Transformations

Resize, crop, convert, color-correct, blur, rotate, remove the background, or strip EXIF, all from the image URL. No re-uploading, no extra storage. Append parameters and each distinct result is cached at the edge worldwide.

How it works

Every upload and media response includes a url field: a direct CDN link on every plan. Append query parameters to transform what it serves.

text

# The url, untouched
https://test.src.img.pro/4j2/abc12345.jpg

# WebP, 800px wide, a touch brighter
https://test.src.img.pro/4j2/abc12345.jpg?format=webp&w=800&brightness=1.1

One source can produce many cached variants with no per-transform charge. Each distinct URL is cached independently and parameter order doesn’t matter. Safeguards may ignore unsupported values or fall back to the source/default rendition.

Named sizes

The shortcut: ready-made variants you request with ?size=, no parameters to pick. The three responsive sizes appear in the response’s sizes object. social is requested the same way.

?size=s short side 320px: Thumbnails and avatars. Aspect ratio preserved.
?size=m short side 640px: Inline previews and feed cards. Aspect ratio preserved.
?size=l short side 1080px: Full-width hero and detail views. Aspect ratio preserved.
?size=social 1200×630: Fixed OpenGraph / social-card frame. Any source aspect is padded to fit.

text

https://test.src.img.pro/4j2/abc12345.jpg?size=m

Format

Convert on the fly. format overrides the extension in the URL path, so …/abc.jpg?format=webp serves WebP even though the path says .jpg.

format jpg · png · webp · avif · gif: webp is the best quality-for-size balance (recommended); avif is the smallest; png is lossless with transparency; jpg is universal; gif is for legacy or animation. auto is intentionally unsupported: a cached transform is shared across clients, so it can’t depend on one client’s Accept header.

text

https://test.src.img.pro/4j2/abc12345.jpg?format=webp

Resize & crop

Resize and crop with explicit dimensions. With the default scale-down fit, images are never enlarged: requesting a larger size is a no-op.

w 1–4096 px: Target width. Set w or h alone to scale proportionally; set both with fit to control how the image fills the box.
h 1–4096 px: Target height. Same rules as w.
fit default scale-down: How the image fills w×h: scale-down, contain, cover, crop, pad, squeeze. cover/crop fill and trim; contain/pad letterbox; scale-down never enlarges. An explicit cover/crop/pad will upscale to fill the box.
gravity default auto: Which part to keep when cropping: auto, face, left, right, top, bottom. face centers on detected faces.
zoom 0–1: Crop tightness around gravity=face. 0 = most context, 1 = tight.

text

# 400×400 square crop centered on faces
https://test.src.img.pro/4j2/abc12345.jpg?w=400&h=400&fit=cover&gravity=face

Quality

Control output quality for lossy formats (JPEG / WebP / AVIF). Accepts a number or a named level.

q 1–100or high · medium-high · medium-low · low: Higher = better and larger; ~80 is a good web default. Named levels map to roughly 90 / 75 / 60 / 45. No effect on PNG: PNG output is always lossless truecolor, so use a lossy format for a smaller file.

text

https://test.src.img.pro/4j2/abc12345.jpg?format=webp&q=80

Color & tone

Each is a decimal multiplier centered at 1 (no change). Omit a param to leave that channel untouched.

brightness 0–10 · default 1: 0 = black, 2 = twice as bright. Useful range ~0.5–2.
contrast 0–10 · default 1: 0 = flat mid-gray, >1 = punchier. Useful range ~0.7–2.
gamma 0–10 · default 1: <1 lightens midtones, >1 darkens them (0 and 1 are both no-ops). Useful range ~0.5–2.5. Direction is opposite to what some editors call “gamma”.
saturation 0–10 · default 1: 0 = grayscale, >1 = vivid. Useful range ~0.5–2.

All four are capped at 0–10, but values near the top produce visually useless extremes. Stay inside the “useful range” unless you’re after a specific effect.

text

# Subtle warmth
https://test.src.img.pro/4j2/abc12345.jpg?brightness=1.1&contrast=1.05&saturation=1.1

# Black and white
https://test.src.img.pro/4j2/abc12345.jpg?saturation=0

Filters

Soften or sharpen detail without re-uploading.

blur 0–250 · default 0: Gaussian blur radius in pixels. 10–20 for a subtle wash, 40+ for a heavy placeholder. Useful for hero overlays and privacy masking.
sharp 0–10 · default 0: Unsharp-mask intensity. 1–3 subtle, 4–6 noticeable, 7+ aggressive (can ring on smooth areas).

text

# Blurred hero background
https://test.src.img.pro/4j2/abc12345.jpg?blur=15

# Detail boost
https://test.src.img.pro/4j2/abc12345.jpg?sharp=4

Orientation

Rotate, mirror, or trim borders. Rotation at the supported angles involves no resampling; the output is re-encoded in the requested format like any transform (use PNG for a fully lossless result).

