Short answer: How to watermark photos: add your name or logo to a copy of the image, never the master, at 30โ50% opacity, sized to 15% of its width and placed where a crop cannot remove it. Visible marks deter theft; invisible and forensic marks track copies. A watermark is evidence, not access control. Pair it with reverse image search.
Every photo you publish can be copied in two clicks, and a watermark is the one defence that travels with the image itself. It survives downloads, reposts and screenshots in a way no site setting ever will. But not all watermarks are equal: a small logo in the corner does a different job from a hidden forensic signal, and a mark placed badly protects nothing at all. This guide explains how to watermark photos properly, the three types, where to put them, how strong to make them, the free tools and batch workflows that actually work, and a real test in which we watermarked a photo and then reverse-searched it.
What is a watermark, and what can it actually do?
A watermark is a name, logo or hidden signal added to a photo so that copies of the image carry proof of who it belongs to. It does three jobs, and they rank in this order. First, attribution: when a photo spreads, the mark carries your name with it, so viewers who see a shared image can find its owner. Second, deterrence: a visible mark raises the cost of casual theft, because a scraper or lazy blogger wants a clean image and a watermarked one is less attractive than the next result. Third, branding: a consistent mark turns every shared image into free advertising.
The honest limits matter more than the sales pitch. A watermark is evidence and deterrence, not access control. It does not stop anyone determined from using your photo. A corner mark can be cropped out in seconds on a phone. Modern AI removal tools can erase a semi-transparent mark from a smooth area with a few clicks, and Google Research demonstrated in 2017 that a watermark placed in the same position across many images can be removed automatically. As the WebCopyrightChecker guide puts it plainly, no watermark is 100% removal-proof. So treat a watermark as one layer of a system, not the whole system: keep your masters private, write a copyright line into the photo’s metadata, publish only watermarked or downsized copies, and monitor for stolen copies with reverse image search. Our guide to protecting your photos from being stolen covers the rest of the layers.
One mistake to avoid before you start: never watermark your only clean copy. Keep the master file private, a local archive or private cloud storage, never web-accessible, and publish only watermarked or downsized derivatives. A typical safe public size is 1,500โ2,048 pixels on the long edge at quality 80โ85. If the clean full-resolution version is already public, a watermark on a copy protects nothing, because thieves will simply take the clean one.
Visible, invisible or forensic: which type do you need?
There are three kinds of watermark, and choosing the wrong one is the most common mistake in this whole subject. A visible mark is a name or logo anyone can see. An invisible mark hides data inside the pixels themselves, so the photo looks untouched. A forensic mark is an invisible mark built for legal evidence: it is designed to survive compression, cropping and format changes, and to identify the exact copy that leaked. This table is the decision to keep, it is the ranking asset for this guide, built from the methods and prices verified for this article.
| Type | Who it is for | Cost | Survives cropping? | Survives compression and AI removal? |
|---|---|---|---|---|
| Visible (text or logo) | Photographers, sellers, social creators | Free (built into Lightroom, GIMP, free online tools) | Only if tiled or centred; a corner mark is cropped out in seconds | No, a determined thief can clone or inpaint it out |
| Invisible (metadata and basic embedded signals) | Anyone who wants attribution without spoiling the image | Free (EXIF copyright field) to low-cost tools | Yes, if embedded in the pixels; no, if it is only metadata, which is often stripped | Partly, survives light compression, not heavy re-editing |
| Forensic (Digimarc and similar) | Stock agencies, publishers, high-value commercial work | Enterprise, WebCopyrightChecker’s guide lists Digimarc Guardian plans as starting at $5,000 a year | Yes, within limits, the signal is spread across the image | Yes, by design, Digimarc’s own site describes hidden signals that survive compression, resizing, print-and-scan and screen capture |
Most readers need the first row, and some need the first two together: a visible mark on previews and an invisible one on delivered files. That combination is called dual watermarking, it deters casual theft while leaving you a way to track a copy if the visible mark is removed. Forensic watermarking is a different purchase, not a bigger version of the same thing. Digimarc, the best-known forensic provider, embeds hidden signals that its own site describes as tracing leaks, detecting unauthorised copies and supporting provenance through the C2PA standard. You can read what it offers on Digimarc’s official site, but unless you are licensing images at volume, its enterprise pricing puts it out of reach, and a visible mark plus good records will do the job.
Do not confuse an invisible watermark with the copyright field in a photo’s metadata. Writing “ยฉ Your Name 2026” into the EXIF copyright field is free, takes seconds, and is worth doing, but social platforms and scrapers routinely strip metadata, so it is not a forensic mark. It is a label on the file, not a signal in the pixels. The opposite job, tracing a watermarked copy back to its clean original when you have the right to use it, is covered in our guide to finding the original image without a watermark.
Where should a watermark go, and how strong should it be?
Placement decides whether a crop kills your mark, and strength decides whether anyone notices it. The rules below come from ImageGuide’s watermarking guide, and they hold across every tool:
- Opacity between 30% and 60%. Below 30%, the mark vanishes on bright images. Above 60%, it fights the photo for attention. For most work, 40โ50% is the sweet spot.
