Assorted Stable Diffusion notes and Tutorials
Header image by anon
πππππTUTORIALSπππππ
Installing Stable Diffusion (Forge)
PonyXL notes and prompt templates
ππππππππππππππ
Have the tutorials helped you?
Gen my Fluffbold: Hymonie Fluffscale
And then drop a gen on the booru/thread/discord/email/times square.
Fangen Gallery
Boring Socials
Send me hatemail or feedback: fluffscaler@proton.me
Fluffscaler on Discord.
My personal Discord https://discord.gg/xnm7An55
Most of my older gens are on the booru [NSFW]: https://e6ai.net/posts?tags=fluffscaler
Moved to Twitter for new uploads [NSFW]: https://x.com/Fluffscaler
If you wanna know the base gen prompt data for something on the booru then feel free to message on Discord, I check on occasion.
Most of them use one of the 2.5D kemono templates from the PDXL guide, with altered weights or a minor lora added so they really aren't terribly complicated.
Various links to /sdg/ genners
/trash/ Booru Uploads
Not everything under "anonymous_director" is /sdg/ of course.
Chubby Snek Anon
Big...Some huge fatties, but also some moe.
Bre Anon (Andrew)
!WARNING!WARNING! BOY ALERT WEEWOO WEEWOO BOYS IN THE VICINITY
Lop Anon
Yep. It's Lopunny.
Invoke Anon
Invoke high series. King effortgen.
Anomalous_Anonymous
Gigachad. Makes very high quality loras. Actually cares enough to remove watermarks from training data, something that can't be said about 90% of lora makers.
WARNING! Blogposting ahead!
What's Meta?
PDXL with inpainting. The tools to get quality are too easy not to use. Variation seed spam has cooled down these days at least.
Krita & Invoke
Been trying both since I inpaint a lot
WebUI(forge) | Krita | Invoke |
---|---|---|
+ Pretty fast | +An actual functioning image editor | +Best looking UI |
+ All in one place | +No need to switch back and forth | + Fast |
+ Can cancel the gen if I see pseudolimbs | - Questionable control over variables. Somewhat awkward UI for prompting, loras, etc. | - Different text encoder means style differences |
- Pretty fucking bad UI, really buggy on firefox | - Slow ass workflow having to create context boxes if you are a control freak (me) | - Switching between WebUI and Invoke is still awkward, memory hungry |
- I have to either eyeball values for padding or create ghetto bounding boxes with little pixels specify context boxes |
Surprisingly I like the WebUI the most. Might be because I have used it enough that I pierce the veil on what denoise and padding will really bring out the pores on that anus, but it is what it is. The other two definitely have their place, but I'll probably stick with WebUI inpaint and gimp for manual edits.
Sloppa
I am the slop of my gen. 1girl is my body and on_back is my blood. I have created over a thousand variation seeds. Unknown to OOM, nor known to Inpaint. I have withstood pain to create many prooompts. Yet, those hands will never hold anything. So as I pray, Unlimited Slop Works.
SD3
Trash. "It's salvageable" is cope. If SD3 training was a reality in the future, then we would have a competent SDXL model by now. StabilityAI is likely just gonna go bankrupt soon. Their licensing terms certainly feel like that's the expected future, designed to be palatable to a buyer, who can then exploit the terms.
Flux
Neat, very evident that half the staff is just ex Stability employees. Basically how SD3 should have been, if they make it fully open that would be sick since it's probably not hard to train on by itself, problem is more the hardware requirements being far bigger than SDXL.
Trolling&Derailing
Not sure what's up with the thread these days, half of it is just bait.
THE ELUSIVE SHITPOSTER CAUGHT?!
Advanced Autism. Quite impressive really.
https://desuarchive.org/trash/thread/68583843/#q68591719
https://desuarchive.org/trash/thread/68588962/#q68593723