Curves and Levels: The Decisions That Matter

Curves and levels are the two tonal adjustment tools that sit underneath almost every professional image edit. Levels sets three things: the black point, the white point and the midtone balance between them. Curves goes further and lets you move any point along the tonal range, from deep shadow to bright highlight, and it lets you do that separately for the red, green and blue channels. Between them, they control brightness, contrast and color balance more precisely than any slider labeled "brightness," "contrast" or "vibrance" ever will.

This matters to anyone whose images have to look right and look consistent: an ecommerce manager whose returns rise when a navy sweater arrives looking black, a marketer assembling a campaign from three photographers, an in-house designer who has inherited a folder of files that all look slightly different. It also matters to practitioners who already know where the Curves dialog lives but want a clearer way to decide which point to move, how far, and in which channel.

Rather than list features, this article follows one illustrative project from brief to delivery and explains every decision along the way. The project is invented for teaching purposes and is labeled as such; the numbers in it are realistic for a small product set but they describe no real client. The techniques, the order of operations and the traps are the same ones that apply to a single hero image or a catalog of thousands.

The Illustrative Brief: 48 Linen Pillows and a Color Problem

Illustrative example. A home goods brand sends a set of product photographs for its spring collection: linen throw pillows in six colorways, eight images per colorway, 48 images in total. Each colorway has a front view, a back view, three detail crops of the weave and piping, and three lifestyle frames on a sofa. The photographs were shot over two days in the same studio, but on day two a large window was uncovered for the lifestyle frames, and the afternoon light mixed with the strobes. The client's brief is short: the pillows must match the physical swatches, the white background must be clean, and the six colorways must read as one family on the collection page.

That brief is typical, and it is almost entirely a curves and levels job. There is no compositing, no heavy skin work, no object removal beyond a few loose threads. What the brand actually needs is correct tone and correct color, repeated 48 times without drift. The fastest way to fail is to open each file, drag sliders until it "looks good," and move on. The fastest way to succeed is to decide the targets first, build the corrections as reusable adjustment layers, and check every result against numbers rather than impressions.

Before anyone opened Photoshop, three questions were answered in writing. First, what is the reference for "correct"? Here it was the physical fabric swatches the brand mailed, viewed under a daylight-balanced lamp, plus a gray card and color target photographed at the start of each day's session. Second, what is the output? Web product pages in sRGB at 2,000 pixels on the long edge, plus a smaller set for a printed lookbook later. Third, what counts as done? Every background within a narrow band of neutral white, no clipped detail in the white or charcoal colorways, and each colorway matching its swatch closely enough that the brand's own team signs it off side by side.

If you want to see how this kind of job fits into a larger editing workflow, our image editing and retouching services page describes the full range of work, from tonal correction through clipping paths and retouching. For framing decisions that sit alongside tonal consistency on a collection page, see the guide on consistent product image framing.

Project Timeline From Intake to Sign-Off

The schedule matters because tonal work goes wrong when it is rushed at the end. In this illustrative project, the correction work was front-loaded: most of the thinking happened on a handful of reference images, and the remaining files were processed from that thinking rather than from scratch.

Four working days for 48 images sounds generous, and for a single straightforward colorway it would be. The time goes into the reference images and the mixed-light frames, not the bulk of the set. A studio that skips the reference stage often finishes the first pass faster and then spends twice as long in revisions, because each image was corrected in isolation and they no longer agree with one another.

Reading the Histogram Before Touching Curves and Levels

The histogram is a bar chart of how many pixels sit at each brightness value, from pure black on the left to pure white on the right. It does not tell you whether an image is good; a moody charcoal pillow on a dark sofa should have a histogram weighted to the left. What it does tell you is where the tonal information lives, whether anything is already clipped, and how much room you have to move things before you lose detail.

On the illustrative set, three patterns showed up immediately. The white-background front views had a tall spike near the right edge, which is the background, and a separate hump for the pillow itself. The detail crops of the natural linen colorway had a narrow, centered histogram with nothing near either edge, which is the signature of a flat, low-contrast image that needs its endpoints set. And the day-two lifestyle frames showed the red and blue channel histograms offset from each other, which is how a color cast looks when you switch the histogram to show individual channels.

What to look for, and what to ignore

• A spike stacked against either edge. That is clipping: pixels that have been forced to pure black or pure white and carry no detail. On a background it may be intentional. On the white pillow's piping, it is damage.
• Gaps at the ends. Empty space between the last bar and the edge means the image does not use the full tonal range. That is what Levels fixes first.
• Comb-like gaps across the middle. Regular empty bars mean earlier edits have stretched the data. A few are harmless; heavy combing warns that further stretching may produce visible banding.
• Channel offsets. In a neutral area, red, green and blue should overlap. If one channel's hump sits consistently to one side, there is a cast.

