own curve. Neither starting point is universally neutral or superior.
White balance: flexible, not infinite
On a RAW file, white balance mainly works by reweighting the colour channels. The value recorded by the camera generally serves as a starting point, not a lock. This makes it possible to cool or warm a scene, correct a cast or restore a more believable relationship between snow, rock, sky and vegetation.
This flexibility does not recreate missing signal, however. If one channel is genuinely clipped, the corresponding colour information is missing in that area. At the other extreme, a nearly black sample contains too little signal to provide a reliable reference. A scene mixing warm direct light, sky-lit shade and artificial light does not necessarily have one single “correct” white balance for the whole frame.
Temperature
It shifts the overall balance toward warmer or cooler rendering. Use it to establish a believable relationship between areas of the scene, not to reach a memorized number.
Tint
It corrects an axis usually running green–magenta, or its equivalent depending on the engine. It becomes useful when an image looks greenish or magenta while the overall temperature appears correct.
Neutral reference
A genuinely neutral area can help, provided it is not clipped. In landscape work, you can then move away from this neutrality if the scene’s light calls for it.
Tonal base: reveal information without rewriting the light
The adjustment called “exposure” in processing software is not the physical exposure made at capture. It changes how the recorded signal is rendered. It can lighten or darken the image very naturally, but it does not send new photons to the sensor and does not recreate dynamic range that was never captured.
Capture, the aperture–shutter speed–ISO choice and general bracketing belong in Exposure in mountain photography. Here, we start from the file already recorded.
Histogram: look at the rendering you are creating
In many RAW processors, the visible histogram describes the current rendering and changes when you alter exposure, curve, balance or other settings. Some software also offers tools closer to the RAW data, but their definition and place in the pipeline vary. It is therefore better to ask “which histogram is this software showing me at this stage?” than to assume that a histogram called RAW would be universally independent of all rendering.
Highlights: recovery or reconstruction?
A highlight that looks white is not necessarily lost. It may simply be mapped too high by the current rendering. If data is still present, reducing exposure or highlights, or changing the curve, can make it visible again.
When only one or two colour channels are clipped, some engines attempt to reconstruct colour or structure from the remaining channels. This is useful, but it is no longer complete recovery of intact data: the software infers part of the result. If all channels are saturated in an area and no structure was recorded, there is no faithful detail to reveal.
Shadows: the cost of an extreme lift
Brightening a shadow reveals both the signal and its defects. The weaker the starting signal, the more luminance noise, colour noise, read irregularities or banding may become visible. A dark backlit valley is therefore not a free reservoir of detail.
A useful diagnosis is to temporarily raise the shadows, inspect what the file genuinely contains, then return toward a coherent rendering. If the material becomes dirty before the area reaches the brightness you want, you have reached a limit of the file — not a lack of determination on the slider.
| Situation | What is possible | Limit to watch |
|---|---|---|
| Highlights rendered too bright | Remap recorded data toward lower tones. | Do not turn snow grey simply to display every nuance. |
| One channel clipped | Some engines can use the other channels to reconstruct a plausible colour. | The reconstructed area is not equivalent to three intact channels. |
| All channels saturated | Reposition the area as white in the rendering. | Detail that was never recorded cannot be faithfully restored. |
| Very weak shadows | Brighten the signal that is present. | Noise, colour shifts and banding may become dominant. |
Keep a hierarchy: not every area needs equal readability
The trap of modern RAW is confusing latitude with an obligation to show everything. A mountain face in deep shadow can remain dark. A cloud around the Sun can stay close to white. Processing often gains credibility when it preserves priorities: some deep shadows, luminous highlights and midtones separated enough to read the relief.
Whites and blacks adjustments position the extremes of the rendering. A small white specular reflection or a deliberately black shadow is not automatically a mistake. Check what is clipped and ask whether that area genuinely needed to contain texture.
Tone curve: organize contrast rather than apply a shape
The curve describes a relationship between input and output values. Raising one part of the curve brightens the corresponding range; lowering it darkens it. It can strengthen midtones, protect one extreme or create a more gradual separation between planes.
An “S-curve” is only one possibility. On a misty morning, it can destroy softness. On very graphic relief, stronger contrast may be coherent. The right curve is the one that produces the desired hierarchy without crushing useful texture.
Contrast, texture and haze: three different scales
Global contrast organizes large tonal masses: dark slope, bright sky, snow, valley. Microcontrast acts more on local variations and the perception of detail. Confusing the two often leads to an image that is technically detailed but visually exhausting.
| Family | Primarily used to | Risk when pushed |
|---|---|---|
| Global contrast / curve | Structure large luminance relationships. | Blocked shadows, harsh highlights, excessive hierarchy. |
| Fine texture | Strengthen or soften relatively fine details depending on the algorithm. | Grainy rock, “sandy” snow, nervous vegetation. |
| Clarity / local contrast | Strengthen separation around structures and midtones. | Halos around ridges and a crunchy rendering. |
| Dehaze / haze correction | Modify contrast associated with atmospheric haze and often perceived colour. | Over-saturated sky, distant planes stuck together, atmosphere removed. |
Names and algorithms vary between software. The diagnosis remains valid: zoom in on a dark ridge against a bright sky. If a light band appears on one side and a dark one on the other, if the rock becomes harder than the light could have produced or if distant planes lose all softness, local contrast is probably too strong.
Colour: preserve the relationships of the scene before intensity
Colour in the mountains provides valuable control points: snow, sky, rock and vegetation have familiar behaviours. That does not mean they should have fixed colours. It means that a colour shift becomes easier to spot when several elements stop telling the same story about the light.
Global saturation, adaptive saturation and colour mixer
A conventional saturation adjustment increases or decreases the intensity of all colours. Tools often called Vibrance, intelligent saturation or adaptive saturation tend to protect colours that are already saturated and act more on others — but their algorithms are not standardized between software.
The colour mixer or HSL allows more targeted work: shifting the hue of a family, changing its intensity or luminance. It is powerful precisely because the effect is selective. Excessive movement of blue can turn the sky and snow shadows cyan; an overly broad change to yellows/greens can contaminate grass, lichens and some rocks at the same time.
Snow
It should first remain bright if the scene was bright, without losing important texture. A shadow can be blue because of the sky, and snow at sunset can genuinely be warm. Avoid neutralizing every area toward the same chromatic target.
Sky
A deep blue can be natural, but watch for a shift toward cyan or saturation stronger than everything else in the image. Dehaze, white balance and HSL can all accumulate on the same sky.
Rock
Material quickly disappears under a double excess of warmth and microcontrast. Check less-lit areas: if every rock turns orange or magenta, the rendering has probably gone beyond the local light.
Vegetation
Mountain greens often contain yellows, cool greens and very low-saturation areas. Pushing them toward a uniform “radioactive” green removes depth instead of adding it.
Warm light: keep a cast without tinting the whole file
At golden hour, the goal is not to make the terrain neutral and then inject orange again. Start by preserving the relationship between directly lit areas, shadows and sky. If the whole image becomes equally warm, the chromatic contrast of the light disappears.
Colour grading that separates shadows, midtones and highlights can sometimes refine this relationship. It remains optional: if white balance, profile and colour mixer already produce a coherent result, adding another colour layer simply because it exists strengthens nothing.
Local corrections: act where the problem exists
A local correction is relevant when a global adjustment improves one area but damages the rest. If darkening the sky makes the foreground too dark, or strengthening rock texture makes the sky dirty, a mask lets you separate the needs.
Sky, relief and foreground are not three different files
The trap of modern masks is optimizing each area independently: perfectly detailed sky, perfectly opened mountain, perfectly readable foreground. The result can then lose the coherence of the moment’s light.
Instead, work from one question: which area is currently preventing the image from reading properly? Correct it just enough to restore the hierarchy, then look at the whole image again.
Luminance, colour and subject masks
Available tools vary, but three principles transfer widely:
- luminance: target a brightness range, for example the highlights of a cloud or the shadows of a slope;
- colour: target a chromatic range, for example certain greens or blues;
- subject / sky / landscape: use automatic detection when available, then inspect and refine the selection.
An automatic mask is not a geometric truth. Inspect ridges, trees, spires, possible cables and fine transitions. A contour that is too hard around a ridge immediately reveals the processing.
Optics, geometry and framing: correct without pretending to change the capture
Lens profiles can compensate for known flaws of a camera–lens combination. They are not all the same kind of problem: distortion, vignetting and chromatic aberration require different diagnoses.
| Defect | What the correction does | Useful check |
|---|---|---|
| Distortion | Geometrically transforms the image to straighten barrel or pincushion curvature. | Check edges, stretching and loss of field after correction. |
| Optical vignetting | Brightens darker edges or corners caused by the optics. | Decide whether the darkening is a technical defect or contributes to the reading. |
| Chromatic aberration | Realigns or reduces coloured fringes depending on the type and algorithm. | Inspect high-contrast edges without desaturating genuine coloured detail. |
Software perspective: a transformation, not a new viewpoint
A geometry tool can level the horizon, correct converging verticals or horizontals and change aspect, scale or framing. It does so by transforming pixels, sometimes leaving empty areas or requiring a crop.
It cannot go back in time and move the photographer. It does not change the occlusion relationships that existed between two summits, recreate the hidden part of a slope or turn the physical perspective linked to the viewpoint into a different perspective that was authentically captured. That distinction belongs in Choosing focal lengths and managing perspective; here, software correction remains a finishing geometric operation.
