
Every frame has a hierarchy of attention, and the eye travels to the sharpest, highest-contrast line it can find. That is why a crisp rectangle of footage dropped on top of a soft background reads as an edit rather than a composition. Softening a perimeter — a vignette on a talking-head shot, a feathered crop for a vertical cutdown, a diffused title card, or a blurred matte that blends a keyed subject into a new scene — tells the viewer where to look and hides the seams.
After Effects remains the reference tool for this work because edge treatments have to be frame-accurate. A feather that looks perfect on a locked-off shot can crawl on a moving one. A vignette that works at 1080p can band at 4K. A matte that holds on a plain wall can fall apart on hair. With generative video tools now producing entire shots, the useful question is no longer which tool blurs edges, but where edge control should happen: at generation time or in post. This guide answers that with concrete settings, a repeatable workflow, and a decision framework.
There are four reliable approaches, and they solve slightly different problems. Pick by asking whether you need a soft boundary on a layer, a soft transition inside a layer, or a soft matte between two layers.
Draw a mask on the layer, then raise Mask Feather. At 1080p, a subtle vignette usually sits between 60 and 180 px; at 4K, double those numbers. Mask Expansion is the underrated control here: pulling the mask in by 20–60 px after feathering removes the hard rim that feathering alone can leave at the frame edge.
Be aware that mask feather is per-axis. An ellipse stretched across a 16:9 frame feathers unevenly — soft top and bottom, tighter left and right. If you need an even falloff, use a radial gradient as a luma matte instead, or split the difference by drawing a circular mask on a square pre-comp and scaling it.
Apply Gaussian Blur, Fast Box Blur, or Camera Lens Blur to an adjustment layer, then mask the adjustment layer so the blur only touches the edges of the frame. Fast Box Blur is the workhorse for speed; Camera Lens Blur is the choice when you want an iris shape and a physically plausible falloff, which matters for anything imitating real depth of field. Radius values that read as "soft" rather than "broken" are typically 20–60 px at 1080p, 40–120 px at 4K, but the real test is watching playback at 100% zoom, not at fit-to-window.
Instead of masking the blur, mask the area you want to keep sharp and use it as a Track Matte (Alpha or Luma) on the blurred layer. This is cleaner when the sharp region is an odd shape, such as a person plus a product on a table. For finer control, the Set Matte effect lets you feed a matte from a different layer without reordering the stack, and it supports invert and "use for matte" logic that track mattes cannot express.
If a blurred element lives inside a pre-comp and you scale that pre-comp up in the parent composition, feathering and blur can stair-step. Toggle Collapse Transformations (the sun icon) on the pre-comp layer so effects, masks, and vector artwork are evaluated at the resolution of the final render. This single switch fixes most "my feather looks chunky when scaled" complaints.
This is the most requested edge treatment and the easiest to get wrong. A workable recipe:
If banding appears in the corners, switch the project to 16-bit color depth before you touch anything else. Most banding in slow gradients is a bit-depth problem, not a blur problem.
Fringes almost always come from premultiplied versus straight alpha. If a masked layer shows a dark halo when blurred, add Remove Color Matting or Shift Channels to strip the matte color, then blur only the alpha channel by placing a Blur effect on a duplicate and using it as an Alpha Track Matte. For keyed footage that needs tightening, apply Simple Choker with a small negative value (around −1 to −2) before the blur so the soft edge does not include leftover background pixels.
Blur produces smooth ramps, and smooth ramps expose 8-bit quantization. Fix it in this order: set the project to 16-bit or 32-bit float, add a Noise effect at 0.5–1.5% with the blend mode set to Soft Light or Overlay to dither the ramp, and only then consider a blur with a wider iris (Camera Lens Blur) instead of a box blur. Grain is not a compromise here; it is the standard fix.
Feathering grows in both directions, so a soft edge eats into whatever is inside the matte. Two masks solve this: a large, heavily feathered outer mask, and a hard inner mask set to Intersect so the sharp subject is protected. Where a person's silhouette matters, combine this with the Roto Brush for the inner matte and keep the outer mask as pure falloff. Bilateral Blur is another option when you want smoothing that respects edges, though it is slow and rarely necessary for vignettes.
Type and vector art need a different approach because their edges are already mathematically clean, and blurring them directly destroys brand legibility. The standard trick is to duplicate the text layer: put a heavily blurred copy behind in a muted color for a soft glow, and keep the top copy perfectly sharp. Animate Mask Feather rather than opacity when you want a title to fade in from a soft edge — keyframing feather from 0 to 140 px keeps the letters intact while the boundary dissolves.
