Playing with Stable Diffusion XL 1.0

August 8th 2023

"Dogwarts"

A dog dressed as a hogwarts wizard Positive Prompt:

award winning photography, a cute dog in the style of a hogwarts harry potter wizard

Negative Prompt:

text, watermark

⬛ Base Image ✅ Refiner Image

(a bad attempt at) Barbenheimer

Barbie walking away from an explosion Positive Prompt:

award winning photography, realistic barbie doll and a 600 lb roll of film being exploded in an atomic blast

Negative Prompt:

text, watermark

⬛ Base Image ✅ Refiner Image

Melting Computer in the style of surrealist painting "The Persistence of Memory"

melting computers in the style of a salvador dali painting Positive Prompt:

melting computers in the style of salvador dali painting

Negative Prompt:

text, watermark

⬛ Base Image ✅ Refiner Image

alt Positive Prompt:

award winning photography, cookie monster from sesame street eating a messy cookie riding fast in a top down convertible car with cookie tires

Negative Prompt:

text, watermark

⬛ Base Image ✅ Refiner Image

Vin Diesel and Grilled Cheese. A classic combo

alt Positive Prompt:

award winning photography, vin diesel with a bunch of grilled cheese falling onto his face

Negative Prompt:

text, watermark

⬛ Base Image ✅ Refiner Image

Vaporwave Ballerina

alt Positive Prompt:

award winning photography, vapor wave aesthetic with dark colors of a ballerina with a glitchy sign lurking in the background

Negative Prompt:

text, watermark

✅ Base Image ⬛ Refiner Image

How were these made?

I used ComfyUI with Stable Diffusion XL Base 1.0 and Stable Diffusion XL Refiner 1.0 to generate these images

ComfyUI is a UI that allows users to create custom pipelines to run different stable diffusion models.

You will need somewhat of a beefy graphics card (recommended) or CPU (very slow) to generate these images. I generated these images on an AMD 6850 XT. They took about 7 minute a piece to generate and then run through the refiner.

Follow the ComfyUI installation procedure
Download the SDXL base and refiner models from Huggingface (the file that ends in .safetensors). I used the ones that don't have vue in the file name.
Place the downloaded models in /ComfyUI/models/checkpoints
Run the UI by following the ComfyUI Readme
Import my workflow JSON download (NOTE: You may need to change the filenames in the JSON to match the files you downloaded)
Voila! Enjoy using SDXL! Your screen should look something like this:

ComfyUI

Suggest a change to this post here! (requires a GitHub account)