The Pokeball Machine

The Pokeball Machine is a Dreambooth model for the pokeball concept (represented by the pkblz identifier). It applies to the wildcard theme. It is fine-tuned from CompVis/stable-diffusion-v1-4 checkpoint on a small dataset of pokeball images (i.e., images of the red-white original pokeball). It can be used by modifying the instance_prompt: a pkblz ball in the middle of a miniature jungle

This model was created as part of the DreamBooth Hackathon 🔥. Visit the organisation page for instructions on how to take part!

Fine-Tuning Details

  • Number of training images: 31
  • Learning rate: 2e-06
  • Training steps: 800
  • Guidance Scale: 10
  • Inference Steps: 50-75

Output Examples

a blueprint photo of a pkblz ball a photo of a cybernetic pkblz ball, wide shot a photo of a pkblz ball in the style vintage disney
a photo of a mosaic pkblz ball lying in an antique temple a photo of a detailed ornate pkblz ball a pkblz ball underwater
a pkblz ball in the middle of a miniature jungle a pkblz ball underwater a mystic pkblz ball, trending on artstation
a pkblz ball underwater, trending on artstation a wooden pkblz ball a pkblz ball hovering over a pond
a pkblz ball on a sunny tropical beach a steampunk pkblz ball, trending on artstation a colored pencil sketch of a pkblz ball
a photo of a spectral ornate pkblz ball, trending on artstation, realistic a sunset photo of a pkblz ball a watercolor photo of a pkblz ball

Usage

from diffusers import StableDiffusionPipeline
import torch

device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
pipeline = StableDiffusionPipeline.from_pretrained('simonschoe/pokeball-machine').to(device)

prompt = "a pkblz ball in the middle of a miniature jungle"

image = pipeline(
    prompt,
    num_inference_steps=50,
    guidance_scale=10,
    num_images_per_prompt=1
).images[0]

image
Downloads last month
1
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.