File: 00039-2816373040.png (1.31 MB, 1112x1248)

1.31 MB PNG

>mfw resource news

>>101606189
She can't see - OK?!

File: 00038-2816373039.png (1.36 MB, 1112x1248)

1.36 MB PNG

>>101606196
OK

File: Frankie Carbone 01.jpg (50 KB, 1152x768)

50 KB JPG

Frankie Carbone from Goodfellas. Not really but okay in the context I guess. CinematicRedmond.

>https://youtu.be/lBeiuhb36RY?t=645
I know youtube is not preferred but this is Yoshiko... she's like 70+ these days now. Amazing artist.

File: 00043-2139679344.png (1.72 MB, 1112x1248)

1.72 MB PNG

>>101606275
>>101606296
He's awfully sweaty anon

File: Ted4.jpg (55 KB, 1344x768)

55 KB JPG

>>101606354
It must be the guilt :)
No
>Frankie Carbone from Goodfellas, crying, deep motivational scene,
>film grain, analog film, cinema, directed by Scorsese
Sometimes these catch on with great detail.
I'm still in awe of this Ted Bundy... (not stroking my own cock but the model and seed).

>>101606368
Well, it wasn't very nice of him to kill those people

>>101606407
Of course but do not judge, he was born that way.
This planet is doomed. Otherwise we wouldn't have these polarized issues everywhere.

Last known picture of Frankie Carbone - guard is giving him the tazer and after this he was supposed to show up in Rikers Island.

File: Frankie Carbone 03.png (1.04 MB, 1152x768)

1.04 MB PNG

>>101606435

>>101606354
uooohhhh catbox?

File: Lucchese 01.jpg (94 KB, 1152x768)

94 KB JPG

Still quite brappy.

Does SD3 have potential? SDXL was bad for about 1-2 months.

>>101606455
fuckhound

>>101606455
what's the prompt for this kind of hair/eyes?

>>101606811
I don't think it has any potential, just waste of bandwidth.
If you want something cool use ipAdapter with SDXL etc.

File: leosamsMoonfilm_filmGrain(...).png (1.91 MB, 1024x1536)

1.91 MB PNG

>>101606811
yo. potential? everything has potential. SD3=DOA

File: file.jpg (151 KB, 1024x1024)

151 KB JPG

>>101606841
it's a long fringe that covers her eyes, so try something like "long fringe that covers her eyes". it will work if the model is good

>>101605239
When using attention couple, how would I go about prompting an interaction betweem two characters? Does one prompt with one character get the action described out with the other just describing the 2nd character or what?

I've been out of the game for months, what's the best anime model now?

File: file.jpg (282 KB, 1024x1792)

282 KB JPG

300m

File: Librarian01.jpg (68 KB, 1152x768)

68 KB JPG

Uhh I thought You already returned all of your library books, anon.
We need to sit down and discuss about it right now.

>>101606926
You need to enforce weights
>dual focus, two characters, interacting,
>long legged puertorican
>short legged mexican
And just work up from there. Then enforce with repeating tokens but different .

File: Example.jpg (78 KB, 1152x768)

78 KB JPG

>>101607074
professional photograph, cinematic,
dual focus, two characters, interacting,
long legged puertorican,
short legged mexican,

film grain, analog film,

File: breakdomainrealistic_M205(...).png (1.5 MB, 1024x1536)

1.5 MB PNG

>>101606959
you compiling a dataset? 300m? gimme a reference.
>>101607046
very nice.
>>101607074
did you make that neon genesis image? thats some serious quality.

>>101607134
I don't know, I don't often generate anime outside joke content.

>>101607127
So basically three prompts if I'm following right?
One for background/composition/interaction.
Then one for character A.
Then another for character B?

>>101607159
No it's all one prompt. Try it out and do more.

>>101607159
Also post your results here and we can dissect them.
Suppose you have a correct setup in the first place.

Generate sex

>>101607187
>https://www.xvideos.com
Generated request.

