Home
Artists
Search Recent Random
Posts
Search DMs Popular Hash Lookup Tags Random
Importer
Import FAQ
Register
Partychan Telegram Find&Fuck Waifus ThePornDude
Home Artists Posts Import Register

About Stable Diffusion: (Patreon)

Published:
2022-08-29 13:03:55
Imported:
2022-09

Content

Hi all. Stable Diffusion was released 7 days ago and everything been so crazy, each new day there is new options and tools build on top of it.

I would like to share some of my thoughts on the project and my build.

First, the model 1.4 seen a little heavier than the leaked one. With my 0.2 I did managed to improve memory a little, but 448X448 was my current limit for 4Vram.
I can probably improve the memory bi digging into the model and changing some stuff, but Emad is aready teasing a 1.5 version, so I see little point on focusing on this right now, instead I will focus on adding tools to the build and easy optimizations.


In 0.2 there is the option to use img2img. I used the official code as base, so I can't say how other scripts implemented this tools. But from my experience and Patron comments it seen that the strength is inverted. Plus I had to use a really high number (99.3, 99.7) to generate a good blend, so I definitely need to work a little more on that.


Future updates will include:

  • Inpainting
  • Upscaler
  • Infinite zoom animation
  • Video input
  • Better interaction and usability
  • Textual Inversion 


Comments

Anonymous

Great work so far!! well done! Can I suggest an idea that location picked for saves is saved on close in a future update?

Anonymous

I struggle to figure out how to use the IMG2IMG well at all, with or without the masking. If I take a photo, crop it to the same size as output, tell it to use as a reference (such as a photo of myself standing on grass) and input something like "Nic Cage standing in grass", very little happens. If I used Photoshop to cut myself out of it so that I'm just a sillouhette on the grass and input it, nothing much happens. If I mask myself into an alpha channel (saved as a TIFF as non-24 bit PNG alpha channels aren't a thing), same thing, nothing happens. I'm probably using this tool completely wrong, does anyone have any pointers or docs to read on what I'm supposed to be doing? I know you can sketch something and have it expound, but I've also seen examples of people replacing someone sitting on a park bench with a panda sitting on a park bench via masking, so I know this stuff should be do-able somehow.