Wall Street Journal Hedcut Style for Stable Diffusion
Readers of the Wall Street Journal (WSJ) are familiar with the distinctive style used to create portaits of their writers and subjects. This is a fine-tuned stable diffusion model that can be used to create hedcut-styled images using the prompt wsj hedcut of <subject>. This model can also be used in a DreamBooth environment to train a face or other subject for custom and unique hedcuts.
*If you use this model, please make an effort to attribute the principal artist of the training images, Noli Novak, along with the other disclosures and restrictions required by the license.
#!pip install diffusers transformers scipy torch
from diffusers import StableDiffusionPipeline
import torch
model_id = "dmillar/wsj-hedcut-v1"
pipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16)
pipe = pipe.to("cuda")
prompt = "wsj hedcut of a woman"
image = pipe(prompt).images[0]
image.save("./woman_hedcut.png")
Sample Images
Prompt: "wsj hedcut of audrey hepburn, portrait, detailed, sharp, black and white, pleasing, white background", Steps: 20, Sampler: "Euler a", CFG scale: 7, Seed: 231828633 Prompt: "wsj hedcut of the Kylian Mbappe, male, portrait, detailed, sharp, black and white, pleasing, white background", Steps: 20, Sampler: "Euler a", CFG scale: 7, Seed: 1863052262 Prompt: "wsj hedcut of tom hanks, portrait, detailed, sharp, black and white, pleasing, white background", Steps: 20, Sampler: Euler a, CG scale: 7, Seed: 224907260
- Downloads last month
- 37