Experimentations in Stable Diffusion

Mad Scientist
For All Humankind
June 2024 - Present

[PAGE UNDER CONSTRUCTION]

Style Transfer

The following images were generated using IPAdapter in ComfyUI using a mixture of methodologies including image-to-image, image + image-to-image, as well as image + predetermined latent image to image. The styles transferred were from Chang Yu, a Chinese painter from the 20th century who was himself notable for combining traditional Chinese painting styles with Western styles. Chang Yu seemed to be a clever choice to experiment with for a project all about style transferring. It was my hopes to not only transfer the style of Chang Yu to sketches and photos, but also to examine how well stable diffusion models can interpret abstract art and painterly styles from different continents.

Yaoguais

A lot of times, we are often left bewildered by what AI might gift to us. In this case, imagine me leaving my computer to churn out what I had hoped would be a batch of Chang Yu-inspired pics of my cat, only to return to... this. I generated a few more batches only to get more of these creatures, or as I like to call them, yaoguais. I did what any reasonable person would do and promptly restarted my computer. Upon generating my next batch, however, I was back to getting cats. What happened? Some faulty NVIDIA drivers? The perfect gamma ray performing an amazing bit shift in my VRAM? Perhaps only the AI Omnissiah shall know. But what I do know is I was left with this collection of yaoguais, which over time, have begun to grow on me! In any case, it's important to document even failures and oddities however funny they are, so here they are!

Acknowledgements

None of this would be possible without the amazing open-source community:
• ComfyUI
• Stable Diffusion
• IPAdapter
• CyberRealistic Pony
• Flux

Tools Used

• ComfyUI
• Stable Diffusion