It's probably actually good, but after this stuff with Qwen-Image-3 and it's ability to things like render images that look like they were rendered with LaTeX or compose many things side by side or in depth, maybe it's perhaps not totally in the right direction to only focus on visual stuff.