After some testing and training, i thought the best solution was to train a generell model that has a limited set of images, but covers a good range of styles and effects. I'll take it as a base for mixy-matchy with other, more focused Loras. Still stuff left i haven't tried, but will come later.Also trained on images i would not normally take for training, meaning to clean, to AI'ish, to much close-up etc. I would say it leans mostly towards realism, but that also depends on the prompt. With my highly convoluted LLM-Instruction setup it's usually no problem to get something, surprising, especially with rough sketches or mashed together pictures created in Photoshop.Funky little Base Model, that actually, for the first time in history, lets you train images with film grain, noise effects, imperfections and does not miraculously make them go "poof" once you try to create an image. Maybe Flux2 Klein is close, but not to this extend. This Lora has not that much of it, another one i tested had plenty of grain, noise and it worked. But .... that stupid name... just can't get over the name of the base model.

Files with SHA256: db01681e6d754ac87827af582d2b8b7b3f39ddbd0cbdb72b025fb2ad416f425d