v10: Changed dataset and training config again. Much more real-looking and more consistent ******* realism. Amazing skin details and further improved chins. Great text writing ability.

Works best with mundane, real-life scenes. 'High-concept images' (e.g. fantasy and sci-fi scenes) can work, too, but will likely have less of an ******* and real look to them and more of a standard FLUX professional photography look.

Recommended inference settings (v10):

Rule-of-thumb:

lower guidance, lower amount of steps = more realism
higher guidance, higher amount of steps = better coherence

This is however not universally true, as depending on seed and prompt this can vary.

1.0 LoRa strength (only lower as a last resort or when needed, e.g. when combining with other LoRa's)
min. 2.5 (more realism) - max. 3.5 guidance (more coherence)
20/25/40/50 steps (these seem to be the optimal convergence points)
euler + ddim_uniform (more realism) OR simple (more coherence)

LoRa description:

This LoRa model was designed to generate images that look more "real" and less "artificially AI-generated by FLUX". It achieves this by making the natural and artificial lighting look more real and making the bokeh less strong. It also adds details to every part of the image, including skin, eyes, hair, and foliage. It also aims to reduce somewhat the common "FLUX chin and skin" and other such issues.