by Heartsync
Open source · 10k downloads · 201 likes
This AI model, called NSFW Uncensored, is designed to generate images without any censorship restrictions, offering an in-depth exploration of AI’s technical capabilities in visual creation. It allows users to test the limits of moderation filters by producing varied content based on provided prompts while assessing the model’s actual image-generation performance. Primarily used for research or development purposes, it caters to users seeking to understand the underlying mechanisms of AI models without editorial constraints. What sets it apart is its minimalist approach to censorship, emphasizing creative freedom and technical experimentation.
This model is a playground that minimizes censorship restrictions, allowing exploration of the technical possibilities of AI-based image generation. Through various prompts, you can test censorship boundaries and verify the actual performance of image generation AI.
# Basic usage example
from diffusers import DiffusionPipeline
import torch
# Load the model (with float16 precision for GPU)
pipe = DiffusionPipeline.from_pretrained(
"Heartsync/NSFW-Uncensored",
torch_dtype=torch.float16
)
pipe.to("cuda") # Move to GPU
# Generate an image with a simple prompt
prompt = "Woman in an elegant dress standing by a window, detailed lighting, 8k"
negative_prompt = "low quality, blurry, deformed"
# Create the image
image = pipe(
prompt=prompt,
negative_prompt=negative_prompt,
num_inference_steps=30,
guidance_scale=7.5
).images[0]
# Save the image
image.save("generated_image.png")
# Advanced example - fixed seed and additional parameters
import numpy as np
# Set seed for reproducible results
seed = 42
generator = torch.Generator("cuda").manual_seed(seed)
# Advanced parameter settings
prompt = "A dramatic scene with explicit details, cinematic lighting, high resolution"
image = pipe(
prompt=prompt,
negative_prompt="ugly, deformed, disfigured, poor quality, low resolution",
num_inference_steps=50, # More steps for higher quality
guidance_scale=8.0, # Increase prompt fidelity
width=768, # Adjust image width
height=768, # Adjust image height
generator=generator # Fixed seed
).images[0]
image.save("high_quality_image.png")