Colour grading a photo to match a reference image

First, separate what you can copy from what you cannot

Before touching a slider, be honest about which parts of the reference are transferable. A grade can move colour and tone. It cannot move a light source, change a lens, or restage a scene.

If the thing you actually admire about the reference is a soft window light falling across a face, no amount of grading will get you there from a flash-lit frame. Recognising this in the first minute saves the evening.

Read the reference before you copy it

Open the reference next to your photo and work out what it is doing, in this order:

1. Where are the blacks? Sample the darkest region. If the darkest pixel reads around 8–12% rather than 0%, the reference has a lifted toe, and that alone accounts for much of what people call "cinematic".
2. Where are the whites? A reference whose brightest area sits at 92% rather than 100% has a rolled shoulder. Together with the toe this defines the tonal envelope.
3. What colour are the shadows? Sample a dark area and read the RGB balance. Blue higher than red means cool shadows; the reverse means warm.
4. What colour are the highlights? Same reading in a bright area. Shadows and highlights leaning opposite ways is a split tone, which is the most common single ingredient in a deliberate grade.
5. What is happening to skin? Sample a midtone skin patch. Note the hue and how saturated it is relative to everything else.
6. How saturated is the rest? Many references that read as "rich" are actually selectively saturated — one or two hues up, everything else down.

Build it in the right order

Grade from the bottom of the pipeline upward: tonal envelope, then global colour, then per-hue, then finishing. Working out of order means re-doing earlier decisions, because every stage changes what the next one appears to do.

1. Match the tonal envelope with the point curve. Lift the toe and roll the shoulder to the levels you read off the reference. Do this before any colour work; it changes every colour reading.
2. Match the global cast with Temp and Tint, judged on a neutral in your photo, not on the overall impression.
3. Split the tones in Color Grading. Set shadow hue and highlight hue first with saturation at zero so you can see the hue clearly, then bring saturation up until it is one notch less than you think you want. Use Balance to decide where the split point falls, and Blending to control how hard the transition is.
4. Shape individual colours in the Color Mixer. Foliage and sky are usually the two that need the most work, and skin usually needs the least.
5. Finish with grain and vignette to match the reference's texture, then step away for five minutes and look again.

When the reference and your photo are the same frame

The whole method above exists because you are inferring a transform you cannot observe. If the reference happens to be your own photo, edited — someone graded your file, or you have a JPEG export whose settings you lost — then the transform is observable, and you should observe it instead of guessing.

Feed both frames to PresetForge and it derives the per-channel curves directly from the pixel correspondence, along with the per-band HSL shifts, colour grading, grain and vignette. Exact before/after extraction is available on the free plan, so this costs nothing to try.

When the reference is somebody else's photo

Then there is no pixel correspondence and no measurement is possible — the differences between the two images are a mixture of grading and of the scene itself, and nothing can separate those two without a reference frame. What you can get is a statistical read of the reference: its tonal distribution, its colour balance across the tonal range, its saturation by hue. That is a genuinely good starting point and a bad finishing point. Take it, then tune by eye against the checklist above.

Related, and worth reading before you try: extracting the colour grade from a film still covers the specific traps in film references, which are more numerous than people expect.