Depth DOF — a lens, not a blur

Most "depth of field" in an image editor is a blur with a shape drawn over it. You mask an ellipse around the subject, feather it hard, and push gaussian at everything outside. It works right up until something inconvenient happens — a shoulder leaves the ellipse, an arm crosses it, a railing runs from the subject's hand to the horizon — and then the picture tells on you. The railing goes soft in the middle of its own length. Nothing in the frame is being asked where it is.

Depth DOF asks. Focus follows the scene, not a shape you drew on the frame.

The circle of confusion

The chain is short and it's the real one. V3KTR infers a depth map from your single image. For every pixel, the effect works out how far that pixel sits from the focal plane, and turns that distance into a circle of confusion — the size of the blur disc a real lens would throw for a point at that distance:

coc = pow(max(|depth − Focus| − FocalRange/2, 0), Falloff)

Everything inside FocalRange has a circle of confusion of exactly zero, which is what a depth of field actually is: a band of acceptable sharpness, not a plane. Outside it, the disc grows, and Falloff decides how fast — the difference between a long lens that drops off a cliff and a gentle one that lets go slowly.

Then the pixel is gathered through that disc. Where the radius is small the original comes through untouched, so anything in focus stays genuinely sharp rather than slightly-blurred- and-sharpened-back.

The aperture is an aperture

The disc isn't a gaussian. It's a hole, and you get to say what shape the hole is.

Samples are laid down on a golden-angle spiral, which fills a disc evenly without the spokes you get from a naive ring. Each one is tested against a polygonal aperture boundary built from Aperture Blades — 3 to 8, six by default, like a real iris. Wind Blade Roundness up and the polygon relaxes toward a circle; wind it down and out-of-focus highlights go hexagonal, the way a cheap fast lens does at f/4. Aperture Rotation turns the whole iris, and Aperture Feather softens its edge.

Two more that only exist because real lenses have flaws worth stealing:

• Catadioptric hollows out the centre of the aperture, so points render as rings. That's the mirror-lens donut bokeh — the one that makes backlit foliage look like a field of little zeros.
• Anamorphic squeezes the disc on one axis into an oval, which is where the widescreen look actually comes from.

Why the highlights matter

Bokeh isn't blur. Bokeh is what a bright point does when it's out of focus, and if you weight every sample equally you don't get it — you get mush.

So samples are weighted by brightness, cubed, scaled by Highlights. A specular hit on a wet road, a fairy light, the sun through leaves: those blow up into discs of the aperture's own shape while the rest of the frame stays civilised. This is the parameter that decides whether the picture looks photographed or filtered, and it's worth pushing further than feels sensible on anything with points of light in it.

Samples sets how many taps build each disc — 8 to 256. Low is fast and grainy, high is smooth and costs real time; the governor scales the count so it stays interactive while you work. Bring it up before you export, not while you're scrubbing.

What to set

Focus first, and set it by eye on the depth, not the picture. Focal Range next — this is the one people leave too narrow, then wonder why an ear is soft. Then Blur Strength (1–49px), which is the only parameter that's purely taste.

The honest caveat: this is inferred depth, so it's an estimate of a scene that was flat before you got here. It reads a hand in front of a chest correctly. It can get confused by glass, by hair against a busy background, by anything where a human would also have to look twice. The gather is graded to keep the model's edge imperfections out of sight, and mostly you won't find the seam — but if a shot fights you, Invert Depth and a wider Focal Range are the two knobs worth trying before you blame the lens.

Then pull it

It's an ordinary effect, so FX MOTION can animate it — and this is the one everyone reaches for first. Arm Focus, run it from front to back, and you have a rack focus on a still photograph: attention moving through a scene that never had a third dimension. Slow. Much slower than you think. The whole trick is that it looks like someone's hand on a barrel.

For the other things the same depth read gives you, see Depth AO, which uses it to find the crevices, and Depth Parallax, which moves the camera instead of the lens.