Adding a watermark looks like “put a line of text on the picture”, yet it often fails both ways: too faint and it vanishes in a thumbnail, too strong and it ruins the picture, and tucked into a corner it can be cropped off in one move. The cause is rarely taste. It is usually a misunderstanding of what the controls mean: what font size is measured against, which point an anchor refers to, and what tiling spacing is relative to.
This guide goes through the ToolGarden watermark controls one by one, with numeric examples, then compares how well each layout resists removal and shows how to diagnose common failures. Everything runs locally in the browser; the image and the watermark asset are not uploaded.
1. Decide the purpose, then the layout
| Purpose | Suggested layout | Starting values (rules of thumb, adjust per image) |
|---|---|---|
| Credit or attribution | Single, nine-grid bottom right or bottom left | Opacity 50% to 70%, font size around 4% to 8% of image width |
| Deterring reuse | Diagonal or tiled across the whole image | Opacity 20% to 35%, moderate spacing so the subject is covered too |
| Draft or confidential marking | Diagonal, large text such as CONFIDENTIAL | Opacity 25% to 40%, with a stroke so it reads on light and dark areas |
| Brand logo | Image watermark, single or tiled | A transparent PNG, about 10% to 25% of image width |
These are common starting ranges, not standards. A thumbnail used on a web page and a large preview image can need font sizes that differ many times over.
2. What the controls actually are
| Control | Range or options | Notes |
|---|---|---|
| Type | Text / Image | Text supports size, bold, italic, color and stroke; an image watermark needs an uploaded logo |
| Font size | 12 to 200 px | Measured in source-image pixels, not as a ratio |
| Stroke | 0 to 12 px | Default 0 (no stroke); the stroke color is black |
| Image watermark size | 5% to 150% of image width | Keeps the logo’s aspect ratio |
| Opacity | 5% to 100% | Applies to the whole stamp; default 50% |
| Rotation | −180° to 180° | Default −20° |
| Layout | Single / Tile / Diagonal | Single can use the nine-grid or free dragging |
| Spacing (tile and diagonal) | 0% to 120% horizontally and vertically | Relative to image width and image height respectively |
| Output format and quality | JPG / PNG / WebP; quality 50% to 100%, default 92% | Export size equals the original; nothing is resized |
The default text is CONFIDENTIAL: 48 px, white, bold, 50% opacity, −20° rotation, a single stamp in the center. Limits: images up to 50 MB and 40 million pixels, one image at a time.
3. The procedure
- Upload an image and watch the live preview. The preview is scaled to the display size, while the export draws at the original resolution.
- Choose a text or image watermark, then type the text or upload the logo.
- Choose a layout. A single stamp can be placed with the nine-grid or dragged anywhere in the preview; tiled and diagonal layouts use the spacing sliders for density.
- Tune font size or image size, opacity and rotation.
- Choose the output format and quality, then click Apply watermark & download. To adjust, change the parameters and download again.
4. Three geometry details you need to understand
1. Font size is in source pixels, not a ratio
A 48 px watermark on an 800 px wide image takes about 6% of the width and looks fine. On a 3000 px wide camera original it takes only about 1.6%, and on a screen it is nearly invisible. The same settings behave completely differently on images of different resolution, so re-tune the font size when the image size changes a lot. A reasonable starting point for a single corner mark is 4% to 8% of image width, which is roughly 120 to 240 px on a 3000 px wide image.
2. Nine-grid anchors position the center of the stamp
Each nine-grid position is a relative coordinate for the stamp’s center: left, middle and right are 8%, 50% and 92% of the width, and top, middle and bottom are 10%, 50% and 90% of the height. The consequence is that a stamp on the bottom-right anchor has its center only 8% of the width from the right edge, and if the stamp is wider than about 16% of the image width, part of it is drawn outside the picture and cut off.
For example, on a 3000 px wide image the right-hand anchor puts the center at x = 2760, 240 px from the right edge, so anything wider than about 480 px is truncated. Take text like © ToolGarden, around 12 characters (average character width roughly 0.6 of the font size, as a rough estimate): at 60 px it is about 430 px wide and just fits, but at 150 px it is about 1080 px wide and a big piece falls off the picture. The fix is to drag the stamp to a suitable spot in the preview rather than relying on the grid.
3. Tiling spacing is a proportion, not pixels
In tiled and diagonal layouts the step between neighbors is: horizontal step = stamp width + horizontal spacing × image width; vertical step = stamp height (about 1.2 times the font size) + vertical spacing × image height. Both steps are at least 20 px. The defaults are 35% horizontally and 28% vertically.