- Keep a margin of 3โ5% of the image width from the edge. Marks glued to the exact edge look like compression artefacts and get cropped with zero effort.
- Scale the mark relative to the image width, not in fixed pixels. A 24-pixel mark disappears on a 6,000-pixel original and looks huge on a 400-pixel thumbnail. A logo around 15% of the photo’s width stays consistent.
- Avoid faces and the main subject. Place the mark over sky, bokeh or empty space when you can, so it does not ruin the photo it is protecting.
- Stay consistent. Same position, same opacity, same logo everywhere. Recognition compounds, and a mark that jumps around looks careless.
The placement table is short because the trade-off is simple. A bottom-right corner mark almost never hides the subject, but it is not crop-proof at all, it is branding, not protection. A centred mark is crop-proof but blocks the photo, so it belongs on low-resolution proofs and stock previews, not on finished work. A tiled pattern repeated diagonally across the frame is crop-proof and only mildly intrusive at low opacity, which makes it the right choice for preview galleries and paid photo delivery. If you sell photos, tile the previews and deliver the clean file only after payment. If you post for attention, use the corner and accept that it is a signature, not a lock.
One more placement warning, from the same Google Research finding mentioned above: if you put an identical mark in the identical spot on hundreds of photos, automated tools can learn the pattern and remove it from all of them at once. Vary the position slightly between images, or use a tiled pattern, if you publish at volume.
How to watermark photos, step by step
The tool matters less than the routine: design the mark once, save it as a preset, and apply it at export so every published copy carries it. Here are the three routes, from the most professional to the freest.
Route 1: Lightroom Classic (photographers)
Adobe’s own Watermark Editor is the cleanest way to do this if you already use Lightroom. Adobe’s help page, last updated on 7 February 2022, gives the steps: in any module, choose Edit > Edit Watermarks on Windows, or Lightroom Classic > Edit Watermarks on a Mac. Pick Text for your name or Graphic for a PNG logo with a transparent background. Set the opacity, size, inset and anchor point, then click Save to store it as a preset. After that, select your photos, choose File > Export, tick Watermark and choose your saved mark. You can read the full instructions on Adobe’s Watermark Editor help page. Lightroom on mobile has the same idea in a simpler form: open a photo, tap Share, choose Export As, and switch on Include Watermark. Save each version, a small corner mark for the web, a large low-opacity mark for client proofs, as its own preset, so you never rebuild it.
Route 2: Free online tools (occasional use)
If you only watermark a few photos, a browser tool is enough, and yes, you can watermark photos online for free. Watermarkly has a free basic tier, PicMarkr is free, and uMark Online is free for batches of more than 40 images, according to WebCopyrightChecker’s guide to watermarking tools. The workflow is the same everywhere:
- Open the tool and upload a copy of your photo, never your only master.
- Type your name or upload your logo as a PNG with a transparent background.
- Set the opacity to 40โ50% and the size to about 15% of the image width.
- Place the mark in a corner for branding, or tile it across the frame for a proof.
- Download the watermarked copy and check it at the size people will actually see it.
Two cautions. First, you are uploading your photo to someone else’s server, so do not use a free online tool for unpublished client work or anything sensitive. Second, free tools rarely expose fine opacity control or relative sizing, so keep your expectations modest, they are fine for a social post, not for a 500-photo client gallery.
Route 3: Batch watermarking with free desktop software (volume)
For a whole folder of photos, a batch tool beats any one-at-a-time method. GIMP is free and does the job: open the photo, add your logo as a new layer with File > Open as Layers, set the layer opacity to 40โ50%, drag it into place, and record the steps as a Script-Fu script for repeats. For a fully automatic pipeline, the Pillow library in Python does the same work in about ten lines, following the pattern in ImageGuide’s guide: scale the logo to 15% of the photo’s width, cut its opacity to 50%, place it in the bottom-right corner with a margin of 4% of the photo’s width, and loop over the folder. ImageMagick does the same from the command line with one composite command per photo. Whichever you choose, run the batch on copies, check the first three outputs by eye before letting it loose on the folder, and keep the clean masters somewhere the batch can never touch.
Worked example: we watermarked a real photo and reverse-searched it
Guides usually describe watermarking without ever testing it, so we ran a small honest test on 9 November 2026. The photo is a real published image from Wikimedia Commons: “Vintage Yashica 35mm SLR Film Camera, Model FX-2, Made In Japan”, 1,920 by 1,552 pixels. It is the kind of photo a seller might list, a product, shot cleanly, on a plain background.
We watermarked it ourselves with Pillow, the free Python imaging library, following the rules in this guide. The mark is the text “ยฉ Image Search Techniques” in bold, 42 pixels high, which is 34% of the photo’s width, in white at 50% opacity with a faint dark shadow so it stays readable on light areas. It sits in the bottom-right corner with a margin of 76 pixels, 4% of the photo’s width, from the edges. We also made two more versions: a tiled version with the same text repeated diagonally across the whole frame at 30% opacity, and a cropped version with 18% cut off the right and bottom edges.
