HLS feels like the best format to actually embed effective watermarks in.
With a traditional "large-file" video format, this is non-trivial, as you need to reencode (or at least remux) the entire video for each user. This is not only computationally intensive, it also prevents you from hosting your videos on a CDN, increasing bandwidth costs.
With a segmented format like HLS, you can just embed the watermark in a few key segments, leaving the rest of the video intact. As long as the user can't distinguish between watermarked and non-watermarked segments, you get most of the benefit at a much lower cost. Crucially, you can keep serving the non-watermarked segments off a CDN.
This makes me wonder whether there are "segment-native" watermarking algorithms. The idea would be to pre-encode each segment multiple times, with each file containing a different watermark. A random variant of each segment would be chosen for each download, with the IDs of the chosen variants saved in a database. THe security of such an algorithm would rely on the combinatorial explosion resulting from the number of allowed variant combinations. For example, with just 3 variants per segment, 15 segments would give you 59 million distinct watermarks.
Lanedo · 4h ago
It is not too surprising that many HTML or JS based watermarking schemes are very easy to remove, some even via a script. For watermarking schemes to be effective, the watermark payload needs to be embedded in the content. Here's one approach that works for images:
https://github.com/tim-janik/imagewmark/
Incidentally, the latter also supports applying the watermark payload to HLS segments.
josephcsible · 3h ago
If you're obfuscating code or writing anti-debugging logic, you are the bad guy. Stop trying to control other people's computers.
mlhpdx · 4h ago
On reading the title, my mind went to crafting image (or video) uploads that effectively “undo“ the water marking. That is, is there a way to alter content before uploading it so that when the watermark is applied it reverts the change and so what is displayed lacks the watermark. That would be crafty.
kevindamm · 3h ago
You could, but then it becomes even easier to capture without the adornment, practically defeating the purpose.
You may be interested in watermarks that are invisible to the human eye but are visible to spectral analysis. You can even encode data within this invisible watermark, and learn a modification that is resistant to various image manipulations and even resistant to lossy compressions, in research that goes back to 2001 and has seen interesting improvements with advances in DNNs, Transformers and Diffusion.
nullbyte · 6h ago
It's a good article, but I was a little disappointed that he only went over crude watermarking technologies. The more advanced systems implant an invisible watermark in the image or video itself by using neural networks. Those are a lot more tricky to bypass than simply removing a floating div.
Mindwipe · 2h ago
To be honest if you are doing client side watermarking outside of a secure hardware enclave I'm not sure you expect any genuine robustness, because you obviously don't have any.
laserspeed · 6h ago
Bypassing of different Watermark implementations; including tricks related to Picture-In-Picture, erroneous assumptions at the time of enforcing client-side protections, and then HLS (HTTP Live Streaming) and ways to reassemble videos offline by looking into m3u8 playlists and encrypted video segments.
With a traditional "large-file" video format, this is non-trivial, as you need to reencode (or at least remux) the entire video for each user. This is not only computationally intensive, it also prevents you from hosting your videos on a CDN, increasing bandwidth costs.
With a segmented format like HLS, you can just embed the watermark in a few key segments, leaving the rest of the video intact. As long as the user can't distinguish between watermarked and non-watermarked segments, you get most of the benefit at a much lower cost. Crucially, you can keep serving the non-watermarked segments off a CDN.
This makes me wonder whether there are "segment-native" watermarking algorithms. The idea would be to pre-encode each segment multiple times, with each file containing a different watermark. A random variant of each segment would be chosen for each download, with the IDs of the chosen variants saved in a database. THe security of such an algorithm would rely on the combinatorial explosion resulting from the number of allowed variant combinations. For example, with just 3 variants per segment, 15 segments would give you 59 million distinct watermarks.
Another that works for audio content is: https://github.com/swesterfeld/audiowmark
Incidentally, the latter also supports applying the watermark payload to HLS segments.
You may be interested in watermarks that are invisible to the human eye but are visible to spectral analysis. You can even encode data within this invisible watermark, and learn a modification that is resistant to various image manipulations and even resistant to lossy compressions, in research that goes back to 2001 and has seen interesting improvements with advances in DNNs, Transformers and Diffusion.