This is the base SDXL checkpoint, with the following changes:<ul><li>DPO Lora merged in at 0.25 model / clip strength</li><li>SPO Lora merged in at 0.75 model / clip strength</li><li>Baked VAE replaced with SharpSpectrumVAE XL</li><li>CLIP-L replaced with ZeroInt's <a target="_blank" rel="ugc" href="https://huggingface.co/zer0int/CLIP-GmP-ViT-L-14/blob/main/ViT-L-14-BEST-smooth-GmP-TE-only-HF-format.safetensors">ViT-L-14-BEST-smooth-GmP-TE-only-HF-format.safetensors</a></li></ul>This is basically an opinionated set of changes I found after a bunch of testing to provide an overall better basis to work off of.

Zoot's Tweaked Base SDXL For Finetuning Or Lora Training

Swaps out regular CLIP-L for ZeroInt's <a target="_blank" rel="ugc" href="https://huggingface.co/zer0int/CLIP-GmP-ViT-L-14/blob/main/ViT-L-14-BEST-smooth-GmP-TE-only-HF-format.safetensors">ViT-L-14-BEST-smooth-GmP-TE-only-HF-format.safetensors</a>All other changes are the same as V1.

Files with SHA256: a5f862cd749ca646282ca6e4d101bf8fc041e93d2ff37be8e28bd8b0da56d653