Image Sizes That Work: Cropping, Watermarks, Grids

2026-11-057 min readImage Tools5 tools covered

One photo behaves completely differently as an avatar, a cover and a grid tile. The difference is not quality but ratio and safe area. Set the ratio, then the pixel size, then the file weight - any other order wastes work.

Key takeaways
  • The crop box is dragged on a preview, but export recalculates with that preview scale, so what you get is always the original-resolution region
  • The 3x3 grid only splits an image into thirds and never squares it; to get square tiles, lock 1:1 first and then split
  • Text watermark font size is in pixels rather than relative units, so one set of settings looks completely different on portrait and landscape images
  • Tiled watermarks start at the top-left with fixed spacing, so the right and bottom edges keep an incomplete gap and are never centred or cropped
  • A share cover should be 1200x630 in PNG, because WeChat, X and Facebook all reject SVG thumbnails

#Get the order right or the rest is wasted

Image work has a fixed order: ratio first, then pixel size, then file weight. Reversing it costs real quality - compress before cropping and you have already thrown away detail; crop before deciding the ratio and the composition is gone.

Each step answers exactly one question, and mixing them creates dead ends:

StepWhat it settlesWhat goes wrong
RatioComposition and safe areaHalf a face cropped away, huge empty bands
Pixel sizeDelivery specThe platform re-compresses and sharpness is lost
File weightTransfer costDetail already lost, and cropping only makes it worse

#Ratio first, pixel size second

Ratio determines composition and pixels determine sharpness, yet platforms validate pixels - so people instinctively change pixels first. The result is a stretched or cropped image with the composition destroyed.

Pick the ratio by purpose, then fit pixels onto it:

  • Avatar: 1:1. Faces usually sit in the upper third, so put the eyes near the upper third line
  • Share cover: 1.91:1, commonly 1200×630. The left third is the text safe area; keep the subject right
  • 3×3 grid: 1:1 per tile, 3:3 as a set
  • Article header: 2.35:1, commonly 900×383
  • Video thumbnail: 16:9, commonly 1280×720
Tip

One rule for pixel size: twice the display size is enough. An avatar shown at 200×200 needs 400×400. Uploading 2000×2000 only slows the upload and the platform compresses it again. Extra pixels do not improve appearance, they only add transfer cost.

#Cropping: always the original resolution

The crop box looks small on the preview, but export recalculates the coordinates with the preview scale - so the exported region is always at original resolution. People often assume a small preview means a small export.

Two behaviours to know with the image cropper:

  • Export format is always PNG: lossless, keeps transparency, but often two to three times larger than the source JPEG. Compress afterwards rather than looking for a format switch inside the cropper
  • Export re-encodes, so EXIF is gone: capture time, camera model and GPS disappear. Good for privacy, worth noting for archives

Also, rotation is already applied when the browser decodes the photo - the orientation you see in the preview is the orientation you get, with no manual correction needed.

#The 3x3 grid splits, it does not square

The grid only divides the whole image into thirds; it never squares anything. This is the most common misunderstanding: a 4:3 source produces nine 4:3 tiles that reassemble into the original, not nine squares.

For square tiles, the correct order is two steps:

  1. Lock the ratio to 1:1 and crop, which fixes the composition
  2. Then split into the grid

Division is by whole pixels with the remainder absorbed on the right and bottom, so all nine tiles match in size and none ends up a pixel narrower.

Warning

This tool does not produce a ZIP archive - tiles are downloaded one at a time. Fine for a single set of nine, painful for dozens of images. That scale is better handled on the command line, where ImageMagick's -crop writes all nine files in one go.

#Watermarks: position versus opacity

A watermark has two opposing jobs: visible enough to prove ownership, restrained enough not to ruin the image. Position and opacity are the two knobs that trade one against the other.

