Important: Usage Responsibility

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

This model is provided without a built-in safety checker. While it is not designed for inappropriate content, users are expected to use it responsibly. By accessing this model, you agree to adhere to the CreativeML Open RAIL++-M license and take full responsibility for your generations.

🌃 SDXL-Lightning-8-Steps (HyHorX Edition)

Developed by: HyHorX AI Lab
Model Type: Faster Diffusion-based text-to-image generation model
License: CreativeML Open RAIL-M

This model is a high-speed merge of Stable Diffusion XL 1.0 Base and ByteDance's SDXL-Lightning 8-step UNet. It is optimized for ultra-fast generation while maintaining high-quality architectural and cinematic details.

🚀 Usage Details

To get the best results (like the Cyberpunk city preview), use the following settings:

Inference Steps: 8 steps (Strictly)
Guidance Scale (CFG): 0 (or 1.0)
Sampler/Scheduler: Euler Ancestral (Euler A)
Timestep Spacing: Trailing

💻 Example Code (Diffusers)

import torch
from diffusers import StableDiffusionXLPipeline

pipe = StableDiffusionXLPipeline.from_pretrained(
    "HyHorX/SDXL-Lighting-8-steps", 
    torch_dtype=torch.float16, 
    use_safetensors=True
).to("cuda")

prompt = "A majestic cinematic view of a futuristic city with neon lights, high detail, 8k"

image = pipe(
    prompt=prompt, 
    num_inference_steps=8, 
    guidance_scale=0
).images[0]

image.save("result.png")

🛠️ Optimization Lab
This model was merged and exported in fp16 format to be lightweight (~6.95 GB), making it accessible for consumer GPUs with lower VRAM (like GTX 1060 6GB/8GB).

Created with ❤️ by HyHorX AI Lab during a deep-learning optimization session.

Downloads last month: 30

Model tree for HyHorX/SDXL-Lighting-8-steps

Base model

stabilityai/stable-diffusion-xl-base-1.0

Finetuned

(1160)

this model