<h1 id="heading-752">What is DPO?</h1><p>DPO is <strong>D</strong>irect <strong>P</strong>reference <strong>O</strong>ptimization, the name given to the process whereby a diffusion model is finetuned based on human-chosen images. Meihua Dang et. al. have trained Stable Diffusion 1.5 and Stable Diffusion XL using this method and the Pick-a-Pic v2 Dataset, which can be found at <a target="_blank" rel="ugc" href="https://huggingface.co/datasets/yuvalkirstain/pickapic_v2">https://huggingface.co/datasets/yuvalkirstain/pickapic_v2</a>, and wrote a paper about it at <a rel="ugc" href="https://huggingface.co/papers/2311.12908">https://huggingface.co/papers/2311.12908</a>.</p><h1 id="heading-753">What does it Do?</h1><p>The trained DPO models have been observed to produce higher quality images than their untuned counterparts, with a <strong>significant emphasis</strong> on the adherence of the model to your prompt. These LoRA can bring that prompt adherence to other fine-tuned Stable Diffusion models.</p><p></p><h1 id="heading-1572">Who Trained This?</h1><p>These LoRA are based on the works of Meihua Dang (<a target="_blank" rel="ugc" href="https://huggingface.co/mhdang">https://huggingface.co/mhdang</a>) at</p><p><a target="_blank" rel="ugc" href="https://huggingface.co/mhdang/dpo-sdxl-text2image-v1">https://huggingface.co/mhdang/dpo-sdxl-text2image-v1</a> and <a target="_blank" rel="ugc" href="https://huggingface.co/mhdang/dpo-sd1.5-text2image-v1">https://huggingface.co/mhdang/dpo-sd1.5-text2image-v1</a>, licensed under OpenRail++.</p><p></p><p></p><h1 id="heading-1573">How were these LoRA Made?</h1><p>They were created using Kohya SS by extracting them from other OpenRail++ licensed checkpoints on CivitAI and HuggingFace.</p><p>1.5: <a target="_blank" rel="ugc" href="https://civarchive.com/models/240850/sd15-direct-preference-optimization-dpo">https://civarchive.com/models/240850/sd15-direct-preference-optimization-dpo</a> extracted from <a target="_blank" rel="ugc" href="https://huggingface.co/fp16-guy/Stable-Diffusion-v1-5_fp16_cleaned/blob/main/sd_1.5.safetensors">https://huggingface.co/fp16-guy/Stable-Diffusion-v1-5_fp16_cleaned/blob/main/sd_1.5.safetensors</a>.</p><p>XL: <a target="_blank" rel="ugc" href="https://civarchive.com/models/238319/sd-xl-dpo-finetune-direct-preference-optimization">https://civarchive.com/models/238319/sd-xl-dpo-finetune-direct-preference-optimization</a> extracted from <a target="_blank" rel="ugc" href="https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/blob/main/sd_xl_base_1.0_0.9vae.safetensors">https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/blob/main/sd_xl_base_1.0_0.9vae.safetensors</a></p><p>These are also hosted on HuggingFace at <a target="_blank" rel="ugc" href="https://huggingface.co/benjamin-paine/sd-dpo-offsets/">https://huggingface.co/benjamin-paine/sd-dpo-offsets/</a></p>

Files with SHA256: d4ec63a436dbd20e5cac4a3e9a1fd0c59fb2513018266252c4ee9ebeb5ce6762