ai video color grading — how to grade ai clips

this post contains a sponsored link to invideo.

ai video color grading works the same way as grading any other footage: balance each clip, then put one look over all of them. the part that's different is the starting point. a generated clip arrives with a look already baked in, and that look can change every time you generate.

this guide covers why ai clips need grading, what to fix first, and how to put one consistent grade across a whole sequence.

why ai clips need grading at all

a camera records what's in front of it, and a colorist shapes it afterwards. an ai model decides the colour for you as part of generating the image. that has two consequences.

the look is baked in. there's no log or raw file underneath, just a finished video. you're correcting an image that has already been graded once, by the model, so small adjustments go further than big ones.

it drifts between generations. each clip is generated on its own. two clips from the same prompt can come back with different white balance, contrast and saturation. on its own, each clip looks fine. cut three of them together and the differences jump out, which is usually the moment people realise they need to grade.

if you're building a sequence out of several generations, whether you're creating ai videos from a text prompt or animating stills, plan for a grading pass the same way you'd plan for one on a shoot.

step 1: balance each clip first

before any look, get the clips to agree on the basics. work on one clip at a time:

1. white balance. pick something that should be neutral, like a white wall or a grey sky, and remove any cast. this is the single biggest source of mismatch between generations.
2. exposure and contrast. bring the midtones to a similar brightness across clips. the histogram guide covers what to look for.
3. saturation. some generations come back noticeably punchier than others. pull the loud ones down to match the quiet ones rather than the other way round.

keep these moves small. generated clips are compressed, and large pushes in exposure or saturation show up as banding in skies and smooth gradients.

step 2: match the clips to a hero shot

pick the clip that looks best and treat it as the reference. match the others to it rather than to an idea in your head.

in luttie, color match does this from a frame. load a frame from the clip you're fixing, drop a still from the hero clip in as the reference, and set the strength somewhere between 50 and 80 percent. it compares the overall colour and brightness of the two images and moves yours toward the reference. it copies the palette, not individual objects, so it works best when the two shots are similar scenes. color match is part of luttie pro.

the free way to do the same thing is by eye: put the two frames side by side and adjust white balance and the color wheels until they sit together. how to match colors between clips walks through it.

step 3: put one look over everything

once the clips agree with each other, apply one creative look to all of them. a lut is the easiest way, because the same file gives every clip exactly the same treatment.

• start from a look. free luts or a pro lut pack give you a finished starting point. for a film feel, faded film and teal and orange are the two looks people ask for most.
• or build your own. grade a frame from your hero clip in the editor until it's right, then save it as a .cube file with export lut (pro). see exporting luts.

then apply that lut to every clip in your editor. using luts in other apps covers the main ones.

if you'd rather not use an editor at all, luttie can render the graded clip directly. drop the mp4 in, grade it, and use export under video (pro) for the full clip. the free preview renders six seconds around the frame you're on, with a watermark, so you can check the grade in motion first. see working with video.

what a grade can't fix

grading changes colour and tone. it won't fix things the model got wrong in the image itself: warped hands, flickering textures, lighting that changes direction mid-shot. if a clip has those, regenerate it before you spend time grading it.

and go easy on film effects. grain helps hide compression in generated footage, but heavy halation or bloom on top of an already-stylised ai image tends to look like a filter rather than a film.

summary

• ai clips arrive pre-graded, with a look that drifts between generations
• balance white balance, exposure and saturation per clip first, gently
• match every clip to your best one, by eye or with color match
• put one lut over the whole sequence so it reads as one piece
• regenerate clips with image problems rather than grading around them

grade a frame from your clip in the editor →