What to ignore is the overall shape. Photographers sometimes talk about an "ideal" bell curve, but there is no such thing. A white pillow on a white background should be heavily weighted right. The histogram is a diagnostic, not a target.

One practical detail: the histogram panel in Photoshop can show a cached approximation rather than the full data. Refresh it before making decisions on a file you have just edited, or you may be looking at a histogram from several steps ago.

Setting Black and White Points With Levels

The first correction on every reference image was a Levels adjustment layer, and it did one job: set the endpoints. Good practice is to set the black and white points first because everything else depends on them. A curve built on an image with muddy shadows and gray highlights will be fighting the wrong problem, and when you later fix the endpoints the curve will be wrong too.

The Levels dialog has three input sliders under the histogram: black on the left, white on the right and a gray midtone slider between them. Moving the black slider inward tells the software that the chosen value should now be pure black, and everything darker is clipped. Moving the white slider inward does the same for white. The gray slider adjusts the midtones without moving the endpoints, which is often called gamma. Adobe's own Help Center documentation on the Levels adjustment describes these input and output controls and the eyedropper options in detail, and it is worth reading once end to end if you only ever use Levels by feel.

Use the clipping preview, not your eyes

The single most useful habit with Levels is the clipping display. In Photoshop, holding Alt (Option on a Mac) while dragging the black or white input slider turns the image into a map of which pixels are about to clip. Dragging the white slider, the image goes black and clipping areas appear in white or color; dragging the black slider, the image goes white and clipping areas appear dark. Colored areas mean only one or two channels are clipping, which often matters less than full clipping but still loses color detail.

On the natural linen detail crops, the black slider was brought in until the deepest shadows in the weave just began to show in the preview, then backed off slightly. The white slider was brought in until the brightest highlights on the raised threads just began to appear, then backed off. The result used nearly the full tonal range without destroying any texture. On the white-background front views, the decision was different, and it is covered below.

Output levels and why they exist

The two output sliders at the bottom of the dialog compress the range instead of expanding it. Pulling the output black slider up from 0 means the darkest value in the file is no longer pure black. This is how you deliberately keep the deepest shadows printable on a press that cannot hold detail near solid black. For web output it is usually left alone; for the printed lookbook, the print team would set output limits in the conversion to CMYK instead. Output limits for print are a prepress decision, so agree them with whoever prepares the press files rather than guessing in the retouching stage.

The Eyedroppers, Auto Options and Target Values

Levels and Curves both offer three eyedroppers: set black point, set gray point and set white point. Clicking the black eyedropper on a pixel tells the software that pixel should be black; clicking the white eyedropper does the same for white; clicking the gray eyedropper tells it the pixel should be neutral, and it adjusts the channels to make red, green and blue equal there. The gray eyedropper is a fast cast remover, but only if you click something that really is neutral.

In this project, the gray card photographed at the start of each session was the neutral reference. Clicking it with the gray eyedropper on a test frame from each day produced a quick first read on how far the two days differed. It was not the final correction, because a single click on a single sample is noisy, but it confirmed the direction of the cast: day two was warmer and slightly magenta in the lifestyle frames.

Target values rather than pure black and white

Double-clicking an eyedropper opens a color picker where you can set the target it assigns. By default, black is 0,0,0 and white is 255,255,255. Many retouchers change these to slightly lighter blacks and slightly darker whites so that the eyedroppers never force a pixel to total clipping. The exact numbers depend on the output: web images can tolerate targets near the extremes, while files headed for print usually need more headroom, which the printer or prepress team should specify. For this project the white target on product highlights was set a few levels below 255 so that the brightest threads kept a trace of texture, and the background was handled separately to reach true white.

What Auto actually does

The Auto button in Levels and Curves is not magic; it runs one of several algorithms you can choose in the Options dialog. One enhances contrast across all channels together, one stretches each channel separately, one finds the darkest and lightest areas and uses them as endpoints, and a newer option aims for a balanced brightness and contrast. The Options dialog also sets how much of the extreme shadows and highlights Auto is allowed to clip, and whether it should snap neutral midtones.

Auto is useful as a diagnostic. If "stretch each channel" produces a dramatically different color than your image, the image probably has a cast. But Auto on a product shot is risky because it cannot know which colors are supposed to be there. On the rust-colored colorway, per-channel auto pulled the fabric toward neutral brown, which is exactly wrong.

