Semantic selections

[rpstrat]

I would like to be able to make a really good high-quality selection of an object by identifying it by name. I can do this with various machine learning nodes for awhile now in ComfyUI. Even with the Quick Selection tool it can be time-consuming to make good selections. The Select Subject and Select Sky modes work pretty well (the former better than the latter). It would be great to say "Select the dog in the bottom left" and then Remove it using the Remove button

[imran_3941]

That’s a great suggestion — incorporating name-based object selection would significantly improve workflow efficiency. While ComfyUI already offers solid tools like Select Subject and Select Sky, a feature that allows users to make context-aware selections using natural language (e.g., “Select the dog in the bottom left”) would bridge the gap between AI-driven precision and user convenience.

Integrating this with the Remove function could make complex edits much faster, especially for scenes with multiple objects. Hopefully, future updates or community plugins will explore this direction, combining visual recognition with language input for more intuitive editing.