John6666

/

rihikari-2d-anime-illustration-v10-sdxl

like 0

For below code, this model can create NSFW image (I must warn you that you’ll need to test it while you’re alone).

from diffusers import DiffusionPipeline
import torch

# Set Seed
seed = 50
# Set random seed for reproducibility
torch.manual_seed(seed)

# Initialize the pipeline
pipe = DiffusionPipeline.from_pretrained("John6666/rihikari-2d-anime-illustration-v10-sdxl")
pipe.to("cuda")
pipe.enable_model_cpu_offload()

# Define parameters
prompt = "rihikari is looking to me"
negative_prompt = "low quality, blurry, distorted, extra limbs, bad anatomy"  # Avoid undesired features
guidance_scale = 7.5  # Controls prompt adherence
width = 1024  # Image width
height = 1024  # Image height
num_inference_steps = 25  # Number of denoising steps
num_images_per_prompt = 1  # Number of images to generate
eta = 0.0  # Controls randomness in some schedulers (e.g., DDIM)
generator = torch.Generator(device="cuda").manual_seed(seed)  # CUDA-specific generator for consistency

# Generate the image
image = pipe(
    prompt=prompt,
    negative_prompt=negative_prompt,
    guidance_scale=guidance_scale,
    width=width,
    height=height,
    num_inference_steps=num_inference_steps,
    num_images_per_prompt=num_images_per_prompt,
    eta=eta,
    generator=generator,
    output_type="pil"  # Output as PIL image
).images[0]

# Display or save the image
image

Please check again and do something that needed.