I'm using Flux now. It uses a language model to process your prompt before handing it over to the actual image generation model.