Short answer

draw = ImageDraw.Draw(img), then draw.rectangle((x0, y0, x1, y1), outline="red", width=3), draw.ellipse(...) with the circle's bounding box, draw.line([(x, y), ...], width=...). For semi-transparent fills, draw on a transparent RGBA layer and Image.alpha_composite() it onto the image.

The basic shapes

draw_shapes.py
from PIL import Image, ImageDraw

img = Image.open("photo.jpg")
draw = ImageDraw.Draw(img)                                   # draws on img itself

draw.rectangle((640, 420, 830, 820), outline="yellow", width=6)          # box around the rocket
draw.ellipse((890, 530, 950, 590), outline="#22d3ee", width=5)           # circle on the water tower
draw.line((80, 1000, 600, 760), fill="white", width=8)
draw.polygon([(1200, 300), (1300, 480), (1100, 480)], fill=(239, 68, 68), outline="white")
draw.rounded_rectangle((1150, 900, 1550, 1020), radius=30, fill=(15, 23, 42))
draw.text((1350, 960), "Launch Pad 39B", fill="white", anchor="mm", font_size=36)
img.save("shapes.jpg")
Tested on ✓ Linux ✓ Windows ✓ macOS
The launch pad photo with a yellow box around the rocket, a cyan circle on the water tower, a white line, a red triangle and a dark rounded label
Real output of the script above.

Every shape takes a bounding box or a list of points in pixels from the top-left corner. Colours can be names, hex strings or tuples. A circle is an ellipse whose box is square: centre (cx, cy) and radius r give (cx - r, cy - r, cx + r, cy + r), or use draw.circle((cx, cy), r).

How the box coordinates work

box_rules.py
import numpy as np
from PIL import Image, ImageDraw

img = Image.new("L", (30, 30), 0)
ImageDraw.Draw(img).rectangle((2, 2, 9, 9), fill=255)
xs = np.nonzero(np.asarray(img))[1]
print("rectangle((2, 2, 9, 9)) covers x =", xs.min(), "to", xs.max(), "->", xs.max() - xs.min() + 1, "pixels wide")

img = Image.new("L", (30, 30), 0)
ImageDraw.Draw(img).rectangle((5, 5, 24, 24), outline=255, width=4)
print("outline width=4, row 12 lit at x =", np.nonzero(np.asarray(img)[12])[0].tolist())

try:
    ImageDraw.Draw(img).rectangle((24, 24, 5, 5), fill=255)
except ValueError as e:
    print("reversed box: ValueError:", e)
Tested on ✓ Linux ✓ Windows ✓ macOS
Output
rectangle((2, 2, 9, 9)) covers x = 2 to 9 -> 8 pixels wide
outline width=4, row 12 lit at x = [5, 6, 7, 8, 21, 22, 23, 24]
reversed box: ValueError: x1 must be greater than or equal to x0

Unlike crop(), drawing boxes are inclusive: both the first and last pixel are painted, so (2, 2, 9, 9) is 8 pixels wide, not 7. An outline's width grows inwards from the box edge, so the box's outer size never changes with the line width. And the first corner must be the top-left one.

Semi-transparent fills

The alpha value in a colour does three different things depending on the image:

transparent_fill.py
from PIL import Image, ImageDraw

blue = (0, 0, 255)
red_half = (255, 0, 0, 128)                                     # 50% transparent red

rgb = Image.new("RGB", (10, 10), blue)
ImageDraw.Draw(rgb).rectangle((0, 0, 9, 9), fill=red_half)
print("RGB image:                ", rgb.getpixel((5, 5)))       # alpha ignored: solid red

rgb = Image.new("RGB", (10, 10), blue)
ImageDraw.Draw(rgb, "RGBA").rectangle((0, 0, 9, 9), fill=red_half)
print("RGB image, Draw(.., RGBA):", rgb.getpixel((5, 5)))       # blended

rgba = Image.new("RGBA", (10, 10), blue + (255,))
ImageDraw.Draw(rgba).rectangle((0, 0, 9, 9), fill=red_half)
print("RGBA image:               ", rgba.getpixel((5, 5)))      # replaced: now see-through!
Tested on ✓ Linux ✓ Windows ✓ macOS
Output
RGB image:                 (255, 0, 0)
RGB image, Draw(.., RGBA): (128, 0, 127)
RGBA image:                (255, 0, 0, 128)

