Did you guys experimented with Kontext Dev as a simple text to image model (it can render images without image inputs)? and if yes, do you think it's worse or better than Flux Dev?