sd15-fullft-human-fullbody

SD1.5 — Human Generation Fine-Tune (CosmicMan-HQ)

This is a diffusion model fine-tuned from stable-diffusion-v1-5/stable-diffusion-v1-5 using the CosmicMan-HQ dataset, specifically designed for generating realistic full-body human images.

---

Overview

• Base model: Stable Diffusion v1.5
• Framework: 🤗 Diffusers
• Dataset: [CosmicMan-HQ 1.0] — a high-quality dataset of full-body human images
• Training objective: Fine-tune the UNet for realistic full-body human generation
• Resolution: 512×512
• Precision: bfloat16
• Epochs: 1 (≈68k images)
• Frozen modules: VAE and text encoder (CLIP)
• Task: Text-to-Image (human-focused generation)

---

Usage

from diffusers import StableDiffusionPipeline
import torch

pipe = StableDiffusionPipeline.from_pretrained(
"DorukBenlii/sd15-fullft-human-fullbody",
torch_dtype=torch.float16
).to("cuda")

prompt = "full-body portrait of a person, studio lighting, white background."
negative_prompt = "lowres, blurry, extra limbs"

image = pipe(
prompt,
negative_prompt=negative_prompt,
num_inference_steps=30,
guidance_scale=6.5,
height=512,
width=512
).images[0]

image.save("sample.png")