Home » posts_tagged Stable Diffusion

Where it all began

- Posted in Gillian by

Somewhere around 1998, on a whim, I bought the X-files season 1 box set and discovered Gillian Anderson, and I've been a fan ever since.
Yes, I even have autographed photos...
Lately, I've been thinking I'd mix up my desire to re-watch X-files and play with Stable Diffusion, specifically the Low-Rank Adaptation (aka LORA) that let you model elements in it.
So I spent a few hours making screenshots and building my first LORA, obviously of Gillian Anderson, aka Scully.
The first results are encouraging (I'm far from a specialist, and the power of Stable Diffusion is matched only by its complexity):
enter image description here

As the X-Files show took place over several years, Gillian evolved visually over time, so I challenged myself to do one LORA of Scully per season.
enter image description here

These two images come solely from a LORA made from the captures of the first two episodes, a job that requires ... a lot of time (capture, crop, describe and then sleep while my 2080 is working).
But I'm pretty happy with the first results, even if I'm learning as I go. See you in a few weeks for the release of the hopefully nice Scully_s01.safetensors.

I want to believe.

More than images, ideas

- Posted in D&D by

Recently I played quite a bit with Stable Diffusion, a program for generating images from text, and it's amazing.
I'm not going to try to show here some unmatched ability (that I don't have) to generate extraordinary images, the Internet is full of tutorials on the subject (I recommend for example Sebastian Kamph's YouTube channel, he is both talented and educational).
However, let's talk about the unexpected fact that these images ... stimulate my imagination.
Indeed, in order to illustrate in an original way the textual descriptions of my imaginary world of Gilianar, I played with Stable Diffusion and besides the quality of the images produced, I discovered with surprise how this tool gave me new ideas.
Let me explain.
Let's say I want an image of a dragon perched on an old tower. With the following prompt

ancient yellow and green dragon with a large wingspan perched on an old broken stone tower

I get this for example:
enter image description here
Not bad for a start.
Stable Diffusion's modus operandi is to generate a lot of images, then once you like one of them, to modify it in a loop until you get a satisfactory result.
So let's modify it, but this time starting not from the textual description of the beginning, but from the image above (yes, Stable Diffusion is amazing, I already said it).
By playing with the enigmatic parameter Denoising strength, one introduce more or less chaos in the following images and so that's where magic happens, it's as if a part of a story was written without us.
After a few generations, an image clicks in my head.
"Oh yeah, that's cool, it looks like a forgotten tomb, let's keep this one!"
enter image description here
So much for the grave, now it looks more like she has boobs and she is instead near the entrance to her nest, on which she watches fiercely.
What if she protects her eggs? Deal!
A few random images later with guarding her eggs added to the description, this one fits my mood:
enter image description here
Definitely a good start for a place and the history that goes with it!

Thank you Stable Diffusion, I was just looking for original images to illustrate my texts, but I actually found more than that, now I own an iterative idea machine!