eolecvk/model-test

1---2language: 3  - en4thumbnail: TBD5tags:6- stable-diffusion7- stable-diffusion-diffusers8- text-to-image9datasets:10- ChristophSchuhmann/improved_aesthetics_6plus11---12 13 14# Mini Stable Diffusion (miniSD)15 16MiniSD is a latent text-to-image diffusion model that has been conditionned on 256x256 images through finetuning.17 18## Examples19 20WIP21 22## Usage23 24```25!pip install diffusers==0.3.026!pip install transformers scipy ftfy27```28 29```30import torch31from diffusers import StableDiffusionPipeline32from torch import autocast33 34# TODO: change model_id to "lambdalabs/miniSD"35pipe = StableDiffusionPipeline.from_pretrained("eolecvk/model-test", torch_dtype=torch.float16)  36pipe = pipe.to("cuda")37 38prompt = "Yoda"39scale = 1040n_samples = 441 42# Sometimes the nsfw checker is confused, you can disable it at your own risk here43disable_safety = False44 45if disable_safety:46  def null_safety(images, **kwargs):47      return images, False48  pipe.safety_checker = null_safety49 50with autocast("cuda"):51  images = pipe(n_samples*[prompt], guidance_scale=scale).images52 53for idx, im in enumerate(images):54  im.save(f"{idx:06}.png")55```56 57