SD 3.5
Stable Diffusion 3.5 from Stability has developed a Rectified Flow image generation model, improving the noise sampling method and enhancing prompt understanding, as well as separating the weight of text arrangement on images.
Initial tests of SD 3.5 Large:
Prompts, such as specifying two people wearing differently colored outfits, were successfully generated. It can also render text on images, though some letters may be missing or words may not display entirely. It can specify font emotions, and hands and fingers are working fine. From the comparison table of multiple models (Elo Score) on the Stability AI website, it shows that SD 3.5 is competitive with Flux1.Dev (SD 3.5 performs better on prompts, while Flux Dev scores higher on aesthetics).
You’ll have to try it out yourself to see what’s what at https://varoriya.com/service/sd35/
The links and prompts used for testing are in the comments 🥰