Oh no, the whole day went on Greenie again!

And it was completely worthless again, too..

-=-=-

Yesterday I tried writing much more descriptive comments for each of the images I'd drawn, and tried to feel my outlined cartoon style Greenie back into the trainer.
After three or four or five attempts, I eventually gave up with my struggles.
From multiple noses, to him wearing glasses again, and the ever present third arm and even a multitude of legs at one point, the whole thing just didn't work well in the slightest.

Honestly, it was a complete waste of time.

The previous LoRA engine is coming up with some nice enough none-outlined Greenies. Using plain Stable Diffusion model to "think" of some pretty pictures, then adding the derek.safetensors LoRA engine can come up with some pretty results.
If I use the derek_j.ckpt model instead of the basic Stable Diffusion, it ends up with a much more Greenie'r Greenie, but tends to forego as pretty a background.
I guess it all depends on how good of an output I'd like.

"A photograph of Derek painting a picture of a beautiful sunset over the town, easel, canvas, oil on canvas, artist" by #DrawThings ( StableDiffusion 1.5 with Derek.safetensors )


1. Original Stable Diffusion output.


2. Redrawn with "Cartoon Derek" replacing Derek, and with the "Strength" of the text description being high, at about 80%


3. Redrawn once more to clean things up a bit, bringing the text's strength all the way back down to 5%.

I'll let you decide which of those is "better"