>>101606525
https://files.catbox.moe/mdjnjr.png

>>101607221
thanks

File: 00051-2026678872.png (1.31 MB, 1112x1248)

1.31 MB PNG

>>101606841
just hair covering eyes/hair over eyes. It didn't work 100% of the time with that lora/model.

File: file.jpg (270 KB, 1792x1024)

270 KB JPG

>>101607134
this one is everything from leonardo, it's 303m now and still going, covered the last 140 days so far. there's already a set up with 274m from playground, 4m from civit and i have 7m from pixai waiting to be normalized and released
i havent looked at the content much, its probably a lot of trash, but thats kinda the point, why are some images trash, what are the best settings and so on, also images can be recaptioned, rated etc and these saas have user ratings already which can help start off the filtering

File: exampless.jpg (94 KB, 742x393)

94 KB JPG

>>101607221
Thanks, I have bunch but I almost never look at them lmao.
Comfy, Ani and Debo has schooled me and I own my own pitiful existence to them.

>>101607163
>>101607178
Thank you, let me set up a workflow and I'll post results in a bit.

>>101607262
No stress.

File: file.png (11 KB, 385x164)

11 KB PNG

there are lots of research avenues, for example pixai has their own nsfw detection, someone could see how effective it is
i think all of the saas have a "title" field, this should be a human description of the image, it would be cool to compare this to the prompt

>>101607281
How familiar are you with SQL?

>>101607290
yes. im using mongodb though. why

>>101607322
I am more interested about what do you think about user tracking done by Amazon for example? Have you ever analysed a session cookie.

File: asde3.jpg (149 KB, 1742x653)

149 KB JPG

>>101607263
So here's how I set up it now, one prompt for everything but only one character was seemingly made. Before when I used this, I had two different positive prompts and while it worked somewhat, the results weren't very good (but I really only tried like 3 quick gens).

>>101607403
>Comfy Couple? two positives? interesing

>>101607403
What the hell is this? Don't use custom nodes - it might seem easier for you but in the end you are going to need to learn it on your own.

>>101607425
>>101607431
Is attention couple not just a custom node? I used comfy couple cause it's just an (easier) fork of attention couple and the anon said he used attention couple. How would I go about setting it up with vanilla nodes?

>>101607403
I will do something brb

File: Screenshot 2024-07-28 at (...).png (142 KB, 631x482)

142 KB PNG

>>101607456
Not familiar with this node and use. looks interesting.

File: 00094-359715098-score_9, (...).png (1.47 MB, 1024x1536)

1.47 MB PNG

mikuslop a day

>>101607481
Which one?

>>101607403
https://litter.catbox.moe/vx3l7h.json

>>101607504
Use this and forget about gay downloaders.
Then when you are ready add more shit into this network. This is how you learn.

>>101607494
Comfy Coupl,e, it has two positive inputs nut the other node has only one exit conditioning, confused.

>>101607504
Wew god damn, I'll be sure to try to study this then. Thank you for this.

>>101607569
Anon in last thread seemed to have good results with it, though he said he used attention but it's kinda the same from what I can tell
>>101604891
>>101605239

>>101607581
Best solution is always the most simple, elegant.
Don't listen to various shit talkers, redditors etc because when you study these things on your own you will learn.
Accept no truth except the right one.

Just woke up from a 2 month coma is SD3 still gay?

>>101607630
You can't kick a dead horse.

is 1080x2520 (PORTRAIT) even possible? I can't find an online SD instance that goes up to more than 1024x1024 and obviously no "mods" to iteratively do it in chunks (which did not work on my machine with a1111 before it broke either anyway)

requesting that someone try to gen anything in this res, upscaling to it is acceptable. preferably hit me with an online link where I can gen it myself (free ofc)

>>101607684
GIve the positive prompt please.

File: Happy3.jpg (82 KB, 1368x768)

82 KB JPG

>https://www.youtube.com/watch?v=f-EfflMI6_4

File: 00127-1076843182.png (173 KB, 512x768)

173 KB PNG

>>101607707
its been a while....forgive how bad this is going to be, off the top of my head, for ultrawide portrait resolution, for phone papes frankly. (since none in this res exist outside the defaults that came with my phone from all my google searching)

black background, minimalism, icon, logo, particle effects, psychedelic, pattern, dark, neon, light beams

negative: people, human, (etcetera)

<-- pic tangentially related, its when I tried to make SD understand the mcguffin from Foundation without a lora.

>>101607773
I don't tolerate trolling, either do it or not and be done with it.

>>101607784
I can't do it myself a1111 broke for me years ago and I haven't set up comfy yet. also what fucking trolling.

>>101607784
https://i.imgur.com/HIw2UPE.jpeg

>>101607810
I'm on it.

>>101607812
https://www.youtube.com/watch?v=HKYYiYcr21Q
This is my song. Do you play guitar? If not... you cannot play this riff.

>>101607813
danke, I just want to know if its possible locally (but preferably online somehow for my purposes currently...guess ill keep looking for that) with a good gpu, which I lack sorely, mine is very mid.
i'm pretty sure I got close to 1080 desktop via upscaling before getting out of vram every time and then a1111 stopped working for me period after that, some bloat in the updates I suppose.
one day ill get it working again, or buy a new gpu.

>>101607853
Danke Schön, Ich bin aus Finland.
I thought you were some braindead yankee etc...

File: First.jpg (74 KB, 1368x768)

74 KB JPG

>>101607853
What is the resolution you want? I also have very slow gpu.

>>101607861
I get that a lot, prolly cause I have some currently undiagnosed mental health issues. i'm not german either I just like "danke" instead of thanks it seems more...genuine, to me.
>>101607888
1080 by 2520 PORTRAIT

>>101607901
Keine problem

>>101607888
*also pretty neat first for this type of gen I like it, it has that futuristic UI look like some kind of "jarvis"

File: final.png (1.04 MB, 1080x2520)

1.04 MB PNG

>>101607933

>>101607951
I know it is too middle but if it is a flyer... you will have text on top and bottom anyway.

File: ComfyUI_temp_hmgha_00002_.png (2.39 MB, 1536x864)

2.39 MB PNG

>>101607684
german anon here. got a free 3090 and some time to spare, just chilling. tell me what you want exactly and i'll gen a few takes for ya

>>101607933
There will be other versions soon but my gpu is really too slow.

>>101607504
So I studied this a bit and when I asked the question about having to do multiple prompts >>101607159, I was told no. But this example here literally shows that, unless I'm missing something. Am I nuts or something?

>>101607986
Use your own brain.

>>101607951
very nice, well this confirms it can be done, I worried the ultrawideness would push current limits, thanks for entertaining my questions for a bit, I have no way of genning this myself right now, need to try different UIs until something works I think
>>101607966
>>101607969
if you are willing to make more go nuts, wallhaven has zero wallpapers of this size and I don't even know where to start looking anymore, its been 4 days since I did so with no good results besides "do it yourself". I don't have any good prompt ideas besides the generic one I posted before right now, unless you want to swap "logo" with planets and space and stuff, its been over a year since I prompted so its like im getting back on a bicycle right now and the chain is coming off.

I should really focus on testing UIs right now until I can at least gen again

File: final3.png (2.06 MB, 1080x2520)

2.06 MB PNG

>>101608012
Gesundheit.

>>101608012
I'm sorry but I don't work for free. These tree examples are all what I can do.

Matteos regional conditioning workflow.
https://files.catbox.moe/8uio7b.json

>>101608031
https://www.youtube.com/watch?v=FRhYSrSQGtw
Ayyy.

File: final4.png (2.1 MB, 1080x2520)

2.1 MB PNG

Last one. mit Respekt.

All right now you want to ask about some fonts?

>>101608110
thanks again fren.
I found a comfy instance via google but I crashed it instantly trying that res so local with 4gb vram again it is.

File: ComfyUI_temp_yekev_00002_.png (2.55 MB, 1072x2512)

2.55 MB PNG

>>101608012
any resolution can be done basically. brute force upscaling this did use up about 18gbs of vram tho

>>101608156
I have the same. I don't want to upgrade because this is so dirty.
Do you want anything? Typography? I can do it while I'm drunk but the time window is small.

