/b/ - Random

Quality is not guaranteed


New Reply[×]
Message
24.3MB total
Tegaki
Password
[New Reply]


If (You) don't like something then post what (You) like instead

These niggas say they heard of me, I ain't heard of you
Get the fuck up out my fuckin' face, 'fore I murder you
Bitch niggas always jackin' Blood, but I know they fu
Whole squad full of fuckin' killers, I'm a killer too


ClipboardImage.png U A
[Hide] (968.8KB, 1422x773)
I play around with AIslop, specifically Stable Diffusion.  I have attempted to make a LoRA file for a certain focks character.

I am dissatisfied with the results, to put it mildly.  You might even recognize her. 

I am using Automatic1111 and Stable Diffusion 1.5.  I trained the LoRA file with kohya_ss.  Source images came from the places you'd expect.

I have access to SwarmUI with SDXL, Flux DEV, and 3.5 available on the back end, but I don't know whether any of the tools I have can make a LoRA for any of those.  And the quality of what I get from them is very disappointing, given how much newer they are than 1.5.

Does anyone else around here mess around with image generation?
Replies: >>585 >>673 >>743
>>>1449
>What's it about stoner chicks that makes them really attractive?
Because the 90s stoner aesthetic is cool(do not bring up seth rogan)
Replies: >>2718 >>2740
>>2717
No it fucking isn't, every other druggie subculture clears high diff (incl. '60s libtard hippies)
i hate junkies
>>>2717
>No it fucking isn't, every other druggie subculture clears high diff (incl. '60s libtard hippies)
Nuh uh stoner metal cool
Replies: >>2741
>>2740
Sounds wooooooke
Replies: >>2747
>>2741
@floof
A word filter to substitute "wookie" for "woke" would be funny.
bully fox robbie 1.png U A
[Hide] (3.5MB, 2304x1792)
not one day has passed without me defiling this character with worthless ai garbage
i made a lora of her on civitai
Replies: >>3101 >>3102
>>3100
>not one day has passed
Bro it hasn't even been a week 💀
>>3100
we need artfags to save robbie before its too late
Replies: >>3105
>>3102
Robbie doesn't deserve to be saved at this point
Replies: >>3107
who?
>>3105
Why?
Replies: >>3108
>>3107
She's an ugly ass nigger bitch that's been shilled by this AI goycattle to the nth degree. The original drawfag forgot about her for a reason
Replies: >>3109 >>3114
>>3108
trvke
>goes into ai slop thread
>complains that there is ai slop
Replies: >>3111
>>3110
Except she hasn't been contained to just this thread
Replies: >>3112
>>3111
guess nobody is allowed to gen averi or robin or any big name furfag because they're not exclusive to this thread
anon said so
Replies: >>3113
>>3112
Glad you understand
>>3108
She may very well be an ugly ass nigger bitch, but being shilled by an AI goycattle (goyïm is out, fyi.) is not a valid justification for not being saved, whatever that means.
bully fox robbie 6.png U A
[Hide] (2.5MB, 1728x1344)
sorry lads i don't consider the opinions of shartycucks as worthwhile
Replies: >>3120 >>3124
>>3116
Nobody here is even a shartycuck
Replies: >>3121
>>3120
im a shartyclucka
Replies: >>3125
1765726183456g.png U A
[Hide] (125.6KB, 600x600)
>>3116
Learn to draw nigger
Replies: >>3140
>>3121
No you're not, you're a shartychad
1765950153015-tegaki.png U A
[Hide] (16KB, 500x500)
>>3124
I learned how to draw nigger.
123422c479a77dd9467558ced38ce6aca5f25ba18ccad53e2e37d8af709ab9fd.png U A
[Hide] (687.2KB, 512x768)
So, I've been playing around with Perchance some more.  And I discovered something.

Perchance, at least with the generator I used, "Image Generator Professional," doesn't know about Frieren if you name her.  If you say "I want an image of Frieren in a pretty sundress," it gives you a generic animoo girl.  But if you describe her:  "short petite Elf woman, pale green eyes, silver hair, pigtails, wearing striped blouse, gold and white cloak, and black leggings," it immediately renders images of perfect Frierens (Frierenen?), with lots of little details, like her earrings, which I did not mention in the prompt, exactly like in the Chinese cartoon.  Not every detail is always exactly the same but it's always definitely her, very recognizable.  Usually when you're trying for a character an AI doesn't recognize, you have to do a very long, detailed prompt, and run dozens if not hundreds of iterations to pick out a few where there is a resemblance.

I am trying to figure out what this means.  Is the AI using a model that has Autism Elf waifu art in it, just not tagged with that name?  What other characters are in there?  I've been trying for Marcile from Dungeon Meshi but not having much luck so far.
Replies: >>3910 >>3911
>>3909
IDFK /g/ has a general for this crap, doesn't it? Ask there.
>>3909
depends on the model. some are trained as Human Syntax, eg; I would like a picture of this girl Frieren. Other models like the one you've described her are Tag Based, eg; Frieren \(beyond journeys end/the japanese name for it\), anime, girl, description, etc. it just depends which model you're using is all. the tag/prompt of \(thing\) just tells the model that you're looking for the specific thing it's from, for example; frankie foster \(fosters home for imaginary friends\), red hair, green hoodie would return something akin to her. Now, I gen locally and use tag based models the most because they're easier to parse when i'm looking for spelling mistakes or whatever than human syntax trained ones.
0e8a5efcedbf3fc3da23d54ad8e7d7bb4061bab8e3e346e64c1ed9fe270a01a5.png U A
[Hide] (750KB, 512x768)
Do LLMs count as slop? They probably do.

