Welcome improved action Send Sequence, which is now combines ALL types of actions combined - it can send clicks, type in text, mouse and click mouse (including links and bindings support).
For about 1.5 years, there were 3 different types of Send actions - Send Input, Send Text and Send Sequence. They all covered their specific set of functionality, Send Input being more focused on clicking somewhere (image location, bindings, etc), Send Text having multiple ways of typing the text (char-by-char, clipboard, etc). Send Sequence was always focusing on “replaying” long sequences of inputs, but without precision of Send Input. This is now in the past as the new reworked version has the same capabilities as all types of Send and it is extremely easy to add new features to it, new “blocks" with any kind of functionality in it - like pick coordinate, do some math and use it to move the mouse.
The main benefit is that now it is possible to construct sequences of inputs way easier than before - you do not have to interchange Send Sequences with Send Inputs if you want to click on some image, for example. Also I will be focusing on improving Send Sequence in the nearest future, making it more flexible rather than spreading time on 3 different types of *Send.
** From now on Send Text and Send Input are on life support** - I will not add new options there and will not touch them at all. At some point in the future, they will be fully deprecated and automatically converted to Send Sequence.
You can do the conversion manually at any moment by clicking “Convert to Send Sequence” button displayed at the top of each and every Send action. For now, conversion does not actually remove the original action, giving you opportunity to review that the new action really does what you're expecting it to do.
As the title says, I've added new option to all these triggers which allows to capture any window without its border. Moreover, all actions which work by linking/binding position of found Image will take into account that there is no border and will adjust coordinates accordingly.
This means that tracking and object on apps which are not full screen is easier than ever.
Added new option, which allows to search for specific “blocks” of pixels of a specific color. There is a restricted set of blocks (like 2x2, 4x4, etc) and the code is optimized to better common use-cases in a more performant way. The optimal (from my testing) are 4x4 blocks.
Keep in mind, that under the hood this mechanism is very similar to how Image Search works, so in some scenarios it may be better to just stick to it. But in other scenarios this new mode in Color Search will shine. And oh, of course you can configure bindings/links to found blocks.
Added quite thorough documentation which should answer most of your questions regarding configuration. The tool is working and can be really handy even with its current functionality.
Initially the tool allowed to train models only locally, on your own PC using your own hardware. But this has its limitations - you have to have this hardware. Training in the cloud allows you to use enterprise-grade GPUs (like A100) to do the training faster and more efficiently in some cases. It has its limitations, but could be useful in many-many cases. For now, the support is limited (e.g. no auto-annotations for this mode yet), but I will work on extending it.