By William del Strother, Head of Innovation, FMTS
AI upscaling - converting low-quality images and video into high-res versions – is a trend that is increasingly impacting piracy, and consequently, watermarking and content security.
AI upscaling tools can distinguish between the picture and any undesirable artefacts resulting from compression, replacing them with more realistic textures. Operators are starting to use this technique to cut video encoding costs by transforming video to a lower resolution, which requires less bandwidth for transmission. AI upscaling at the client level then converts it back to the original quality for the viewer.
This presents a new content security problem for some, as it impacts the effectiveness of any watermarks that are rendered on the server.
However, because FMTS watermarks can be rendered on the client, as well as the server, we give content owners the reassurance that the watermark is preserved with full fidelity after AI upscaling.
Pirates riding the AI upscaling wave
Pirates see compression and AI upscaling as a potential way to remove watermarks. When there is a visual watermark, pirates can simply remove it and repaint the area. This can potentially be done using software such as Google’s latest AI tool, Gemini.
However, when trying to destroy an imperceptible watermark, the pirate is forced to degrade the video quality by adding distracting components into the video, such as white noise, or by downscaling it dramatically. Even after AI upscaling, details will have been unintentionally changed in the process, and video quality will be impaired.
Beating potato quality
Our watermarking solutions were designed with the fundamental philosophy that they must support any level of compression, noise, and low resolution. In fact, independent testing by Cartesian Consulting’s security analysts has demonstrated successful extraction of our watermarks from a 240p video recorded on a handheld camcorder, then compressed down to 125 kbps. This is well within the realm of ‘potato quality’ video (a YouTube term describing content of such terrible quality that it looks as if it was filmed using a potato). Even with the best AI upscaling, this level of compression will irreparably harm the image, and the AI upscaled version will not provide the same viewer experience as the original.
AI video makeover
While we are familiar with the role of AI to create deepfakes, another growing type of manipulation is video restyling, where AI converts video to new styles, including anime, claymation, or Pixar-style visuals. This technology is too computationally expensive to apply at scale to feature-length movies or live sports, but it is already raising questions about how to determine when restyling becomes copyright infringement.
Fortunately, all members of our Innovation team here at FMTS are keeping a watchful eye on this and other developments to ensure our solutions are being updated constantly to prepare our customers for a world with this new set of threats.
For more information about how our watermarking solutions can help you protect your content, get in touch with the team today.
Read our previous blogs on watermarking:
