AuraFlow is a new open-source text-to-image model that has the potential to become the new backbone of image creation workflows such as with ComfyUI. Backed by Fal, it has released a base version 0.1 with the promise of more to come. It’s a 6B parameter model, which is quite large, and some are complaining that it’s hard to run on their own PC due to RAM requirements. However, it’s available through API from Fal and Replicate (who have added additional sizing options).
Currently, it looks promising, especially with illustration and art styles, and it can add text to images. It is trained with positional information and spatial relationships, and the prompts can be reasonably detailed. However, it’s still not up there with the top proprietary models (Midjourney v6 being the current favourite), or even the top open-source models, as ranked on imgsys (being topped by RealVisXL V4.0)