All posts
By Pravit Gandhi··11 min read

How to match video to an edited still photo

How do you match video to an edited still photo? Why it is harder than camera to camera matching, and the step by step workflow that gets you there.

Matching video to an edited still is a different job from matching two cameras, and the reason is that the still is already a finished picture. It is display referred; your log video is not. So the order is: normalize the video and only the video, read the still on scopes before you trust your eyes, match tone before color, then neutralize and match white balance, then hue and saturation by region. All the video tutorials you find are about matching two different cameras, which is why they stall you halfway.

It is different, and the difference is not a detail. Matching camera A to camera B is matching two recordings of the same kind, which is its own job with its own method, covered in why your cameras don't match. Matching video to an edited still is matching a recording to a finished picture. Those are not symmetric problems, and the reason most attempts stall halfway is that people treat them as if they were.

Why this is harder than it sounds

Four things are working against you before you touch a control.

The photo is display referred and the video is not. Your exported still is almost certainly sRGB, and sRGB is defined against a specific viewing situation. The W3C specification sets its reference conditions at a display luminance of 80 candela per square meter in an ambient illuminance of 64 lux, described as a dim viewing environment. That is a standard for how a picture should appear on a screen in a room. Log video is the opposite kind of thing: an encoding designed to preserve a range for later interpretation. ARRI's color FAQ describes Rec.709 material as having a display specific encoding, the "what you see is what you get" case, and that is the category your photo already belongs to. The video has to be brought into that category before comparison means anything.

The two standards are closer than they look and further apart than that suggests. Image Engineering's technote makes the point that sRGB and Rec.709 share identical primary chromaticities and the same D65 white point, and the W3C specification confirms sRGB is built on the ITU-R BT.709 reference primaries. What differs is the transfer function. So a photo and a video frame can be describing color with the same three primaries and still land in visibly different places, and no amount of hue adjustment fixes a tone response mismatch. Fix tone first or you will chase it forever.

The photo has local edits and the video has global ones. A raw photo edit is full of things that have no equivalent in a normal video grade: a gradient over the sky, a brush on one cheek, an HSL move that only touched the oranges, a lens profile correction, a clarity slider that is really a local contrast operation at a specific radius. Some of that is reproducible in video with windows and qualifiers. Some of it is not worth reproducing, and knowing which is which is most of the skill.

The still is one frame and the video is thousands. Photos are chosen. You picked the frame where the light was right and the expression worked. The video includes the two seconds before and after, where a cloud moved. A match tuned on one frame and never checked at the ends of the shot will fall apart in exactly the place you did not look.

None of that makes the job impossible. It makes it a sequence, which is what the rest of this is.

Step one: put the still where you can see it against the video

Import the exported photo into the project as a still image so it lives on the timeline or in the gallery, not in a separate window on a second screen with different color handling. Comparing a picture in a photo viewer against a video in an editor compares two different color pipelines, and you will spend an hour chasing a difference that only exists because two applications disagree.

Resolve's color page gallery is built for exactly this. Blackmagic describes it as a place to organize, share and re-use grades, and it holds reference stills too. The same page has a split screen mode that displays multiple full frames in the viewer at once, which is how you want to do the actual comparison. Side by side beats memory. Nobody's eye holds a color for more than a couple of seconds.

Export the photo at the size you will actually compare at, and export it once. Re-exporting mid-process with a different setting is a classic way to lose a match you already had.

Step two: normalize the video, and only the video

The still is finished. Do not touch it. Everything moves on the video side.

Apply the correct input transform for whatever the camera shot: the ACES Input Transform, a color space transform, or your editor's automatic log detection. The Academy's documentation frames Input Transforms as the step that converts camera native data into ACES2065-1, a single common encoding, before anything creative happens. Blackmagic describes color management in the same terms, as controlling conversion between cameras, monitors, broadcast displays and projectors, with support for both its own system and ACES.

After this step the video should look like a normal, slightly flat, correctly exposed picture. It should not look like your photo yet, and if it does, something has applied a creative LUT you did not intend. If it still looks washed out and gray, no transform has been applied at all, which is a separate diagnosis in why your S-Log3 footage looks washed out, with the conversion routes themselves compared in S-Log3 to Rec.709.

Step three: read the still on scopes before you trust your eyes

Put the still up and look at it on the waveform, the parade and the vectorscope. This is the step almost everybody skips and it is the one that turns a vague matching exercise into a set of targets.

Write down four things. Where do the darkest meaningful pixels sit on the waveform, and where do the brightest. How do the three channels sit relative to each other in the shadows and in the highlights on the parade, since that is where a photo's color grading usually lives. Where does skin sit relative to the vectorscope's skin tone line, and how far out. And how saturated is the image overall, in the sense of how far the vectorscope trace spreads from the center.

Those four numbers are your brief. Everything below is hitting them.

Step four: match tone before color

Contrast first, because contrast changes apparent color and color does not change contrast.

Set the video's black point and white point to match the still's, using the waveform rather than the picture. Photo edits very often lift the blacks, so your video's shadows may need to come up rather than down, which feels wrong and is correct. Then match the midtone placement, which is where most of the perceived difference between a photo and a video frame actually lives, and check it on the face rather than on the whole frame.