For lower thirds, blur the background plate under the graphic instead of the graphic itself. Place an adjustment layer with a blur above the footage, mask an area slightly larger than the graphic, and feather the mask at 80–140 px. This creates the standard broadcast look where the text sits in a soft pocket of defocused video, and it survives a change of graphic without re-rendering the whole composite.
Generative and AI-assisted tools move edge work earlier in the pipeline. Instead of adding depth of field in post, models such as Runway, Pika, Luma, and Kling render bokeh and diffusion as part of the image, which means the falloff is baked into the frame rather than composited on top of it. The result is often more organic: light wraps around a subject, foreground blur bleeds naturally, and a shallow-focus look appears without a single mask.
Some AI features map directly onto classic post tasks:
The limits are equally real. Generative output is inconsistent between shots, so a diffusion treatment that looks right in one clip may shift in the next. Most models do not deliver true alpha channels, so anything that needs to sit over live footage still ends up in a compositor. And frame-exact timing — a logo reveal, a lower third synced to speech — is still manual work.
| Situation | Better fit |
|---|---|
| Frame-exact timing, brand typography, motion graphics | After Effects |
| Organic depth of field in a generated shot | AI generation with baked diffusion |
| Hair-level mattes on live footage | AI matting pass, then After Effects refinement |
| Vertical or square cutdowns from a locked frame | AI outpainting, then a feathered crop in post |
| Series that must look identical shot to shot | Manual templates in After Effects |
| Fast turnaround on a single social clip | AI tools end to end |
Two criteria usually decide it. First, does the edge need to match a previous shot exactly? If yes, build a preset in After Effects and reuse it. Second, does the deliverable need real alpha? If yes, plan on a compositor regardless of how the shot was generated.
AI generation and manual edge control are not competitors; they are sequential stages. A workflow that holds up in production looks like this:
Is feathering the same as blurring an edge?
Functionally, yes for a matte: feathering creates a gradient of opacity along a boundary, which is what makes an edge look soft. The difference is scope. Feather applies to a mask and affects the layer's transparency, while a blur effect changes pixel values. For a soft vignette over a single layer, feathered masking is usually cheaper and cleaner. For softening detail inside the frame, use a blur.
What is the fastest way to blur only the edges of a video clip?
An adjustment layer with Fast Box Blur, masked with a feathered inverted ellipse, takes under a minute and previews in real time. If you need it across many clips, save it as an animation preset and apply it in one click.
How much feather should I use at 4K?
Start around 240–360 px for a full-frame vignette and 120–180 px for a smaller element such as a lower third pocket. Then judge on a real display at full resolution, because the only reliable measurement is whether the transition is visible as a line or reads as a gradient.
Why does my blurred layer show a dark outline?
Almost always a matting color issue. The layer was premultiplied against black or another color, and blurring pulled that color into the soft edge. Add Remove Color Matting, or blur the alpha channel separately and recombine it with the untouched color channels.
Can AI tools create soft edges without any manual masking?
For generated shots, yes — depth of field and diffusion can be produced by the model itself, and AI matting handles most background removal. What AI does not reliably give you is consistency across a series, exact alpha output, or frame-accurate control over a branded graphic. Those still belong in a compositor.
Should I use Bilateral Blur or Gaussian Blur for edges?
Gaussian or Fast Box Blur for softness around a perimeter, because you want the gradient. Bilateral Blur is for smoothing surfaces while preserving detail, and it is a poor fit for vignettes. If your goal is a physically plausible lens look, Camera Lens Blur with a defined iris is the better third option.
Edge control is a small skill with an outsized effect on how professional a finished video feels. Learn one reliable vignette recipe, one matte-cleanup routine for fringes, and one noise-based fix for banding, and you can handle the vast majority of real-world edge problems. Then decide per project whether the softness should come from the model that generated the frame or from a layer you control by the pixel — and use both when the deadline and the brand both matter.

Alexander
Vidu Q4 alternatives compared: Kling 4.0, Seedance 2.5, Wan 3.0, and Runway Gen-4 — 4K support, input modes, durations, audio, and what each costs in credits on Domer.
The top 4K AI image generators on Domer compared: Flux 3 Klein, Nano Banana 2.1, GPT Image 2.5, GPT Image 2, and Nano Banana Pro — 4K credit costs, reference limits, and speed.
Nano Banana 2.1 vs GPT Image 2.5 Flare compared: credit costs per resolution, aspect ratios, reference images, text rendering, and which tasks fit each model.