>>101608159
No it doesn't... generate a multiple and scale it up in photoshop.

File: ComfyUI_temp_nzing_00002_.png (3.77 MB, 1080x2520)

3.77 MB PNG

>>101608012
i fucked up the resolution on the first one, sorry lol

File: ComfyUI_temp_ayhpg_00001_.png (2.25 MB, 1080x2520)

2.25 MB PNG

>>101608224
eh? that one was ok, the golden one wasn't, is 4 off. anyways, I like this one:

>>101608254
Yeah I don't get paid by your opinions. You like what I serve or if you don't - go away.

>>101607630

Yes. Carry on.

>>101606140
anyone have a tutorial series recommendation to learn comfyui from scratch? all of the ones i've seen assume that you have some experience with ai image generation already

>>101608383
https://comfyanonymous.github.io/ComfyUI_examples/

>>101608468
She's rescuing me....

i miss schizo anon

>>101608517
Self-flagellation

>>101608254
>>101608517
https://www.youtube.com/watch?v=AwUyY2WGC00
Finnish song to celebrate summer.

File: 00030-3539361914.jpg (1.1 MB, 1560x2328)

1.1 MB JPG

morning.

File: My Heart.jpg (120 KB, 1368x768)

120 KB JPG

>>101608559

My 4090 sometimes gets choked on 2x hires fix for 1080x1080. Also 1.5x hires takes 40-90 seconds while 1080x1080 takes 10-16 seconds. I feel like I will need to upgrade to 5090

>>101608634
Sure you do, nigger.

>>101608559
good morning, i hope you are feeling well

File: SDXL_CyberRealisticPony_V(...).png (2.69 MB, 1248x1824)

2.69 MB PNG

>>101608634
yeah not that hard to OOM 24gigs of vram, thankfully things are a lot faster now with the tech provided (lighting, hyper, AYS, whatnot). I hope you are not on a1111 tho.

>>101608712
So this is the image you wasting that GPU power?

>>101608712
>yeah not that hard to OOM 24gigs of vram
I can't even picture this having pushed 4 to its absolute limit, can you give some examples of what would trigger a beeg OOM? one guess would be latent upscaling yeah?

File: SD1.5_picxReal_10.safeten(...).png (1.9 MB, 1024x1360)

1.9 MB PNG

>>101608559
good morning. you're just using the weight slider lora at -1 and probably add the zovya age embedding? lol
>>101608554
I like this one lol https://www.youtube.com/watch?v=Ota5eYeiK4o

>>101608761
https://www.youtube.com/watch?v=CENlVKSj-tw
Kiitos / Thank you for remembering.

>>101608761
uh, yeah

>>101608634
Rumors 5090 would be only 28gb which really suck.

File: SD1.5_cyberrealistic_v50.(...).png (1.78 MB, 1024x1360)

1.78 MB PNG

>>101608749
I remember OOMing when trying to resample an SDXL gen after an "upscale by model" node (and without downscaling it, erm), as in at 400% size, before I knew better. and doing a 2x highres fix on SDXL gens in a1111, for some reason, has used up all my vram too, sometimes. DAT/HAT upscalers are also dangerous.
anyways, hyper 12 step sd15 lora test, 12 steps, yay or nay?

What's the best way to upscale? Nothing beats 2x hires fix with remacri/nmkd upscaler, img2img is ok but its start to deform around 0.40-0.50 noise so images are not that crystal crisp.

>>101608913
You have learned the essential. If the result looks like shit then redo .

