A visible watermark does two jobs that pull against each other. It has to say who made the picture, and it has to stay put when someone would rather it did not. A signature tucked into a corner does the first job well and the second hardly at all. A bold mark across the middle survives a crop but spoils the photo for the people you wanted to impress. And a batch of photos stamped identically has a weakness of its own, one that researchers turned into an automatic removal method in 2017.
Decide Which Job the Mark Is Doing
Before choosing a size or a spot, settle what the mark is for. A credit line on a photo you are happy to see shared only needs to be legible and to survive resizing. It can be small, in a corner, fairly opaque, and nobody minds it. A proof sent to a client for approval, or a preview in a shop where the full file is sold, needs to make the unpaid copy unusable. That calls for a mark across the subject or a tiled pattern, at an opacity that leaves the picture readable but clearly not finished.
Most disappointment with watermarks comes from asking a credit line to do the work of a proof mark. A corner signature tells honest people where the photo came from. It does not stop anyone else, because cropping it off takes seconds.
What Each Placement Costs to Remove
Rough effort to get rid of each kind of mark, and what it does to the photo for everyone else.
| Placement | To remove it | Effect on the photo | Suits |
|---|---|---|---|
| Small, in a corner | One crop | Barely noticed | Credit on shared work |
| Along an edge, full width | A slightly tighter crop | A strip of text | Social posts, blog images |
| Over the subject, mid-sized | Careful retouching of detail | Visible but the photo still reads | Portfolios, previews |
| Tiled across the image | Retouching the whole frame | Obvious, the picture looks unfinished | Client proofs, stock previews |
Opacity, Color and the Photo Underneath
Opacity is the setting that decides whether a mark looks considered or stamped on. Over the subject, something between 30 and 50 percent usually keeps the picture visible while leaving no doubt the mark is there. A small corner signature can run at 70 percent or more since it covers nothing that matters. Tiled patterns cover everything, so they are normally kept lighter, around 15 to 35 percent.
Color matters more than people expect, because one batch rarely has one background. White text at 40 percent disappears on a snowy scene and shouts on a night shot. A soft dark shadow around light text fixes most of this without adding a box behind it. When a set mixes bright and dark photos, preview the brightest and the darkest before saving, rather than the first one in the folder.
Keep the mark simple. A name or a short web address reads at small sizes, and a © in front of it is a widely recognised way to say the picture is not free to take. A long sentence turns into a grey smear once it is scaled down to fit.
Size the mark by proportion, not by pixels
A mark 300 pixels wide is a quarter of a 1200-pixel web image and barely a twentieth of a 6000-pixel camera file. If a batch mixes both, set the watermark as a share of each image instead, along with the margin from the edge. Then a phone screenshot and a full-resolution photo come out looking like they belong to the same set.
The Batch Problem: The Same Mark Everywhere
In 2017 Tali Dekel, Michael Rubinstein, Ce Liu and William Freeman, working at Google, presented a paper at CVPR titled On the Effectiveness of Visible Watermarks. Their point was that watermarks are usually applied in exactly the same way to many images: the same graphic, the same spot, the same opacity. That consistency is a weakness. Given a collection of images marked the same way, their method estimates the watermark and its transparency from the images themselves and then recovers the clean pictures underneath, with no knowledge of the mark in advance. Google's write-up says the estimate becomes very accurate from hundreds of images, and that much of the world's stock imagery was susceptible when they tested it.
The obvious counter-measures did not help much. Moving the mark to a random position on each image did not prevent removal, and neither did small random changes in its opacity. What did help was warping the watermark slightly and differently on each image. A removal tool then has to work out the distortion for every photo separately, and the authors report that even very subtle warping produced watermarks their technique could not fully defeat. They were careful to add that this is no guarantee against future methods.
The practical reading for anyone watermarking a set is simple. One identical stamp across a large public collection is something software can learn. It still deters casual copying, and it still credits you, but it is not a lock. Separately, tools that fill in parts of a single image have become far better at painting over a mark, which also favours marks that cross detailed parts of the picture over marks on plain sky or wall.
Watermark one photo or fifty
Text or logo, in a corner or tiled, sized to each image. Preview it live and download the batch as a ZIP.
What Happens to the File
A watermark tool that runs in the browser draws each photo onto a canvas, paints the mark on top, and writes a new file. Three things follow from that. JPG and WebP go through one more round of lossy compression, which at quality 90 or so is very hard to see but adds up if the same file is watermarked again and again, so always start from the original. Metadata does not survive, because a canvas holds only pixels: camera model, capture date and GPS position are gone, which is often welcome before a photo goes online. And an animated GIF comes out as a single still frame.
Keep the unmarked originals somewhere safe. A watermark burned into the pixels cannot be taken out cleanly by you either, and you will want a clean copy for print, for a client who has paid, or for a later edit with a different mark.
A Short Routine for Watermarking a Set
The same few checks catch most problems before the files go out.
Publish smaller copies
A 1600-pixel image is plenty for a screen and far less useful for print. Resize before watermarking, so the copy online is worth less than the original even without the mark.
Cross detail, not empty space
A mark over plain sky is easy to fill in. One that crosses edges, faces or texture takes far more work to remove convincingly.
Check the extremes of the batch
Preview the lightest and darkest photos, and the most vertical and most horizontal, before saving all of them.
Never watermark the master
Work on copies and keep the originals unmarked, so you can produce a clean file or a different mark later.
Frequently Asked Questions
Related Tools
Keep Reading
Why Resized Images Come Out Noisy, and the Fix Nobody Applies
Bilinear sampling reads four pixels no matter how far you shrink, so a 20:1 reduction ignores 396 of every 400. Halving in steps reads all of them.
What the JPEG Quality Slider Is Really Doing to Your Photo
Quality 80 scales the quantization tables to 40 percent, it does not keep 80 percent of the picture. Where the setting stops being free, and why PNG has no dial.
PDF Watermarks: Where They Live in the File and What They Stop
A watermark can be page content, an annotation or an Acrobat object. Where each lives, why marks behind the text vanish on scans, and what they stop.
What a Crop Really Costs: Pixels, JPEG Blocks and Hidden Leftovers
A 2x crop keeps a quarter of the pixels, a JPEG only crops losslessly on block lines, and some cropped files still held what was cut. What each one means for your images.