Every Pillow ImageFilter, shown on the same image

The original

Everything below starts from this

The adjustable filters

These take arguments, so you can tune the strength. For real work, use these.

`GaussianBlur(3)`

A proper Gaussian blur. The one to reach for

`BoxBlur(3)`

Cheaper than Gaussian, slightly blockier

`UnsharpMask(2, 150, 3)`

Sharpening with radius, strength and a threshold

`MedianFilter(5)`

Removes speckle while keeping edges

`MinFilter(5)`

Erode. Dark areas grow

`MaxFilter(5)`

Dilate. Light areas grow

`ModeFilter(5)`

The most common value nearby. Good on flat art

`RankFilter(5, 2)`

The general form of the three above

`Kernel((3,3), [0,-1,0,-1,5,-1,0,-1,0])`

Your own 3 by 3 kernel

The fixed filters

No arguments, small built in kernels. Quick effects rather than tools.

`BLUR`

A fixed 5 by 5 blur. Mild, and not adjustable

`CONTOUR`

Outlines on white, like a pencil tracing

`DETAIL`

A gentle local contrast boost

`EDGE_ENHANCE`

Lifts edges a little

`EDGE_ENHANCE_MORE`

The same, harder

`EMBOSS`

Flat grey with raised edges

`FIND_EDGES`

Edges only, on black

`SHARPEN`

A fixed sharpen kernel

`SMOOTH`

A light averaging

`SMOOTH_MORE`

The same, stronger

How to apply one

from PIL import Image, ImageFilter

im = Image.open("photo.jpg")

soft = im.filter(ImageFilter.GaussianBlur(3))
crisp = im.filter(ImageFilter.UnsharpMask(radius=2, percent=150, threshold=3))
edges = im.filter(ImageFilter.FIND_EDGES)

The adjustable ones are called with arguments, the fixed ones are passed as they are. Both return a new image.

Which to use when

Blurring anything: GaussianBlur. BLUR and SMOOTH are fixed and weak.
After resizing down: UnsharpMask with a small radius. Shrinking always softens, and this puts the crispness back.
Scanned or noisy images: MedianFilter, which kills speckle without smearing the edges the way a blur would.
Cleaning up masks: MinFilter and MaxFilter to shrink or grow the white areas.
Anything else: your own Kernel. The kernel designer lets you try values and copy the code.

Filtering a mode P image filters the palette indexes rather than the colours, which produces nonsense. Convert to RGB first. See concepts.

The ImageFilter module page covers the arguments in detail.

The filter gallery

The original

The adjustable filters

GaussianBlur(3)

BoxBlur(3)

UnsharpMask(2, 150, 3)

MedianFilter(5)

MinFilter(5)

MaxFilter(5)

ModeFilter(5)

RankFilter(5, 2)

Kernel((3,3), [0,-1,0,-1,5,-1,0,-1,0])

The fixed filters

BLUR

CONTOUR

DETAIL

EDGE_ENHANCE

EDGE_ENHANCE_MORE

EMBOSS

FIND_EDGES

SHARPEN

SMOOTH

SMOOTH_MORE