>>101608913
what i have been doing last couple days, is: a 1.25 upscale at .25 denoise, batch count 3. i then pick 1 of the 3 outputs, then repeat the process until the pic is big enough. also have the extra noise multiplier setting at .05

File: ComfyUI_temp_mihqb_00406_.png (1.41 MB, 1080x2560)

1.41 MB PNG

>>101607773
found this, having a poke

File: ComfyUI_temp_mihqb_00410_.png (3.64 MB, 1080x2520)

3.64 MB PNG

>>101608959

>>101608913
well didnt you just answer your own question? NMKD superscale is indeed a good one, remacri brightens image so not so hot for darker/nighttime shit. another good one, albeit more aggressive, would be ultramix balanced. step amount and sampler/cfg you use for the upscaling process. are also of importance.
DPMPP 2M/3M SDE, scheduler SGM uniform or exponential, can try AYS. lower stepcount, like 12, maybe 16. low cfg, denoise 0.3-0.4. bla
obviously also depends on checkpoint and what type of material you work with.
2 chicks I posted above are both NMKD superscale at 0.3 denoise with 12 steps, sgm uniform scheduler, dpmpp 2m sde

>>101608913
kohya hiresfix or hi diffusion

>>101608634

3090 anon. You might need to reinstall A1111 or use Forge.

>>101608913
can change default img2img upscaler model in anus1111, to whichever model you like, like nmkd, also, I forgot to mention controlnet tile guidance.
>>101609023
what is hi diffusion?

>>101609056
upgraded kohya hiresfix, there's a comfy node
https://github.com/blepping/comfyui_jankhidiffusion
and someone on /h/ was patching it into reforged last night but they had to go to bed

File: ComfyUI_temp_mihqb_00411_.jpg (2.24 MB, 1080x2520)

2.24 MB JPG

>>101608998

File: leosamsMoonfilm_filmGrain(...).png (1.99 MB, 1024x1536)

1.99 MB PNG

>>101609063
oo very interesting. thanks!
here have a korean schoolgirl, yes I get korean sd15 schoolgirls with the now defunct leosam model in june 2024.

What's the deal with fooocus? Is it faster than regular webui?

Fuck, someone deleted perfectly good vintage styled lora few hours after it was posted on Civ. I was in the mood to make 50s chars. Is there any way to see name of that user that made that?

>>101609293
that lora was a mistake. just let it go.

File: file.png (63 KB, 660x558)

63 KB PNG

>>101609293

 {'name': 'Vintage style - XL (testing)',
  'creator': {'username': 'unfazedanomaly964'}},

was it this

>>101609078
>>101608998
>>101608959
very nice
i've been trying to come up with a better prompt idea for this while sorting out ringtones (I went with the mgs codec) but honestly black background with something abstract really does work best for phones, so the icons don't drown in the detail in the background
i'll let you guys know if I think of something

>>101609345
Yes. Any chance it's possible to get that deleted lora?

File: 00017-757296536.jpg (316 KB, 1288x1456)

316 KB JPG

Bread

>>101609368
not from me, sorry, it's deleted, my downloader is stopped atm, only getting metadata. it had 13+ downloads though maybe you'll get lucky and someone has it

>>101609431
Alright, thanks for trying.

>>101609401
>*coughts on your fish*
dinners ready

File: 00084-1961941168.jpg (951 KB, 1944x2576)

951 KB JPG

cought's on youre fish

>>101609401
is this real?

File: 00102-1057598377.jpg (956 KB, 1944x2576)

956 KB JPG

teach a man to fish

>>101609870
are you the guy that does the twitter/facebook entry posts?

>take long break from prompting
>return to prompting
>feet just work now, with basically no effort
It's a whole new world

>>101610043
>feet just work
they don't though

>>101610059
Seems to on my end, you get the occasional sixth toe but generally they look pretty much spot on, miles better than last time I prompted.

File: 1706028986676283.webm (3.91 MB, 848x508)

3.91 MB WEBM

>>101609763
yes

>>101610069
are they good enough by a footfag's standards, i doubt. i am glad i am a thigh connoisseur, thighs are so easy to gen

>>101610094
I'd say they're pretty good yeah. I'm happy with them.

>>101610069
are you using SDXL?
I still use maily 1.5 since I don't feel that SXDL has caught up yet. I mean sure Images can be higher quality but they are janky sometimes and a lot of the training data from S.D. 1.x has been omitted.

