The idea behind this model merge is to create a perfect mix capable of generating artistic photography: artful portraits and beautiful landscapes. Both HotArt and HotArt2 as well as HotArt3 are great in their own way, try both with the same seed/sampler = then use the one according to your taste ;) This rich Checkpoint/LORA mix series is an experiment that worked out very well ;)For HotArt2 better hands - use both of the suggested resources (LORAs) at low ~0.333 level.See below for more examples of generation capabilities 👇

HotArt

This version adds <a target="_blank" rel="ugc" href="https://civitai.com/models/510261/spo-sdxl4k-p10eplorawebui">Step-aware Preference Optimization (SPO) LORA</a>, [a novel post-training approach that independently evaluates and adjusts the denoising performance at each step, using a step-aware preference model and a step-wise resampler to ensure accurate step-aware supervision... blah, blah, blah = makes the output way artfully beautiful, a.k.a. - NICER !!!]P.S.: The SPO approach is better than <a rel="ugc" href="https://civitai.com/models/242825/dpo-direct-preference-optimization-lora-for-xl-and-15-openrail?modelVersionId=273996">Direct Preference Optimization (DPO)</a> that has extended its success from aligning large language models (LLMs) to aligning text-to-image diffusion models with human preferences. Unlike most existing DPO methods, which assume that all diffusion steps share a consistent preference order with the final generated images, this assumption neglects step-specific denoising performance and that preference labels should be tailored to each step's contribution.

Files with SHA256: 96ffa5bb688edbff5abb71053c7cce4daf42f44c07b9ff0a516c7ea8d045eb66