Understanding Neural Filters
If you worked through the earlier lessons in this book, you’ve used Photoshop filters such as Surface Blur, Smart Sharpen, Clouds, and Liquify. Those are conventional filters that produce their results with algorithms—procedural programs where the code itself determines the result.
The newer Neural Filters produce results in a different way, combining traditional algorithms with machine learning and other advanced techniques. Machine learning means that a filter in Neural Filters can be trained using many examples of desirable and undesirable results, potentially creating better images than what could be achieved with procedural computer code alone.
Neural Filters are different from other Photoshop effects in the following ways:
Neural Filters are trained by machine learning and neural networks.
Some filters need to be downloaded before the first time you use them. This is partly to save space on your computer, because some Neural Filters and their machine-learning models can be large. If you need to download a Neural Filter, you can do it with one click in the Neural Filters workspace.
Some filters display a message saying that they process image data in the cloud (on Adobe Creative Cloud servers). They may need more power than a desktop computer has, or a machine-learning model may be too large to download.
It’s possible to use some Neural Filters without an Internet connection, but you get the most options when your computer is connected to the Internet.