Create Stylized Vector Logo Chibi Characters with Stable Diffusion

This tutorial demonstrates a workflow for creating stylized chibi character artwork with Stable Diffusion, including custom model training, prompting, inpainting, outpainting, and a hypernetwork. It then uses Adobe Illustrator’s Image Trace and export options to convert the raster result into vector artwork.

Syllabus

Introduction
Examples of personal realistic artwork/motivation for creating heavily stylized artwork
Preview of possibilities
Materials and tools needed for the workflows
More detailed explanation of workflow
Reasoning for using Adobe Illustrator to convert raster image to vector image
Blender scene render overview
Don't use renders shot from the rear perspective
How to properly l subject for best image training
Using diffusers vs. checkpoint files
Enable git LFS to make sure large files are cloned to your machine
DreamBooth training script overview
Difference between two workflows
Method 1
Method 2
Two trained models as an experiment, left model is superior
Text prompts in AUTOMATIC1111
Finding the hero image from prompting
Creating the inpainting model from your trained model
Testing the inpainting model
Analyzing the sample outpainting outputs
Outpainting upwards
Short timelapse full timelapse in description
Using the waven chibi hypernetwork
Compositing the new hypernetwork output layers
Adobe Illustrator introduction
Image Trace Demonstration
File export options
Closing thoughts