Stable Diffusion AI Image Generator

This is an interesting (free \ no signup) site which gives you a "street view" style panorama of an environment generated from your prompts.

Latent Labs
 
I just love IMG2IMG in Auto1111... I remastered some more sprites from 90s video games. Quite interesting how the modern iterations of these games have redesigned the creatures when the original designs were already quite impressive but just lacking detail.

Here's a more "faithful" version of the classic Doom Imp.

Doom Imp_cleanup.jpg

Imp - Doom video game monster - Critter profile - Writeups.org.jpg
 
OnlyOneKenobi said:
That's almost definitely Emma Roberts, did you use her name in your prompts? I guess they moved on to the next Emma after "Watson" when training this model.
Click to expand...
Yeah, so far she works the best. Emma Watson , not so much hehe. She looks kinda off.
Will post later.
 
Yip, looks like SD does okay with both Emmas. This particular model seems a bit anime-influenced so the proportions of the facial are a bit off, but still ok. Amusingly, the combined Emma(3) Roberts/Watson looks more real than the other two.

Emma1.jpg
Emma2.jpg
Emma3.jpg
 
ControlNet for Stable Diffusion is MINDBLOWING. It allows so much more control over your generations from TXT2IMG and IMG2IMG, and the possibilities are virtually endless using this.

Here are two videos which demo how to install and how to use it.


 
OnlyOneKenobi said:
ControlNet for Stable Diffusion is MINDBLOWING. It allows so much more control over your generations from TXT2IMG and IMG2IMG, and the possibilities are virtually endless using this.

Here are two videos which demo how to install and how to use it.


Click to expand...
Cool!!
 
LORA model training is a game changer, quick and easy to train new faces and styles right in Automatic1111 and the resulting model files are much much smaller than Dreambooth models. Furthermore, you can combine various LORA models with checkpoints and even other LORA models so the possibilities are virtually endless.

I trained a new LORA model for a face most South Africans should know pretty well.

News reader 3.5 is almost ready for duty ;)

02163-482552821-Riaan Cruywagen, cyborg, cyber punk, mechanical face and exposed circuitry, wi...png
 
OnlyOneKenobi said:
LORA model training is a game changer, quick and easy to train new faces and styles right in Automatic1111 and the resulting model files are much much smaller than Dreambooth models. Furthermore, you can combine various LORA models with checkpoints and even other LORA models so the possibilities are virtually endless.

I trained a new LORA model for a face most South Africans should know pretty well.

News reader 3.5 is almost ready for duty ;)

View attachment 1478213
Click to expand...
Dri@@n Cyborgen 2.0
 
You must log in or register to reply here.
Back
Top
Sign up to the MyBroadband newsletter
X