LoRA for Stable Diffusion 3.5L/Turbo trained on the works of Léon Bakst (1866-1924), the iconic painter & illustrator best known as the scene/costume/character-designer for Diaghilev's Ballets Russes
Trained on Replicate using Ostris' ai-toolkit, via the express interface found here:
This adapter is usable at a relatively broad range of LoRA scaling (0.5 to roughly 1.2), with varying effects and impacts on aesthetic character/style.
Generally speaking, this S.D.3.5 version seems to more closely & consistently extrapolate Bakst's aesthetics than our Bakst Flux LoRA, trained on the same data-set (albeit a smaller subset thereof).
Do note that, like most SD3.5 LoRAS thus far, this version seems to further disbalance S.D.3.5's ever-fragile knowledge of hands/fingers, and sometimes other anatomy (dep. on scale + settings/prompting). In this respect, Bakst Flux LoRA is substantially stabler, whilst still delivering some measure of Bakst's touch.Nonetheless, the outputs' tangibly painted quality, this S.D.3.5 variants's key forte, may well render suchlike anatomical imprecision irrelevant, forgivable, or simply undiscernible, depening on how you prompt and run it.
Learn more about Léon Bakst and check out his work here or here or, say, here.
Trigger words
You should use Bakst or 'Bakst style illustration' or 'by Léon Bakst' or 'Bakst style character art, by Léon Bakst" to trigger the image generation.