I think I killed Llama 3.  It cut off in mid-sentence:

>>> There are some individuals who call themselves "furries," who have great int
... erest in anthropomorphic characters, and seem generally to want to be anthro
... pomorphic characters themselves.  Consider a second group, who have an inter
... est in anthropomorphic characters, who say they are distinct from the first 
... group because they do not themselves want to be anthropomorphic characters. 
...  Does this claim have merit?  If so, is there any good name for the second g
... roup that makes the distinction clear?
A fascinating topic! The distinction between these two groups of 
individuals with an interest in anthropomorphic characters does have 
merit.

The first group, often referred to as "fursonas" or "fursuiters," actively 
engage in creating and role-playing their own anthropomorphic characters. 
They may create artwork, write stories, or participate in costuming and 
cosplay events featuring these characters. Many furries see themselves as 
part of a community that celebrates creativity, imagination, and 
self-expression through anthropomorphized animal forms.

In contrast, the second group might be referred to as "anthropomorphism 
enthusiasts" or "anthro fans." These individuals are drawn to the concept 
of anthropomorphic characters but do not desire to become them themselves. 
They may appreciate the art, storytelling, or entertainment value of anth
Replies: >>4581
ClipboardImage.png U A
[Hide] (111.2KB, 957x713)
Is it over, 'foo frens?
Replies: >>4576 >>4580
>>4575
No.
>>4575
It's only just getting started tho
>>4355
what a cute katte
Replies: >>4582
>>4581
I can take no credit. I just prompted for it.  I was hoping for a recognizable Black-chan eating taiyaki at a festival.  Sometimes the AI generates a good one.  I liked it.  🐈‍⬛
c8bfdd233f37277896765103d15e879574b8c8e6724c167a40eb11fdaeb1eccb.png U A
[Hide] (329.6KB, 512x768)
A wild fops appears.  Does she look familiar?

I ran a bunch, specified brown hair and a green turtleneck, and a few came out sort of like this one
My ai art creations:
https://mega.nz/folder/Vg5BQAbR#buNlVdoECji9vw_kxmGj0g
Replies: >>4632 >>4640
04e5e273-24fd-4105-bb12-9d589c2d712b.jpg U A
[Hide] (365.7KB, 1376x752)
>>4631
Fuck, forgot an example of my works
Replies: >>4639 >>4640 >>4658
>>4632
Nicely done. What tools did you use?

I've been experimenting with large language models using Ollama.  I had a bit of difficulty because it wanted to download and install enormous large language models and stick them all into my boot drive.  Eventually I figured out that I needed to make a symlink and point it at a big spinning platter drive that had more available space.  Now really big* models like Magistral take minutes to load instead of seconds, but I'll deal with it.  They run fine once they're all in VRAM.

*24 billion parameters is yuge if we're talking about something to run at home on consumer hardware.  I know there are models with trillions of parameters, for which you have to rent a dedicated server.
00358-1549471974.png U A
[Hide] (4.7MB, 2048x2048)
>>4631
>>4632
neat.
22ac2f8d-6180-48b8-951b-4cfb6d46275c.jpg U A
[Hide] (215.6KB, 1248x832)
>>>4632
>Nicely done. What tools did you use?
Since my PC is ancient and can't handle making ai stuff locally I use online tools. In the past I was using civit.ai but since now it prohibits "NSFW" stuff to free accounts I use chatgpt and grok for image-to-image stuff and these ones for text-to-image stuff:
https://pixai.art/en/generator/image
https://perchance.org/ai-text-to-image-generator

By the way do you have twitter or discord?
Replies: >>4667
trolling2.png U A
[Hide] (782.8KB, 1024x1024)
ede638fc-8bd0-4a9b-b696-f507d77134a8a.jpg U A
[Hide] (348.3KB, 848x791)
>>1526
I like this one
427b5393583a3ea9270c3dda19ab746b0e96be1cd270241cc051593d73e0fc4e.png U A
[Hide] (643.5KB, 512x768)
>>4658
Thanks.
>twitter or discord?
No.  I don't use social media.  I prefer antisocial media.
Replies: >>4668
>>4667
Oh, damn. That's nice.
catgirl test doodle.png U A
[Hide] (6.1KB, 512x512)
After many tries I finally got Framepack installed.  I am not sure what I expected, but it wasn't this.  This post has the quick doodle I put into it.  Prompt was "The catgirl dances gracefully, with clear movements, full of charm."
Replies: >>4785
catgirl test video.mp4 U A
[Hide] (1.1MB, 608x640, 00:04)
>>4784
And this is what it made over the course of several minutes.
Replies: >>4786
>>4785
The arm movements are interesting, the pace of them. The most potential that I see in generative AI [as a medium for human expression, and cheap novelty in general] is image prompting/better prompt interpretation like we have now combined with the abstract results of 2022-3 models.
Replies: >>4795
414000c784a0d6b3fbc3e4db47d5ac13e1fc0deabfd5ba29f31854fccdc4ade8.mp4 U A
[Hide] (877.6KB, 512x768, 00:04)
>>4786
There is a certain similarity to the way the animations come out, yes.  Asking for anything much longer or more involved than this tends to result in characters rotating their heads three hundred and sixty degrees.  It's cute, though.
1bb3e8a680e8601d90cd920eb2d709a7eae32a53dc5200912a909b4e6a9dc587.mp4 U A
[Hide] (2.2MB, 512x768, 00:04)
Sometimes the animations just come out cursed, even short and simple ones.  I should stop spamming, though.

The thing about generating video is, it takes a long time.  It's not like image generation where you can do big batches and sort through them for nice ones.  The time it takes even for a five-second video makes that prohibitive.
[New Reply]
195 replies | 80 files
Connecting...
Show Post Actions

Actions:

Captcha:

- faq - rules -
anthrochan 1.8.0