Do not use a curve to do all three at once yet. Lift, gamma and gain, in that order, checking the waveform each time. When the two waveforms have similar shapes you are most of the way home even though the picture may still look wrong in color.

Step five: neutralize, then match white balance

Find something in both images that should be neutral. A wall, a road, a white shirt, a gray card if you were lucky enough to shoot one.

Balance the video so those neutrals sit with the three channels together on the parade. Then, and only then, introduce whatever white balance offset the still has, because most edited photos are not neutral. They are deliberately warm or cool, and often warm in the highlights and cool in the shadows. Reading it off the parade tells you which, and by how much.

Doing it in one move instead of two is how people end up with a video that is the right temperature overall and wrong in the shadows.

Step six: match hue and saturation by region

Now the parts a global control cannot reach.

Work in this order: skin, then the dominant color of the scene, then everything else. For skin, compare the angle and length of the vectorscope trace between still and video, and correct it with a hue versus hue and hue versus saturation adjustment rather than by rotating the whole image. For the dominant color, usually foliage or sky or a branded object, use a qualifier. Blackmagic describes qualifiers as selecting parts of the image by hue, saturation or luminance, and power windows as shapes drawn around specific objects, with a tracker to follow movement.

If your photo edit used a gradient over the sky, this is where you reproduce it with a window rather than pretending a global move will do. If it used a brush on one part of a face, consider whether the video needs it at all. Usually not.

Step seven: check it in motion, and at both ends of the shot

Play the shot. Then scrub to the first frame and the last frame and compare each against the still.

Anything keyed with a qualifier is the first thing to fail here, because the key was built on the frame you were looking at. Anything windowed is second. A match that only holds on one frame is not a match, it is a screenshot.

What reliably fails

Automatic white balance matching on mixed light. If the scene has daylight through a window and tungsten from a lamp, there is no single white balance that is correct, and any tool that solves for one will get half the frame right and half wrong. That includes the eyedropper in your editor and it includes automated matching. The manual answer is to balance for the light on the subject and use a window to handle the other source separately.

Matching a photo that was heavily denoised. Photo denoising is aggressive by video standards, and a video frame next to it will look noisy even when the color is a perfect match. That is a texture difference, not a color one, and no amount of matching fixes it.

Matching grain, halation and lens character. If the still got its look partly from a film emulation, part of what you like is not color at all. Reproducing the color and skipping the texture is the honest version, and the gap is usually smaller than people fear once the tone is right.

Matching from a photo exported in a different color space than you think. If a photo comes out of a raw editor in ProPhoto or Adobe RGB and lands in a video pipeline expecting sRGB, everything will read oversaturated and the match will fight you the whole way. Export the reference in sRGB and confirm the file is tagged before you start.

Questions that come up halfway through

Why does my video look more saturated than the photo even after matching?

Usually a color space mismatch on the reference rather than a grading error. Confirm the still is sRGB and tagged as such. If it is, the next likely cause is that the video's transfer function is doing something different in the upper midtones, so recheck the waveform match before touching saturation.

Can I just make a LUT from the photo?

Only loosely. A LUT is a fixed mapping from input color to output color, and it can encode the tone curve and global hue relationships of your photo look reasonably well. It cannot encode the local edits, and it will only behave correctly on footage in the same input state it was built against. Build it from normalized footage, apply it only to normalized footage.

How do I match video to a photo when the lighting changed between them?

You do not, entirely. Match tone and white balance, accept that the direction and quality of light are not gradeable, and decide which parts of the photo look you actually wanted. Most of the time the answer is the tone curve and the skin rendering, both of which travel fine.

Do I have to do this for every shot in the edit?

For a scene, no. Match one hero shot to the still properly, then propagate that grade to the rest of the scene and adjust per shot. That is the same normalize-then-grade split covered in do you have to color grade every clip.

The automated path, and what it does not do

Reference image matching is the automated version of all of the above: hand a tool a picture, hand it footage, and let it solve for the transform between them. This is the space photo-first tools have occupied for a while, and it is also the center of our own product, so treat the next paragraph as disclosure rather than review.

Leumos AI is a browser-based AI color grading studio: upload your edit, and it detects scene cuts, grades every shot to match a reference image, and renders 4K ProRes or H.265 in the cloud — no GPU and no install required. The reference image is exactly the case this page is about, an edited still rather than another clip, and the input side is ACES based, with transforms for Canon Log, Log 2 and Log 3, S-Log2 and S-Log3, ARRI LogC, V-Log, F-Log, N-Log, Apple Log, Log3G10, BMDFilm and GoPro Protune. The part that matters at edit length is that the match is applied per shot across the whole timeline rather than to one frame, which is the difference between a look and a graded film.

What it does not do is worth saying with the same directness. It will not reproduce local edits from your photo, because those are decisions about specific pixels in a specific frame. It will not add grain, halation or lens character that came from a film emulation. It does not edit, so picture has to be locked elsewhere first. And it is in closed beta as of Q3 2026, so what exists today is the waitlist at leumos.ai.

The wider category, including what these tools can and cannot transfer, is covered in what AI color grading actually means, and the browser-based options generally in the best browser color grading tools.

Sources