#749 Sampler no es que haya probado mucho, se que cambie alguna vez por probar pero creo que con esas imágenes fue el por defecto k_euler_a
Steps suelo generar siempre al máximo, además que con las últimas versiones juraría que genera mucho más rápido.
Está la 1.6 de nmkd
SD GUI 1.6.0 Changelog:
Added Dreambooth training (24 GB VRAM required for now!)
Added support for prompt emphasis - () for more, {} for less
Added Model Quick Switcher: Press Ctrl+M to quickly change the current AI model
Added model folder manager: You can now add additional model folders to load models from
Pop-Up Image Viewer: When slideshow mode is enabled, Left/Right arrow keys change images
Pop-Up Image Viewer: Window can now be resized/zoomed in 25% steps using mouse wheel
Pop-Up Image Viewer: Added "Always on Top" option to keep window always on top
Added lots of hotkeys, documented on Github Readme
Words in prompt field can now be deleted using CTRL+BACKSPACE, like in most text editors
Model Pruning Tool: Added option to delete input file if pruning was successful
Fixed an issue where the Stable Diffusion process would be killed when cancelling
Fixed prompt queue not working after running first entry
I'm confident that Dreambooth training will require less VRAM in the future, but currently the only lightweight implementations are Linux-only. The one I included here is VRAM-heavy but quite fast, you can get basic results after 6 minutes, and the highest quality preset takes 80 minutes to train on a 3090 (around 50 minutes on RTX 4090).
A ver si pronto desciende la RAM que se necesita para el Dreambooth.
Dejo unas pics random de bichos más feos que dios.
