I present to you this model based on an old creation of mine that tries to combine the versatility of pony models with the realism of sdxl models, focused on NSFW content, has a good adherence to prompts and very good knowledge of anatomy, accepts a good range of samplers, the most used by me being dpm ++ sde karras and euler a simples, good images are created with 30 steps and CFG from 3 to 5, accepts the use of lora DMD2 well generating good images with only 8 steps and CFG of 1.5. It has great performance with IPadapter and instantID that I used in many creations to maintain style and facial consistency.
Description
I present the fast generation version of the caligula reborn with the embedded lora DMD2 allowing the generation of images in at least 4 steps, I recommend using 8 steps with CFG of 1.5 and the dpmpp sde, dpmpp sde 2m or 3m samplers with the sgm uniform scheduler. The images seem to have an analog style moving away from the pony style of the base model which makes the images in some cases more realistic.
FAQ
Comments (2)
it has issues making realistic looking people. people it makes look like what most pony models do, instead of what SDXL can do
so far this does a good job, but its not holding instructions. Like the images are clear but if I ask it for - have a person in the room surrounded by other people, it puts in the person (with a single expression the faces all seem to have the same look), but the surrounding cast isn't in the room, hall etc. Its just them all alone.