In short, the toolkit on this project came down to six things:

• Levels Sets black point, white point and midtones in one dialog
• Curves Adjusts any point along the tonal range, not just three
• Channel adjustments Correct casts in red, green and blue independently
• Histogram Shows tonal distribution and reveals clipping
• Gray eyedropper Neutralizes a cast only when clicked on a truly neutral area
• Clipping preview Alt or Option while dragging a slider shows what will be lost

Neutralizing the Day-Two Cast With Channel Curves

Once the endpoints were set, the next job was color. The reference material is clear on the right tool: correct casts with channel curves. A master RGB curve changes brightness and contrast; a curve on a single channel changes how much of that color is present at each tonal level. Because a cast is rarely uniform, with daylight contamination often stronger in the shadows than the highlights or the other way around, a curve lets you correct it differently across the range, which Levels cannot do as finely.

Measure, do not guess

The procedure was built around color sampler points. Photoshop lets you place up to ten fixed samplers on an image, and the Info panel then shows their values before and after each adjustment. On the day-two lifestyle frames, samplers were placed on three areas that should be neutral: a highlight on the white sofa cushion, a midtone on the gray throw draped behind the pillows, and a shadow in the fold of the sofa. The sofa and throw had also been photographed on day one under strobes alone, so there was a known reference for how they should read.

The readings showed red running higher than green and blue in all three samplers, most strongly in the midtones, and blue slightly low in the highlights. That pattern, warm overall with a highlight-specific blue deficit, is the fingerprint of window light mixing with flash. A single white balance shift cannot fix it cleanly because the ratio is different at each level.

A good rule of thumb for how far to go: in a truly neutral area, the three channel values should be within a few points of one another. Absolute zero difference is neither necessary nor always desirable, because a studio white that reads as perfectly neutral on a sampler can look cold next to warm wood furniture. The furniture image editing guide deals with exactly that tension between measured neutrality and a believable interior.

Building Contrast With a Gentle S-Curve

With endpoints set and color neutral, the images were accurate but a little flat, especially the detail crops, where the weave depends on subtle tonal differences between raised and recessed threads. The standard tool here is the S-curve: lower a point in the quarter tones, raise a point in the three-quarter tones, and the midtones get steeper, which means more contrast where most of the detail lives. The reference guidance says gentle, and it means it.

How gentle is gentle

For the detail crops, the two points moved only a handful of levels from the diagonal, enough to see the thread texture separate from the shadow between threads when the layer was toggled on and off, not enough that the effect was obvious without the toggle. The front views took a slightly weaker curve, because the pillows' form already had shading from the light, and the lifestyle frames took a weaker one still, because stronger contrast made the sofa look harsh. A useful test is to toggle the curve layer at 100 percent zoom and again at the size the image will actually be viewed. If the difference at viewing size is dramatic, the curve is probably too strong.

Why the blend mode matters

Steepening a curve in RGB also increases saturation, because it pushes the three channels further apart in colored areas. On the rust and mustard colorways, a standard S-curve made both fabrics noticeably more saturated than their swatches. The fix was to set the contrast curve layer's blend mode to Luminosity, which applies the tonal change without the saturation shift. This is one of the most useful, and most overlooked, decisions in curve work: the same curve does two different jobs depending on blend mode.

Shape, not just amplitude

The position of the points matters as much as how far they move. Placing the contrast points closer to the shadow end adds punch to the darker colorways without affecting the highlights much, which suited the charcoal pillows. Anchoring the highlight end with an extra point near the top, and leaving it on the diagonal, protected the white colorway's piping from brightening further. Curves are powerful precisely because you can decide where contrast goes. Levels' midtone slider can brighten or darken the middle, but it cannot add contrast to one part of the range while leaving another alone.

It is tempting to assume that a steeper curve always means a crisper, more professional image. It does not. Contrast is borrowed from somewhere. Every part of a curve that gets steeper forces another part to get flatter, and strong curves stretch the tonal data until smooth gradients break into visible steps. On fabric, that shows first in soft folds and shadows as banding, which reads as cheap rather than crisp.

Matching Six Colorways Across the Collection Page

The hardest part of this illustrative project was not any one image. It was making six colorways look as if they belonged together while each still matched its own swatch. Those two goals pull against each other: the brand wanted a consistent, clean look across the page, but a navy pillow and a natural linen pillow cannot share an identical tonal treatment without one of them looking wrong.

