Every gui abstraction trades away control of your drawing buffer. Which means every abstraction sucks in some way. The best library is one that gives you a 'put_pixel(x, y, color)' function.
From there you can have additional tools. Lines, circles, boxes, anti-aliasing, font rendering, window resizing, etc but it should always come back to "here's your surface, do what you want"
Unfortunately the graphics pipeline makes this a pain. Final control belongs to pixel shaders on the GPU, and for most applications, it isn't fun or interesting to write low level Vulkan boilerplate.
It makes me want to design my own CPU that cares less about performance and more about simplicity. Things might actually be more performant if we didn't constantly have 5 layers of abstraction protecting us from implementation details.