/hdg/ - Stable Diffusion

Anime girls generated with AI

Index Catalog Archive Bottom Refresh
Name
Options
Subject
Message

Max message length: 12000

files

Max file size: 15.00 MB

Total max file size: 50.00 MB

Max files: 4

Supported file types: GIF, JPG, PNG, WebM, OGG, and more

E-mail
Password

(used to delete files and posts)

Misc

Remember to follow the Rules

The backup domains are located at 8chan.se and 8chan.cc. TOR access can be found here, or you can access the TOR portal from the clearnet at Redchannit 3.0.

US Election Thread

8chan.moe is a hobby project with no affiliation whatsoever to the administration of any other "8chan" site, past or present.

(2.06 MB 1824x1248 catbox_13g9wv.png)

(2.14 MB 1824x1248 catbox_44619z.png)

>>26995 Yeah this is great, I love the Roxy one.
>>26995 Do you have a clara lora?
>>26998 nah, illustrious is enough on its own these would be better with face detailer but lazy prompt: >clara \(honkai: star rail\), white hair, very long hair, red eyes, loli, [[flat chest]] red jacket, long sleeves, long_coat, trench_coat, open coat, breasts out, open clothes, 1girl, solo, naked_coat,
>>26999 impressive that the model can just gen her like that
>>27000 I was also surprised that it went ahead and did the hair intakes on its own when double checking the prompt, let alone the general jacket design. It's just missing the collar-choker
>>27000 >>27001 I've heard the dataset cutoff is actually April 2024 though the base dataset is danbooru2023, so that's not too surprising since she should have >300 images. I'm not sure what characters would be best to test this theory, probably whatever BA chars released at the beginning of the year in JP. Also super clean inset, inpainted or RNG? Can I get a box for that style mix?
>>27002 >probably whatever BA chars released at the beginning of the year in JP. Shupogakis probably? Their first picture was the last week of march so might barely make the cutoff
>>26999 >>27001 What do you use for negative?
>>27002 if it's the x-ray you're talking about, daram usually does x-ray variants so it got baked in. still surprised how well baked it got in considering this is an "undercooked" model https://files.catbox.moe/bmzpwk.png there's an undercooked nababa lora I was using at .4 strength and I think I fucked up the settings for it since it's 400mb. I can litterbox that if you want >>27004 general negative I'm running is >multiple views, sound effects, lowres, (bad), ( dark_skin, dark-skinned_male, sagging_breasts, pubic hair ) text, error, fewer, extra, missing, worst quality, jpeg artifacts, low quality, watermark, unfinished, displeasing, oldest, early, chromatic aberration, signature, extra digits, artistic error, username, scan, white_pupils, monochrome, limited_palette, greyscale, blurry, blurry_foreground, motion_blur, censored, bar_censor, also, sampler: euler cfgpp, 1.7 cfg, sgm uniform. I feel like sampler makes a bigger difference than negatives
Illustrious is a lot like NAI where it can gen characters with less data using detailed tags. Not perfectly obviously but pretty good
>>27006 It's also pretty useful to have a list of associated tags for certain outfits
Is it possible to remove Nahida's hair accessory with Illustrious? The thing is literally glued to her head with pony
Are best quality and masterpiece replacing score_9 and score_7 in my prompts or is the era of quality tags gone?
>>27008 Yes. >>27009 No. Luckily with this one it's a short neg and just adding "masterpiece, best quality" at the end of your prompt. Maybe "very aesthetic", but I'm too lazy to do A/B to see how much that affects the style.
>>27011 Holy shit, we finally have the technology.
>>27012 May I present to you, ojou-sama Nahida? (tailed hairstyles still try to add them but you can get rid of em with a replacement decoration)
>>27014 Perfect, definitely going to have fun with the model later.
I couldn't get it to spit out shupogakis but it does know Dress Hina, a character released in January 2024 so the cutoff is somewhere in 2024.
It knows dress kayoko too and that's probably as far as I go because I can't think of a character released after that and it doesn't know Hikari so it's likely the cutoff is somewhere in march 2024
One thing I'll complain about is that I hate the "default" artstyle of illustrious. It feels like base pony where the default/base art style was ass and I basically needed to prompt an artist to get anything resembling good.
>>27018 imo leave "good" default styles to shitmixes, base model should just be malleable
>>27019 I am pretty blown away by illustrious ability to just prompt artists Pony needed loras for basically everything related to characters and artists while illustrious doesnt
>>27020 yeah i havent trained anything or downloaded any finetunes ive just been playing with the base model since it came out
>>27007 >pretty useful to have a list of associated tags for certain outfits https://danbooru.donmai.us/related_tag?commit=Search&search%5Bcategory%5D=General&search%5Border%5D=Cosine this has been extremely useful for nai prompting and should be similarly useful for illustrious
after skipping out on naiv3 i have quite a lot to learn in the ways of style-fu hopefully there's a few in here who aren't immediately well known: https://files.catbox.moe/15jjm5.txt
>>26928 me again, I've decided to make a dataset for this but I'm not sure if my captioning will be much good. I won't be able to train the lora myself since my PC is on the lower end and my ram sticks may be going bad. If anyone would like to try with this dataset I can provide it.
(2.34 MB 1248x1824 catbox_l29n58.png)

(1.87 MB 1248x1824 catbox_bl7c4u.png)

(1.99 MB 1248x1824 catbox_hwov6y.png)

(1.83 MB 1248x1824 catbox_fw1o3a.png)

>>27024 for illustrious? sure why not, i'm not using my GPU right now anyways
https://mega.nz/folder/OoYWzR6L#psN69wnC2ljJ9OQS2FDHoQ/folder/DkJ2nLTQ this style wasn't intended for /h/ but this *is* /hdg/ :^) >>27024 >>27025 ah I was gonna offer to bake it as well if you compile a dataset, the third nahida in the post above uses onono imoko as the base style and you can tell it's pretty good for mixing already but it's pretty good, eg it understands the usage of halftones: https://files.catbox.moe/ke2ng7.png so it probably won't need to be baked too hard to get to 100% accuracy either way post the dataset here when you make it i'm hoarding
(1.80 MB 1920x1248 00174-3744801659.png)

(2.30 MB 1248x1920 00140-1245248580.png)

(2.60 MB 1440x1920 00482-1481276938.png)

(1.11 MB 1024x1280 00646-1160354527.png)

>>26938 >>26939 I can't believe I missed kamiposting hours >>27026 as a certified sugoi stan I can't wait to try this out
>>27023 >>27025 cute >>27027 sexo Help, I'm uohverdosing hard, I need a break.
(1.71 MB 1152x1632 00017.png)

>>27025 >>27026 Thanks, either of you are free to take a shot at it. Again I'm kind of new so I'm not sure if the tags are fine (I scraped them from gelbooru and pruned some bad ones), I also separated alt sets from the main ones since they range from small details like glasses to alternate clothing etc. I made sure to tag any images with text or copyright just in case. https://files.catbox.moe/q94jq6.rar
Prompting musouzuki on illustrious seems to generate text all over even if I put text and speech bubbles on neg
bake bread
(2.01 MB 1248x1824 catbox_1fnm50.png)



Forms
Delete
Report
Quick Reply