Then we uploaded the watermarked copy to Google Lens, without signing in to any account, to see whether the mark interfered with finding the original. It did not. Lens identified the camera immediately, its AI Overview said the image shows “a vintage Yashica FX-2 35mm SLR film camera equipped with a Yashica DSB 50mm f/1.9 manual focus lens”, and the “Exact matches” tab found the original source: three exact matches, including the Wikimedia Commons file page for this exact photo and a Flickr copy dated 22 March 2015. The only related search Lens offered was “Yashica FX-2”, and the visual matches were led by eBay listings for the same camera model. The watermark text itself was never mentioned anywhere in the results.
That result cuts both ways, and both halves matter. A visible watermark does not break reverse image search: Lens reads the photo, not the mark, so a watermarked copy still traces back to its source, which is exactly what you want when you are hunting stolen copies of your own work. But it also means a watermark does not hide a photo from reverse search either. If you watermark a photo and wonder whether anyone can still find where it came from, the answer is yes. The mark is visible to people, and effectively invisible to the matching algorithm.
The cropped version confirmed the oldest weakness in watermarking without needing any tool at all: cutting 18% off the right and bottom edges removed the corner mark completely, and the photo still looked fine. If a competitor does exactly that to your photos, our guide to what to do when a competitor crops out your watermark covers the after-it-happened route, from the evidence pack to the escalation ladder. The tiled version survived the same crop, because the pattern repeats, cutting one corner leaves marks everywhere else. That is the whole placement lesson in two files: a corner mark is a signature, and a tiled mark is protection.
Which watermark workflow fits you?
Photographers, marketplace sellers and social creators need different marks, because their photos face different threats.
Photographers: build two presets in Lightroom, a small corner mark for the web and a large, low-opacity tiled mark for client proofs. Watermark the previews a client picks from, and leave the photos they paid for clean. Keep the masters private, write the copyright line into the metadata, and never publish the full-resolution original. If a client image is high-value, consider an invisible or forensic mark on the delivered file as well.
Marketplace sellers: your threat is another seller copying your listing photos, so keep the mark small, in a corner, and never over the detail a buyer needs to see, the lens etching, the label, the stitching. Batch-mark a whole listing’s photos at once with a free tool, and use the same mark on every listing so buyers learn to recognise it. Then reverse-search your own listing photos every few weeks: if your photo appears on another seller’s listing, you have found the thief, and our guide to protecting your photos online explains the full protection routine, including what to do next.
Social creators: consistency is everything. Use the same mark, position and opacity on every post, so your audience recognises your work at a glance and reposts carry your name. Remember that platforms recompress everything you upload, which can weaken invisible marks and soften visible ones, check how your mark looks after uploading, not just in your editor. And keep the originals: if a repost goes viral without credit, the original file is your proof.
What do you do when someone steals your watermarked photo?
A watermark only helps if you act on it. First, find the copies: reverse-search the photo through Google Lens and TinEye, using the steps in our how to reverse image search guide, and save screenshots of every infringing use with the date visible. The watermark itself is your evidence in that screenshot, it shows the thief had clear notice of who owned the image. Second, ask for removal: most platforms have a copyright report form, and a DMCA takedown notice is the standard tool for a host that ignores you. Our DMCA takedown guide for stolen photos walks through exactly what to send and where. Third, if the same image keeps reappearing, consider an invisible or forensic mark on future work so you can prove ownership even after the visible mark is cropped out. A watermark does not stop theft by itself, but a watermarked photo, a saved screenshot and a takedown notice together are a system that does.
FAQs
Do watermarks stop photo theft?
No, not on their own. A watermark deters casual copying and proves ownership when a photo is misused, but a corner mark can be cropped out in seconds and AI tools can remove semi-transparent marks. Treat it as evidence and deterrence, and pair it with private masters, reverse image search monitoring and a DMCA plan for removals.
Can I watermark photos online for free?
Yes. Watermarkly has a free tier, PicMarkr is free, and uMark Online is free for batches of more than 40 images. Upload a copy rather than your master, set the opacity to 40โ50%, and check the result at the size people will see. Free tools are fine for occasional social posts, but they rarely offer fine controls for large batches.
What is an invisible watermark?
An invisible watermark hides identifying data inside the photo’s pixels, so the image looks untouched while carrying proof of ownership. Basic versions survive light compression but not heavy re-editing, and they need special software to read. Forensic versions, such as Digimarc’s, are built to survive compression, cropping and format changes, but they are sold as enterprise services with enterprise prices.
Where is the best place to put a watermark?
Where a crop cannot remove it. A bottom-right corner mark suits branding but offers no real protection, while a centred or tiled mark survives cropping and suits proofs and previews. Keep opacity between 30% and 60%, size the mark to about 15% of the photo’s width, and never place it over a face or the main subject.
A watermark travels with a copy, but it does not stop another site from loading the image from your server. For that, see our guide to how to stop hotlinking images, which shows how to find the embeds in your access logs and block them.







