sdxl-DPO-LoRA

DPO LoRA Stable Diffusion XL

Model trained with LoRA implementation of Diffusion DPO Read more here

Base Model: https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0

Running with 🧨 diffusers library

import torch
from diffusers import AutoPipelineForText2Image, DPMSolverMultistepScheduler
from diffusers.utils import make_image_grid

pipe = AutoPipelineForText2Image.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0",
    torch_dtype=torch.float16,
    use_safetensors=True,
    variant="fp16",
)
pipe.scheduler = DPMSolverMultistepScheduler.from_config(
    pipe.scheduler.config,
    use_karras_sigmas=True,
    algorithm_type="sde-dpmsolver++"
)

pipe.to("cuda");

seed = 12341234123 
prompt = "professional portrait photo of a girl, photograph, highly detailed face, depth of field, moody light, golden hour, style by Dan Winters, Russell James, Steve McCurry, centered, extremely detailed, Nikon D850, award winning photography"
negative_prompt = "3d render, cartoon, drawing, art, low light, blur, pixelated, low resolution, black and white"
num_inference_steps = 40
height = 1024
width = height
guidance_scale = 7.5

pipe.unload_lora_weights()
pipe.load_lora_weights(
    "radames/sdxl-DPO-LoRA",
    adapter_name="sdxl-dpo-lora",
)
pipe.set_adapters(["sdxl-dpo-lora"], adapter_weights=[0.9])
generator = torch.Generator().manual_seed(seed)
with_dpo = pipe(
        prompt=prompt,
        guidance_scale=guidance_scale,
        negative_prompt=negative_prompt,
        num_inference_steps=num_inference_steps,
        width=width,
        height=height,
        generator=generator,
    ).images[0]
with_dpo

Adaptor Weights effect

adapter_weights

ComfyUI

https://huggingface.co/radames/sdxl-DPO-LoRA/raw/main/workflow-sdxl-dpo-lora.json