Separate what should match from what should differ

The solution was to split the adjustments into shared and per-colorway layers. The shared layers were the background treatment and the day-specific cast correction: every image from day two got the same channel curves, and every front view got the same background cleanup. The per-colorway layers were the Levels endpoints and the contrast curve, tuned on each reference image to match its swatch. That way, anything that should be identical across the page was literally the same adjustment, and anything that should differ was deliberately different.

Match by numbers, then by eye on a calibrated screen

For the backgrounds, "consistent" was defined as a numeric band: sampler readings within a few levels of one another on every front view, all neutral. For the fabrics, numbers were used to catch drift between images in the same colorway, by placing a sampler on the same area of the pillow across the eight images and comparing, and the eye was used for the match to the swatch. That last comparison only means something on a calibrated display, which is why calibration was the first job on day one rather than an afterthought.

The table is the kind of document worth writing down on any multi-colorway job. It is short, it records the reasoning, and it means the person who touches these files in six months does not have to rediscover why the charcoal pillow has a different black point from the rest. The same discipline applies to organizing the files themselves; our article on organizing image libraries by product covers naming and folder structures that keep masters, variants and exports from getting mixed up.

Keeping Every Adjustment Nondestructive

Everything above was done on adjustment layers, never directly on the pixels. The reference guidance lists destructive adjustments as a mistake to avoid, and on a project like this it would have been an expensive one. When the client asked in the review round for the sage colorway to be "a touch greener," the change took minutes: open the sage reference, nudge one point on one channel curve, and apply the updated layer to the other seven sage files. Had the correction been baked into the pixels, the only options would have been to redo the work from the raw files or pile a second correction on top of a first, stretching the data twice.

A layer stack that stays readable

Each final file carried three adjustment layers above the retouched pixel layer, always in the same order and always with the same names: a Levels layer named for endpoints, a Curves layer named for cast correction, and a Curves layer named for contrast on Luminosity. The background cleanup lived on a fourth, masked layer on the front views. A consistent stack means anyone can open any file and understand it at a glance, and it means layers can be dragged from one file to another or recorded into an action without surprises.

Work in 16 bits, deliver in 8

The raw files were converted to 16-bit working files. Each tonal adjustment stretches some part of the data, and in an 8-bit file there are only 256 levels per channel to begin with; stack three or four adjustments and gaps appear in the histogram that can turn into banding in smooth areas. A 16-bit file has vastly more levels to spare, so the same adjustments leave smooth gradients intact. The conversion to 8-bit sRGB happened only at export, after all tonal work was complete.

Where raw editors fit

Most of the endpoint and white balance work could have happened in a raw editor such as Lightroom, Camera Raw or Capture One, all of which include a tone curve and, in some cases, levels-style controls. For this project the raw editor did the white balance from the gray card and a neutral starting profile, and the finer curves and levels work moved to layered files because of the per-colorway masking and the need for a reusable, inspectable stack. On a simpler job with a single colorway and even light, doing everything in the raw editor with synced settings is often faster and entirely adequate. When exposure itself is the problem rather than tone, the techniques in our guide to HDR merging and exposure blending are the better starting point.

Quality Control: Clipping, Posterization and the Screen You Trust

Before the client saw anything, every file went through the same checks. None of them required judgment calls, which is the point: they catch the problems that tired eyes miss at the end of a long day.

Clipping check

With all adjustment layers visible, the histogram was checked for spikes at either edge in each channel, and the Levels clipping preview was used on the composite to confirm that only the background was reaching pure white. Clipping shadows or highlights is the first mistake on the reference list for a reason: once detail is gone, no later adjustment brings it back, and on a product page the lost detail is often the stitching, texture or finish that sells the item.

Posterization check

Strong curves that create posterization are the second mistake on the list. The check here was visual and targeted: zoom to 100 percent on the smoothest areas of each image, typically soft shadows on the pillow and the gradient of the background falloff on lifestyle frames, and look for steps instead of a continuous tone. Two lifestyle frames showed faint banding in the shadow under the sofa arm. The cause was an extra contrast point that had been added to rescue the mixed light; softening that curve and moving the correction into a masked channel adjustment solved it.

The screen you trust

Correcting color by eye on an uncalibrated screen is the third mistake, and it is the one that most often turns into a dispute with a client. A display that runs cool will lead an editor to warm everything; the files will look right on that screen and wrong everywhere else. The editing display in this project was calibrated with a hardware colorimeter to a standard white point and a moderate brightness suitable for a dim room, and the swatches were viewed under a daylight-balanced lamp next to it. Neither step is expensive compared with the cost of a reshoot or a round of returns.

