The system automatically scans the image, finds the specified watermark(using a template) with pixel-perfect accuracy, and generates a removal mask. The watermark area is reconstructed by a neural network so that the background structure, texture, and gradients remain completely natural. Immediately after cleaning, the image is sent to an AI upscaling unit, where a neural network algorithm removes compression noise and enhances detail.
The standard chaiNNer nodes have their limitations, so a custom set of Python nodes was designed and written for this task.
As a result, with just one click and a short wait, you get a highly detailed image, free of watermarks. There's also a version that can process all files in a specific folder.

The standard chaiNNer nodes have their limitations, so a custom set of Python nodes(more than 50% of the entire project) was designed and written for this task. A classical mathematical approach to the subtraction (removal) of translucent visible watermarks is used, which is based on the model of linear mixing of layers.
As a result, with just one click and a short wait, you get a video in its original resolution, free of watermarks.