Crop and horizon: early in the thinking, revisable in the workflow
Straightening the horizon and choosing the final ratio fairly early helps judge composition, masks and the areas that will actually remain visible. But a non-destructive workflow generally lets you revisit the crop. Just remember that a substantial crop reduces the number of available pixels and that a strong geometric transformation may require even more cropping.
Dust and distractions: distinguish cleanup from transforming reality
A sensor spot, hot pixel or small accidental element at the edge of the frame can be cleaned without particular difficulty. Removing a road, moving a summit or rebuilding a large area, however, is no longer simple RAW correction: you are entering a retouching or compositing workflow that deserves a separate editorial decision.
Noise and sharpness: preserve useful detail, not the perfect pixel
Luminance noise mainly appears as brightness variation or grain. Chromatic noise produces more coloured speckles or variations. Engines do not all process them in the same way, and some modern systems combine demosaicing and denoising very early in the pipeline.
Denoising: reduce a defect without erasing the material
Noise reduction always creates a trade-off. Too little and noise distracts in a sky or shadow. Too much and rock, forest or snow become smooth surfaces that never existed.
First inspect the areas where noise is distracting at 100%, then return to output size. Grain visible at pixel level may disappear in a web image or a print viewed from a normal distance; conversely, smoothing that looks attractive at 100% can look plastic once printed.
AI denoising: a family of tools, not a guarantee
Some engines now use neural networks directly on RAW data and may combine denoising and demosaicing. Others apply AI later or generate a new intermediate file. There is therefore no universal “AI step”.
The right control is visual and comparative: microtexture in rock, tree needles, granular snow, stars or fine ridges. If the software extracts very spectacular detail but the material becomes repetitive, waxy or less natural, reduce the effect or return to a simpler mode.
Three moments of sharpening: a useful model, not a software requirement
Separating sharpening into three functions helps reasoning:
- capture sharpening: compensate for part of the blur inherent to the sensor, optics, any anti-aliasing filter or demosaicing;
- creative sharpening: selectively strengthen an important structure in the image;
- output sharpening: compensate for the rendering of resizing and the final medium, screen or print.
Not every software package exposes these three phases in the same way. The important idea is not to ask one global adjustment to solve capture, visual hierarchy and final output at the same time.
Amount / strength
Controls the strength of sharpening on detected edges. Too high and it amplifies noise and halos.
Radius
Roughly controls the width of the area affected around edges. A radius that is too large produces visible borders.
Threshold / masking
Reduces the application of sharpening to small variations or smooth areas so noise and sky are not treated as details to strengthen.
Order of operations: think in dependencies, not a list of clicks
It is useful to have a working method, but misleading to claim that one slider order is universal. In many non-destructive processors, the order in which you move controls in the interface is not the order in which the engine actually processes the pixels. Some pipelines have a fixed internal sequence; others allow certain modules to be moved.
What matters is understanding a few dependencies:
- RAW interpretation: demosaicing, profile, balance and operations specific to raw data condition the starting point;
- global base: processing exposure, tonal extremes and curve let you judge the overall structure before multiplying masks;
- colour and local corrections: they are easier to judge once the tonal base already works;
- noise and sharpness: they interact, and their effective position depends on the engine; avoid oversharpening noise you intend to smooth later;
- output: resizing, destination profile, compression and output sharpening depend on the final file you need to produce.
You can return to an earlier stage. If a new profile changes the shadows, it makes sense to revisit contrast. If soft proofing reveals a loss of saturation, adapting the rendering may be useful. Processing remains iterative.
Series: synchronize a base, not an entire image
On a sequence shot in the same light, synchronizing profile, starting white balance, optical corrections or part of the tonal base can save a great deal of time. But a sky mask, crop or local correction depends on the geometry and brightness of each frame.
The coherence of a series is judged from the result: the same contrast logic, the same colour family, the same sense of texture. It is not measured by ten files showing the same numerical value.
Master and export: decide what the file must be used for
The original RAW is not the final delivery file, and the final JPEG should not become your only archive. A robust workflow preserves at least the source file and the editing state that can reproduce the rendering. If the work passes through a pixel editor with layers or retouching that cannot be described in the RAW catalogue, a TIFF/PSD master or equivalent can also be part of the archive.
| Destination | Common format / space | Important decision |
|---|---|---|
| Web / sharing | JPEG is highly compatible; sRGB remains a common choice for screen compatibility. | Pixel dimensions, compression and output sharpening suited to the platform. |
| High-quality JPEG or TIFF depending on the laboratory’s requirements; ICC profile requested by the provider when available. | Follow the laboratory’s actual specifications rather than a universal preset. | |
| High-quality intermediate | 16-bit TIFF with lossless compression when the downstream workflow justifies it. | Preserve sufficient depth and the expected colour space without unnecessarily multiplying heavy files. |
| Working archive | Original RAW + reproducible editing state; optionally a pixel master if external retouching is used. | Do not depend only on the final JPEG. |
JPEG or TIFF: the destination decides
Standard JPEG is an 8-bit format with lossy compression: excellent for distributing a finished photograph at a reasonable file size. TIFF can retain 8 or 16 bits per channel and use lossless compression; it is useful as an intermediate or when a laboratory, publication or other software requires more latitude.