rotate 90 · 180 · 270: Clockwise rotation in degrees. Other values are ignored.
flip h · v · hv: Mirror horizontally, vertically, or both.
trim border: Auto-crop a uniform border color around the image (scanned documents, letterboxed screenshots).

text

# Fix a sideways photo, then mirror it
https://test.src.img.pro/4j2/abc12345.jpg?rotate=90&flip=h

Privacy

Cameras and phones embed GPS, device info, and timestamps. Control what travels with the image.

metadata default copyright: copyright keeps copyright/author tags and strips GPS + device info (the default). none strips everything (recommended before sharing any photo publicly). keep passes all original EXIF through, including GPS; use only when you need it preserved.

text

# Strip all EXIF before sharing a phone photo
https://test.src.img.pro/4j2/abc12345.jpg?metadata=none

Background

Remove the background with a single param (image segmentation runs on the edge) and optionally fill it with a solid color.

segment foreground: Replace the background with transparency. Add format=png so the alpha channel survives; otherwise the output follows the path extension and JPEG flattens transparency.
background: Fill color for transparent areas: a CSS named color (white, navy), hex (URL-encode # as %23, e.g. %23f5f5f5), or rgb(…) / rgba(…). After segment=foreground it swaps the background; on its own it fills any transparent pixels. Omit it and alpha-capable formats stay transparent, while others fill white.

text

# Cutout on a white background (Amazon / Shopify ready)
https://test.src.img.pro/4j2/abc12345.jpg?segment=foreground&background=white&format=jpg

# Transparent PNG cutout
https://test.src.img.pro/4j2/abc12345.jpg?segment=foreground&format=png

Effects

One-param composites: the subject is separated from the background with the segmentation model, the effect is applied to the background only, and the sharp subject is layered back on top.

fx blur-bg · color-pop · darken-bg: blur-bg = portrait-style background blur; color-pop = grayscale background, full-color subject; darken-bg = dimmed background, spotlit subject.
strength per recipe: Tunes the effect (only meaningful alongside fx). blur-bg: blur radius 0–250, default 60. darken-bg: background brightness 0.05–1, default 0.5 (lower = darker). color-pop: no strength (the background is fully desaturated). Out-of-range values clamp.

text

# Portrait-mode blur behind the subject
https://test.src.img.pro/4j2/abc12345.jpg?fx=blur-bg

# Spotlight: dim the background to 30%
https://test.src.img.pro/4j2/abc12345.jpg?fx=darken-bg&strength=0.3

Effects follow the URL’s output format like any transform: the path extension, or format= when set (JPEG in the examples above). The composite fills every pixel, so no alpha is needed; use format=png only if you want lossless output. Effects compose with sizing (w/h/fit), q, and tile; other adjustment parameters (rotate, blur, brightness, …) are ignored while fx is set. Each fx + strength combination is computed once and cached at the edge.

Watermark

Repeat one of your own images across the base as a tiled overlay, the standard way to watermark. Upload the watermark once (typically a transparent, seamlessly-tiling PNG with opacity baked in), then reference it by id.

tile image id: The id of another image you host, repeated edge-to-edge at its native size over the base. The overlay must be in the same workspace as the base; an id that isn’t yours (or doesn’t exist) is ignored and the base is served unchanged.

text

# Tile your watermark (id wm0listo1) across a photo, served as WebP
https://test.src.img.pro/4j2/abc12345.jpg?tile=wm0listo1&format=webp

The overlay is tiled axis-aligned at its native pixel size, so design it as a seamless tile with the angle, color, and opacity baked in. PNG output is always lossless truecolor.

Combining transforms

Chain parameters freely: order doesn’t matter, and the CDN runs everything in a single pass. (The one exception is effects: while fx is set, only sizing, quality, format, and tile apply.)

text

# Privacy-blurred preview with EXIF stripped
https://test.src.img.pro/4j2/abc12345.jpg?blur=40&metadata=none

# Faded social card
https://test.src.img.pro/4j2/abc12345.jpg?format=jpg&contrast=0.9&saturation=0.8

# Rotated, sharpened, exported to AVIF
https://test.src.img.pro/4j2/abc12345.jpg?rotate=90&sharp=3&format=avif

Caching & formats

Caching: Each unique URL is cached independently at the edge. A cache miss generates the requested rendition; later matching requests reuse the cached result.
Upscaling: The default scale-down fit never enlarges beyond the source; cover / crop / pad upscale to fill the requested box.
Accepted uploads: JPEG, PNG, GIF, WebP, AVIF, and supported HEIC originals up to 20 MB (20,000,000 bytes). The API proves decoder capability before creating the image.