For a whole talking-head episode, a skin mask that smooths inside each face is the better tool. Frequency separation protects texture, and it's the technique people call the "real" way to retouch, usually with a little attitude. On moving footage, though, it turns into a repair job for one shot at a time.
A VanityFilter result at close-up range. Drag the handle across the cheek. The skin is much smoother and much of the pore texture is gone, while the red patch is still there, a little more even. The lashes, brows and fine hairs at the temple stay sharp. For more pore texture, lower Strength or raise Detail restore.
Short version: frequency separation is a still-image technique. The published video version is a repair workflow. One Resolve tutorial tracks a reference frame, freezes it, paints the fix by hand, and places the cleaned patch back over the shot. That's a tool for one shot. For footage across an episode, a skin mask that smooths inside the face and adds detail back covers every frame without a per-shot repair.
How frequency separation works
The idea's simple. Any image can be split into two stacked layers. The low frequency layer holds broad information: color, tone, shadow, the blotchy red patch on a cheek, the shape of the light. The high frequency layer holds fine detail: pores, stubble, fine lines, individual hairs, the grain of the sensor.
Retouchers build it by blurring a copy of the image to make the low frequency layer, then subtracting that blur from the original to isolate what's left, which is the texture. Once the layers are separated, you can paint out a blotchy patch on the low layer, and the texture on top stays where it was. That's why the result doesn't go plastic. The detail was never blurred. Only the color underneath it got fixed. If you retouch stills in Photoshop, our guide to Photoshop skin smoothing software sets frequency separation against the Skin Smoothing neural filter and Portraiture.
It's also why the technique has a following. It treats uneven color and unwanted texture as two separate problems, which they are, and gives each one its own fix. Our guide to smoothing skin in video explains the main routes and how to keep the result natural.
Why it doesn't port cleanly to video
Frequency separation has no concept of time. It works on one image. There's no frame two.
You can run the same math on every frame of footage. The problem is what comes next. The value of frequency separation comes from a person painting on the low frequency layer, sampling nearby clean skin, and deciding about one patch of one image. Repeat that on every frame and you're painting the whole clip by hand. Do it once, hold it, and the fix slides off the face the moment the subject moves.
The published video version is a repair workflow, not a grading step. The clearest example is an editorial Fusion walkthrough for DaVinci Resolve 20. It starts by setting a reference frame and tracking it with a planar tracker, then bakes that track into a planar transform. Next it freezes the frame in a Time Speed node, separates it into high and low frequency layers with blur and Channel Booleans nodes, paints the color and clones the texture by hand, and masks the clean patch. The planar transform goes in just before the final merge, so the cleaned plate lands back over the moving footage. It's a tutorial, not a one-click Resolve feature.
Blackmagic's Fusion product page confirms the parts: a node-based workflow, retouching among its uses, and a built-in planar tracker. It doesn't mention frequency separation, so the technique comes from the tutorial, not from Blackmagic.
You wouldn't run it on 40 minutes of two-camera footage. A planar track assumes the tracked area stays on one flat plane. The tutorial draws its polygon around texture on the same physical plane. A head turn takes the skin off that plane, and hair crossing the face hides it. So each shot needs its own paint and its own track. The result is a fix for one blemish in one shot.
The tell
If a tutorial promises frequency separation on video and never mentions a freeze frame, a clean plate, or a tracker, it's either a plain high-pass sharpen with a new name, or a photo tutorial with "video" in the title.
How a masked smoothing pass works in VanityFilter
A masked pass attacks the same problem from the mask side. VanityFilter detects the faces in each frame, builds a mask for each one from the face outline, forehead and neck, and keeps only the pixels that read as skin. It smooths inside that mask. The mask is recomputed on every frame, and the Mask stability setting blends each mask with the previous frame to reduce flicker on moving subjects.
Under the hood, the engine uses some of the same high and low frequency ideas. It runs an edge-preserving bilateral filter inside the mask, a Texture repair step that pulls residual texture toward a blurred base, and a Detail restore step that adds fine detail from the original frame back in. The difference is where it runs: on every frame, inside the face mask, with the amounts you set.
VanityFilter leaves eyes, lips, hair, hats, glasses, clothing and background untouched. Lips, eyebrows and eyes are never smoothed. Protect eyelids and Protect ears are on by default.
Each detected face gets its own mask and its own sampled skin color, up to five faces per frame. The second guest doesn't share the first guest's skin sample. The sliders are shared, though. Strength, Detail restore and the rest apply to every face you smooth. The only face-selection controls are Primary face only, which smooths just the largest, most central face, and Ignore small faces, which skips faces below a set fraction of the frame height. A fallback detector catches some turned and profile heads. If both detectors lose a face, its mask fades out.