A 16-bit TIFF is not automatically “better” for every print. If the laboratory asks for a high-quality sRGB JPEG, sending a large TIFF in another space can complicate the workflow without improving the output.
Colour space and embedded profile
A colour space describes how to interpret the file’s numerical values. For broad web distribution, sRGB remains a highly compatible choice. For printing, follow the laboratory’s instructions: working space, possible output profile and requested format.
Embedding the profile allows colour-managed applications to know how to interpret the file. Soft proofing using the output ICC profile can simulate gamut losses or tonal changes before printing.
This simulation is not absolute proof: its reliability also depends on the display, its profile, ambient lighting and the quality of the output profile.
Dimensions, PPI and compression: do not create pixels with metadata
For screens, pixel dimensions are the main data. For a print, physical dimensions and the resolution requested by the laboratory together determine the number of pixels required. Changing only a PPI value without resampling creates no new detail; resampling creates new pixels through interpolation, not new optical information.
JPEG quality sets a trade-off between size and compression losses. There is no magic value: inspect the final file at its distribution size and pay particular attention to sky gradients, fine edges and detailed areas.
Output sharpening: after defining the output
Resizing can soften an image, and a print does not behave like a screen. That is why output sharpening makes more sense once dimensions, medium and resolution have been defined. A large print can also reveal halos or noise smoothing differently from a web image: assess the result with the final size and likely viewing distance in mind.
For a large print, the specifications of the actual printing chain remain the reference: dimensions, medium, profile, expected resolution and viewing distance can all affect output decisions.
Diagnosis: identify the tool that broke the image
When processing looks artificial, do not lower every slider at random. Find the symptom, then return to the family of corrections that could have created it.
| Symptom | Likely cause | Test / correction |
|---|---|---|
| Light or dark halo around a ridge | Clarity, Dehaze, sharpening or mask too hard. | Disable these families one by one; reduce radius/amount or soften the mask transition. |
| Very cyan or electric-blue sky | White balance too cool, HSL, saturation and Dehaze accumulating. | Return to the base profile/WB, then reintroduce colour corrections one by one. |
| Grey snow | Excessive highlight compression or processing exposure too low. | Restore the snow’s brightness while controlling only the texture that genuinely matters. |
| Uniformly blue snow | Global correction too cool or blue/cyan pushed too far. | Compare sunlit and shaded areas; preserve a natural cast where it comes from the light. |
| Shadows with an “HDR” look | Global or local lifting too strong, blacks raised everywhere. | Restore depth to areas that do not need full readability. |
| Crunchy rock / aggressive trees | Microcontrast, texture and sharpening stacked together. | Reduce local contrast first, then reassess sharpening at 100%. |
| Plastic-looking surface | Excessive noise reduction or AI denoising too strong. | Return to microtextures; accept some grain if necessary. |
| “Radioactive” colours | Global saturation + adaptive saturation + HSL stacked together. | Reset targeted corrections and rebuild colour relationships in small steps. |
| Loss of texture in snow or clouds | Real clipping or rendering that pushes values to white. | Check channels and available data; distinguish recovery from reconstruction. |
| Banding in a sky or shadow | Weak signal strongly amplified, overly aggressive gradient or output with insufficient depth/compression. | Compare RAW/high-bit-depth master and exported file; if the defect appears only on export, revisit format, bit depth or compression. |
| Very clean file but no relief | All shadows opened, all highlights held back, uniform local contrast. | Re-establish a global hierarchy: which areas genuinely need to dominate, stay bright or remain dark? |
| Image looks “sharp” at 100% but harsh in output | Sharpening designed for pixels rather than the destination. | Reduce creative sharpening and redo output sharpening after resizing. |
HDR and panorama: distinguish assembly from processing
If you have a bracketed sequence, an HDR merge can create a file containing complementary information from the exposures; moving elements can produce differences or ghosting. The choice to bracket and the general management of dynamic range are made at capture, as explained in the guide to settings and exposure in mountain photography. Once the HDR merge has been produced, the resulting file is then processed according to the same principles of tone, colour and natural rendering.
For a panorama, stitching the frames is a separate stage from RAW processing. Once the file has been stitched, its global processing follows the principles in this guide.