Example: a 3000×2000 image with the default 48 px CONFIDENTIAL (roughly 375 px wide, using the 0.6-times-font-size estimate). The horizontal step is about 375 + 0.35 × 3000 ≈ 1425 px and the vertical step about 58 + 0.28 × 2000 ≈ 618 px. That gives around two stamps per row and three or four rows, each only 375 px wide, so most of the picture is empty and easy to crop around. Raise the font to 120 px and set spacing to 10% horizontally and 12% vertically: the horizontal step is about 940 + 300 ≈ 1240 px and the vertical step about 144 + 240 ≈ 384 px, a dense layer whose stamps are large enough to land on the subject. Spacing is proportional so the same numbers keep a similar density on images of different sizes.
Diagonal versus tile
- Tile: stamps form a regular grid and the Rotation slider sets the angle.
- Diagonal: rotation is fixed at −30° and the Rotation slider has no effect; every other row is offset by half a step, giving a staggered brick pattern that is harder to cut away with one straight crop.
5. How hard is each layout to defeat?
| Layout | Cropped around | Removed by AI inpainting | Impact on the picture |
|---|---|---|---|
| Single, corner | Very easy; crop one side | Very easy; a small rectangle is what inpainting does best | Small |
| Single, centered, semi-transparent | Cannot be cropped out | Harder, though a stamp over the subject hurts repair quality | Medium; covers the subject |
| Tile (grid) | Hard | Harder; the whole image must be processed | Larger |
| Diagonal (staggered) | Hardest | Hardest; the most area is covered | Largest; control it with low opacity |
To be honest about it: no visible watermark is impossible to remove. Its job is to raise the cost of reuse and leave a source hint, not to replace copyright protection. Small watermarks are exactly what AI inpainting handles best; the watermark removal article explains why. For stronger proof, combine it with original files, shooting records or platform copyright registration instead of relying on the mark alone.
6. Color, stroke and format
- On images with mixed light and dark areas, white text with a thin black stroke is the safest. The default stroke is 0, so raise it yourself; at around 48 px try 2 to 4 px first.
- In JPG output, small text edges can show noise. Keep quality at about 85% or higher, and use PNG when crisp watermark text matters.
- If the source is a PNG with transparency and you export JPG, the transparent areas are filled with white before the watermark is drawn. Export PNG or WebP to keep transparency.
- The watermark is drawn into the pixels, so it cannot be toggled or moved afterwards. Keep an unwatermarked original.
- The export is a new file re-encoded through a Canvas, and the original EXIF data is not kept.
| Output format | Good for | Watch out for |
|---|---|---|
| JPG | Photos, web publishing | Lossy; transparency becomes white |
| PNG | Screenshots, keeping transparency, further editing | Large files |
| WebP | Web publishing with small files | A few older programs cannot open it |
7. Common problems and how to diagnose them
| Symptom | Cause | Fix |
|---|---|---|
| The watermark is nearly invisible | Font size is in source pixels and looks tiny on a large image; or opacity is too low | Set the font to 4% to 8% of image width; raise opacity above 50% |
| Part of the watermark is outside the picture | The nine-grid positions the center and the stamp is too wide | Drag it into place, or reduce the font size |
| Tiling is sparse and can be cropped around | Default spacing is proportional to image width and looks sparse on large images | Reduce the spacing and increase the font size |
| Tiling is too dense and covers the subject | Spacing too small and opacity too high | Increase the spacing and drop opacity to 20% to 30% |
| Non-Latin text turns into boxes or mixed fonts | The tool uses the system default font and the device lacks the right one | Use a device with the fonts installed, or use an image watermark |
| A logo watermark looks blurry | The source logo has too few pixels and is being enlarged | At 25% of a 3000 px image the logo is drawn about 750 px wide, so supply a larger source |
| The exported PNG is huge | PNG is lossless and compresses photos poorly | Use JPG or WebP for photos |
8. When not to use this approach
- You need verifiable ownership: a visible mark is no substitute for copyright registration, original files or content credentials, and it cannot prevent removal.
- Processing hundreds or thousands of images: the tool handles one at a time, so scripts or professional software suit bulk work better.
- Invisible watermarks or digital fingerprints: these are outside what a visible watermark can do.
- PDF documents: use PDF Watermark, which works page by page.
- Photographic originals whose full metadata must survive: export re-encodes the image and drops EXIF.