How can CTM be turned into GenAI for text and image?

[BradKML]

Since LLM (both diffusion and conventional autoregressors), as well as image generators (both rectified flow and conventional diffusion models), are what people are used to as output, as the most common usecase of these models are for creative purposes, how can CTM be trained to handle text and image generation? Like can it handle pre-training + SFT + RL for text? Can it handle image model training when the internals are very different from what people usually expect? Or to be more general, can it handle high-fidelity outputs?