On an RGB image the alpha is silently ignored. ImageDraw.Draw(img, "RGBA") blends instead. On an RGBA image, drawing replaces pixels, alpha included, so a 50% fill punches a half-transparent hole into an opaque image. The reliable pattern is a separate transparent layer, composited on top:

highlight.py
from PIL import Image, ImageDraw

photo = Image.open("photo.jpg").convert("RGBA")
overlay = Image.new("RGBA", photo.size, (0, 0, 0, 0))           # transparent layer to draw on
d = ImageDraw.Draw(overlay)
d.rectangle((640, 420, 830, 820), fill=(250, 204, 21, 90), outline=(250, 204, 21, 255), width=5)
d.ellipse((870, 520, 970, 620), fill=(34, 211, 238, 90), outline=(34, 211, 238, 255), width=5)

result = Image.alpha_composite(photo, overlay).convert("RGB")
result.save("highlight.jpg")
Tested on ✓ Linux ✓ Windows ✓ macOS
The sunset photo with a translucent yellow box over the rocket and a translucent cyan circle over the water tower
Real output: highlights drawn on a transparent layer, then composited.

Smooth (anti-aliased) circles

ImageDraw does not anti-alias shapes, so curves and diagonals have jagged edges. Draw at 4× size and shrink:

antialiased_circle.py
from PIL import Image, ImageDraw

SCALE = 4
size = 120

jagged = Image.new("RGB", (size, size), "white")
ImageDraw.Draw(jagged).ellipse((10, 10, size - 11, size - 11), outline="black", width=3)

big = Image.new("RGB", (size * SCALE, size * SCALE), "white")    # draw 4x larger...
ImageDraw.Draw(big).ellipse((10 * SCALE, 10 * SCALE, (size - 11) * SCALE, (size - 11) * SCALE),
                            outline="black", width=3 * SCALE)
smooth = big.resize((size, size), Image.LANCZOS)                 # ...then shrink: smooth edges

sheet = Image.new("RGB", (size * 2 + 10, size), "white")
sheet.paste(jagged, (0, 0))
sheet.paste(smooth, (size + 10, 0))
sheet.resize((sheet.width * 3, sheet.height * 3), Image.NEAREST).save("aa.png")   # enlarged to show the pixels
Tested on ✓ Linux ✓ Windows ✓ macOS
Two circles enlarged three times: the left with stair-stepped edges, the right with smooth grey edge pixels
Real output, enlarged 3×. Left: drawn directly. Right: drawn at 4× and resized down.

Text is the exception: it is anti-aliased already.

Thick lines and corners

thick_lines.py
from PIL import Image, ImageDraw

img = Image.new("RGB", (620, 200), "white")
d = ImageDraw.Draw(img)
points = [(30, 160), (120, 40), (210, 160), (300, 40)]
d.line(points, fill="#2563eb", width=24)                                  # default joints: notches at corners
d.line([(x + 300, y) for x, y in points], fill="#2563eb", width=24, joint="curve")
img.save("joints.png")
Tested on ✓ Linux ✓ Windows ✓ macOS
Two thick blue zigzag lines: the left with notched corners, the right with rounded joints
Real output. Left: default joints. Right: joint="curve".

A labelled box (as in object detection)

textbbox() measures the label before drawing, so the background tag fits the text exactly:

labelled_box.py
from PIL import Image, ImageDraw

img = Image.open("photo.jpg")
d = ImageDraw.Draw(img)
box, label = (640, 420, 830, 820), "rocket 0.97"

d.rectangle(box, outline=(250, 204, 21), width=5)
left, top, right, bottom = d.textbbox((box[0], box[1]), label, font_size=28, anchor="lb")
d.rectangle((left - 4, top - 6, right + 6, bottom), fill=(250, 204, 21))        # tag sized to the text
d.text((box[0], box[1]), label, fill="black", font_size=28, anchor="lb")
img.crop((560, 340, 960, 860)).save("label.jpg")
print("text box:", (left, top, right, bottom))
Tested on ✓ Linux ✓ Windows ✓ macOS
Output
text box: (640, 400, 782, 420)
Close-up of the rocket inside a yellow box with a yellow tag reading rocket 0.97 above its top-left corner
Real output (cropped).

The ImageDraw reference lists every drawing method; the text guide covers fonts, centring and wrapping.

GuideAdd text to an image ReferenceImageDraw, with tested examples GuideOverlay a transparent PNG onto an image

The outputs and images on this page are the real results of running every example with Pillow 12.3.0 on Python 3.13 (Linux) on 2026-10-15. Every example is also re-run automatically on macOS (Darwin 25.6.0, Pillow 12.3.0), Linux (Linux 6.17.0-1022-azure, Pillow 12.3.0), Windows (Windows 2025Server, Pillow 12.3.0); the line under each example shows the result per operating system. The DejaVu Sans font files sit next to each script so font examples behave the same everywhere. Sample photos: NASA/Sam Lott, public domain. Spotted something wrong? Tell us.