Four things worth deciding before using the watermark tool:

  • Tile it if it must survive cropping. A single corner mark disappears with one crop; a tiled pattern leaves traces
  • Text size is in pixels, not relative units. One set of settings looks completely different on portrait and landscape images
  • Logo width is a percentage of the source width, which adapts far better than a fixed pixel size
  • Export a small region to check anything below 15% opacity. The preview canvas caps at 520 pixels wide, so pixel-level sharpness is unreliable and very faint marks become invisible in preview but obvious in the output

Tiling has one built-in consequence: it starts at the top-left with fixed spacing (about 0.8 of the font size for text, 0.3 of the width for logos), so the right and bottom edges keep an incomplete gap and are never centred or cropped. If that bothers you, adjust the spacing or use a single centred mark.

Note

The font is fixed at bold sans-serif and cannot be changed. For a nicer typeface, render the text to a transparent PNG in a design tool and apply it as a logo - which is how brand watermarks are normally made anyway.

#Format and weight: what comes after cropping

The cropper exports PNG, usually two to three times larger than the source JPEG, so a compression pass is almost always required next. That step cannot come earlier.

With image compression, shrinking pixels beats lowering quality by a wide margin: going from 4000 to 1600 wide saves far more than dropping quality from 90 to 70.

When switching formats with the image converter, remember the three hard limits:

  • PNG is lossless and keeps transparency, and the quality slider does nothing for it
  • JPEG is lossy and cannot hold transparency; a white background is painted first, so transparent areas become white rather than black
  • WebP is usually smallest, but a low-quality image converted to WebP can actually grow

The interface reports an increase or decrease in bytes - read the number, not just the direction. An already-compressed image growing is normal.

#Which tool to use

What you needToolKey constraint
Crop an avatar or cover to ratioImage cropperExport is PNG, EXIF is dropped
Split into a 3x3 gridImage cropperLock 1:1 first for square tiles
Add a brand watermarkWatermark toolPixel-based font size; tiling leaves an edge gap
Shrink the cropped PNGImage compressionPixels beat quality
Switch to a lighter formatImage converterThe quality slider is inert for PNG
Screenshot a web page as artworkWeb screenshotFull-page shots get very tall; use the viewport mode
Package finished images into a documentImages to PDFEmbeds as JPEG, slight loss on text

For self-made visuals the order is: capture or take the source - crop to ratio - add watermark - compress - deliver. One job per step, so a problem is easy to walk back.

Key point

With web screenshot, note the privacy boundary: the image is produced by a third-party service, so the URL leaves your browser. Never paste intranet addresses or links carrying tokens or signatures. Full-page mode captures everything below the fold and produces a very tall image; the viewport mode is better for thumbnails, cropping to roughly 16:9 by width times 0.9.

#Frequently asked questions

#My avatar crop always cuts off part of the face. Why?

Ratio and position both matter. Lock 1:1 first, then move the frame up so the face sits in the upper third. Most platforms render avatars as circles, so the corners get cut away - which is the safe area concept: keep anything important away from the edges.

#The 3x3 grid does not give me squares.

Because it only splits into thirds and never changes the ratio: a 4:3 source gives nine 4:3 tiles. Lock 1:1 and crop first, then split. Remember it does not bundle a ZIP, so tiles download one by one.

#The image looks blurry after adding a watermark.

Export is PNG and therefore lossless, so blur usually comes from two places: the preview canvas is only 520 pixels wide, so what you see there does not reflect the output; or the watermark itself is too small and too transparent and blends together when scaled down. Export a small region to check.

#Why does my file get bigger when I compress it?

Three possibilities: it was already compressed and has nothing left to save, you converted to PNG (lossless re-encoding always grows), or you converted a small low-quality image to WebP, which is not always smaller. Read the percentage the interface reports rather than the direction.

#Why do social platforms show no thumbnail?

Usually format or dimensions. Share cards expect PNG or JPEG at around 1200×630; SVG is not supported by WeChat, X or Facebook. The image URL must also be an absolute https:// address, because crawlers cannot resolve a relative path on your site.

Tools used in this guide

Related guides