It is worth being honest about what calibration does and does not do. It makes your screen accurate; it does not make your customers' phones accurate. The goal is to ensure that the file itself is correct, so that on any reasonably accurate display it looks the way the brand intended, and on inaccurate ones it is wrong in the same way every other correct file is wrong.

The final checklist, run on every file before review:

• Histogram checked per channel for spikes at either edge
• Clipping preview confirms only the background reaches pure white
• Background samplers neutral and within a narrow band across all front views
• Same-area fabric samplers consistent across all eight images in each colorway
• Smooth areas inspected at 100 percent for banding
• Each colorway compared to its swatch on the calibrated display
• Layer stack named and ordered consistently
• Export converted to sRGB with the profile embedded

Delivery, Revisions and What the Illustrative Numbers Show

The client review was done side by side: the brand's product lead had the physical swatches, the retoucher had the calibrated display, and the six reference images were shown first before the full set. Starting with the references means the conversation is about six decisions, not 48 images. In this illustrative project there were two notes: sage slightly greener, and the charcoal front view a little lighter so the weave was visible at thumbnail size. Both were single-layer changes applied to the whole colorway.

The delivered set was sRGB JPEGs at the agreed pixel dimensions, with the color profile embedded, plus the layered 16-bit masters archived for the later print lookbook. Embedding the profile is a small step that saves a lot of confusion: a file without a profile is interpreted however the receiving software chooses, which is exactly the inconsistency the whole project was designed to remove. If the files will also carry product codes, colorway names or usage rights, writing that into the metadata at export is far easier than retrofitting it; the guide to bulk metadata editing explains how.

Where the time actually went

Spread across the four days, the rough split in this illustrative project was about a quarter of the effort on intake, calibration and raw conversion, about a third on the six reference images, a quarter on applying and fine-tuning the remaining 42, and the rest on quality control and the review round. The mixed-light lifestyle frames took disproportionately long, roughly as much time as all the white-background front views combined, because each had a slightly different mix of window and strobe. That split is typical of well-run tonal jobs and a useful planning guide: if a quote allocates almost nothing to the reference stage, expect revisions.

Cost for work like this depends on the number of images, how many distinct lighting conditions the shoot contains, whether raw files and color targets are available, and how many colorways must match physical samples. A set shot in one session with a gray card in frame is quick. A set assembled from several shoots, with no targets and a demanding brand team, is not. Our image editing and graphic design service page explains how we scope and quote this kind of project.

When Curves and Levels Work Needs Outside Help

A single image with a flat histogram and a mild cast is well within reach of anyone who has read this far. The reference guidance draws the line in a specific place: bring in help when product colors or skin tones are inconsistent across a shoot. That is the situation this illustrative project described, and it is where a structured process pays for itself. One-off corrections are forgiving; a set of dozens or hundreds that must match each other and a physical sample is not.

Signs you have crossed the line

• The same product looks different across images and correcting one makes it disagree with the next.
• Mixed lighting means one white balance cannot fix the whole set.
• Brand colors must match a physical reference or a specification, and "close enough on my laptop" is not an acceptable standard.
• Skin tones vary between frames in a campaign where the same person appears several times. Tonal consistency matters as much as retouching; our guide to eye and teeth retouching covers the detail work that follows once tone is right.
• Output goes to both web and print and the two need different endpoints.

What to ask a retouching partner

Ask how they set targets before they start, whether they work on calibrated displays, whether they deliver layered masters or only flattened files, and how they handle revision rounds on a colorway. A partner who answers in terms of reference images, samplers and adjustment layers is working in the way this article describes. One who talks only about turnaround time may be applying a preset and hoping. The guide on choosing a retouching partner goes through the full set of questions, and our image editing and retouching articles cover the neighboring techniques in the same level of detail.

The bottom line is that curves and levels reward order more than skill: set the endpoints first with the clipping preview on, remove casts with channel curves measured on true neutrals, add contrast with a gentle S-curve on Luminosity, keep every step on an adjustment layer in 16 bits, and judge the result on a calibrated screen against a real reference. Do that on one reference image per variant and apply it across the set, and consistency follows.

Where this comes from

• Adobe Help Center — Curves adjustment
• Adobe Help Center — Levels adjustment

The figures and practices above come from the sources listed.

Working on something like this?

We take on Image Editing & Retouching work for teams who want it done once, properly. Tell us what you are building and we will tell you honestly whether we are the right studio for it. Start a project.

Where to go next

• Blend Modes: What Actually Works
• Dodge and Burn Retouching, Done Properly