Where frequency separation still wins: a masked pass doesn't paint a repair on one patch. Blemish removal heals small acne and blemish spots it detects inside the mask, and results vary with skin tone and acne type. Reduce redness evens out red blotches and flush toward the person's own skin tone, also inside the mask. Neither control lets you pick one spot by hand. For one specific spot in one hero shot, a manual repair is still the more precise tool.
| Frequency separation | Masked smoothing | |
|---|---|---|
| Works on | One frame at a time | Every frame automatically |
| Handles motion | Planar track, clean plate | Mask re-detected every frame |
| Texture | Untouched by design | Depends on the settings |
| Best at | One blemish, one shot | Even skin, long takes |
| Cost | Manual work per shot | Set once, then render |
| Multiple people | Setup repeated per face | Per-face masks, shared sliders |
When frequency separation is still the right call
- A single distracting mark on a close-up that viewers will lock onto.
- A near locked-off shot where a tracked clean plate will hold.
- Product or beauty work where one hero frame gets pulled as a still anyway.
- A fix that's about color and blotchiness only, with texture that's already fine.
If you already grade in Resolve, doing the repair on the Fusion page keeps everything in one project. Our Resolve skin smoothing comparison covers Face Refinement, a DaVinci Resolve Studio feature, next to VanityFilter.
A practical workflow for long footage
- Set exposure and white balance at the shoot. Smoothing only works on the pixels it gets, and a highlight clipped to pure white has no skin texture left to keep.
- Open a representative clip in VanityFilter and check the Mask overlay view. It's the fastest way to check coverage. The Detected faces count should match the people in the shot.
- Start from the Standard preset, the balanced everyday setting. Lower Strength until pores still read at normal size.
- Process the source files in a batch before you cut. Each file gets one pass, and every clip you cut from it inherits the result. If you need the mask in your editor, export the mask video in the same run. The skin matte guide for Premiere Pro and Resolve covers how to use it. For editing, export ProRes 422 HQ (.mov). For delivery, use H.264 (.mp4).
- Cut the episode. If one close-up still has a mark that pulls focus, that single shot is where frequency separation earns its keep.
- Check the uploaded version, not only the export. Platform compression also removes fine detail, so a pass that looked right in your editor can look different after upload. Our YouTube compression guide covers what to look for.
Doing the correction before the timeline means one pass per source file instead of one per clip. Our pre-edit workflow guide covers the batch setup.
How to check you didn't flatten the texture
Whatever route you take, the check is the same. Play the result at speed, at the size people will watch it. Then step through a few frames and look at four places: the pores across the cheek, the fine hairs at the temple and jaw, the edge where skin meets hair, and the background just past the face. Texture should be readable, hairlines should be crisp, and nothing outside the skin should have moved.
Then watch for temporal wobble, which stills never show. If the correction breathes or shimmers as the head turns, the mask is unstable. Raise Mask stability in the Mask shape panel to reduce that, then check the Mask overlay on the turn. Flicker gives the work away much faster than a slightly heavy setting does.
FAQ
Can you do frequency separation on video?
Yes, but it's a repair technique, not a per-frame beauty pass. One published DaVinci Resolve Fusion tutorial tracks a
reference frame with a planar tracker, freezes that frame, separates it into high and low frequency layers, paints the
fix by hand, and applies the planar transform so the cleaned plate sits back over the moving shot. Use it for a specific
blemish in a specific shot, not across an entire episode.
What is the difference between frequency separation and AI skin smoothing?
Frequency separation splits an image into a low frequency layer holding color and tone and a high frequency layer
holding texture such as pores and fine hair, so you can fix blotchy color without touching the detail. It's manual and
frame-based. AI skin smoothing detects faces, builds a skin mask, and applies a smoothing pass inside that mask
automatically on every frame. Frequency separation gives you more control per frame. A masked pass gives you consistent
treatment across every frame of a long take.
Does frequency separation avoid the plastic skin look?
It avoids it by design, because in the basic method the texture layer is left alone while you work on color and tone.
Plastic skin comes from blurring the texture away along with the blotches. A masked pass can avoid it too, if it keeps
the fine detail and stays inside the skin. Judge any tool by whether pores and facial hair survive at normal viewing
size.
How do you retouch skin across a whole video without redoing it for every clip?
Process the source files once, before the edit, instead of applying an effect to every clip on the timeline. Every cut
you make afterward uses the corrected file, so you don't re-apply the effect to each clip instance.
VanityFilter processes video locally and masks each detected face. A one-time license is $49, currently on sale for $29.95, with no subscription. See current pricing.