This merge is based on <a target="_blank" rel="ugc" href="https://civarchive.com/models/1789765/bigasp-v25?modelVersionId=2025412">bigASP 🧪 v2.5</a> and requires <a target="_blank" rel="ugc" href="https://github.com/comfyanonymous/ComfyUI">ComfyUI</a> with the ModelSamplingSD3 node to function.First step was a tensor sum between these two models using <a target="_blank" rel="ugc" href="https://github.com/ljleb/comfy-mecha">comfy-mecha</a>.<a target="_blank" rel="ugc" href="https://civarchive.com/models/277058/epicrealism-xl">epiCRealism VXVII CrystalClear</a><a target="_blank" rel="ugc" href="https://civarchive.com/models/617652?modelVersionId=690475">NatViS 1.0</a>Second step was an add difference of the result, minus SDXL base, on <a target="_blank" rel="ugc" href="https://civarchive.com/models/1789765/bigasp-v25?modelVersionId=2025412">bigASP 🧪 v2.5</a> and a slerp with the latter again. The results are very close to <a target="_blank" rel="ugc" href="https://civarchive.com/models/1789765/bigasp-v25?modelVersionId=2025412">bigASP 🧪 v2.5</a> but more stable in my testing.A workflow is attached in each example image, some tweaks are set at specific settings that help the generation using <a target="_blank" rel="ugc" href="https://github.com/pamparamm/sd-perturbed-attention">PAG</a>, <a target="_blank" rel="ugc" href="https://github.com/Reithan/negative_rejection_steering">NRS</a> and <a target="_blank" rel="ugc" href="https://github.com/pamparamm/sd-perturbed-attention">PLADIS</a>.<a target="_blank" rel="ugc" href="https://gitlab.com/-/snippets/4879090?plain=1">Click here for an ngram list.</a> (tags it was trained on.)

Files with SHA256: 6b5d7b3e53f967c35c60c7cda890985ca3590a277de402e54d9b0372cd4419a5