>>101610133
Nah not using XL

File: 00006.png (2.55 MB, 1280x1920)

2.55 MB PNG

>>101610043
based
had trouble with it when I tried at the time but some results. I need better hardware.

>>101610177
Good lord, my results look nothing like that, thank God

File: 09179-1811777377.png (1.34 MB, 1024x1280)

1.34 MB PNG

>>101610152
post some feet

>>101610071
is this real???

>>101610133
model censorship has set us back years from where we should be now, thanks US congress.

File: file.jpg (301 KB, 1024x1792)

301 KB JPG

>>101609439
ive started the downloader again. working through latest to oldest loras. metadata is already scheduled for every 3 hours, might make it hourly to be sure nothing is missed. downloader will pick up on anything new added while its in progress
will set it up for other types as well
only grabbing the file marked as primary, looks like most only have 1 version anyway and checkpoints with more than one is just a different dtype or theyve uploaded the vae or text encoder separately as well as being included in the checkpoint itself, ill change that if someone shows me an example with multiple versions that are actually useful
probably dump everything on hf at some point

File: 1704387187879152.webm (923 KB, 480x854)

923 KB WEBM

>>101610214
Yes

File: 00072-1170786755.png (1.39 MB, 1024x1536)

1.39 MB PNG

>>101610198
Just one, it feels kinda lewd

>>101610221
There has been some good underground colabs though. But yeah things could be better I guess. From a coomer perspective.
What we need is a DALL-E leak and a way to merge different base models.

>>101610238
is that Will Smith?

>>101610239
That is kinda lewd. Very nice though.

File: 00134-807328759.jpg (963 KB, 1944x2576)

963 KB JPG

me on right
>>101610266
maybe the chicoms are our best hope

File: 1701779365689048.webm (297 KB, 720x996)

297 KB WEBM

>>101610281
Yes

File: 01761-peanuts comic strip(...).jpg (133 KB, 984x984)

133 KB JPG

Morning anons

>>101610323
I think my life has definitely taken a turn for the worse ever since stable diffusion released.
I mean I've always lead a cloistered existence.
But it feels worse now somehow...

>>101610228
Oh awesome, thanks. What's your username to look up?

>>101610468
I was always an underachiever, and since Stable Diffusion, I feel vindicated.
Rather than attempt anything I just need to wait until some autist figures out how to digitize consciousness and send us into our own personal universe that we can shape to our desires at will.

File: 1691525909174445.webm (3.82 MB, 960x540)

3.82 MB WEBM

>>101610503
same

File: 1696242259746436.webm (1.73 MB, 720x720)

1.73 MB WEBM

>>101610468
I just want the world to be a better place and seeing this stupid shit makes me think we might just make it.

>>101610503
I guess this kind of encapsulates it. I mean all that effort. Simply negated.
It's not perfect yet. But soon it will be.

>>101610523
meme2vid is my absolute favorite AI genre

>>101610525
toenails look kind of wierd. Who's the broad though?

Thinking of trying out ComfyUI. Anyone know if it's relatively self-contained? Automatic has already shat things all over my C drive, and I'm low on space.

>>101610614
there is a portable install option, which means it doesn't install jackall and can run off a disc like how I ran emulators at college that one time out of sheer boredom, so you're good.

>>101610614
not sure but it uses gradio. and the shitting all over your hard drive is a gradio issue.

File: 08799-3477633176.png (979 KB, 1024x1024)

979 KB PNG

Tits and ass...
>It gets me every time

File: may2feb.webm (3.81 MB, 576x1024)

3.81 MB WEBM

>>101610523
>>101610547
>we might just make it.
The rate of progress seems pretty decent
Y'know, it's okay

>>101610715
ironically the older tech AI is better tho for memes, maybe it's also that people use newer AI to simply produce normal shit instead of using it for things that cannot be done in reality

File: 00154-1262232559.jpg (1.6 MB, 1944x2576)

1.6 MB JPG

kinda feel like singing

>>101610715
>1 year
>will smith is now white
what a difference a year makes

top_indices = ops.cast()(ops.arange(top, top + height, 1)(), "int64")
left_indices = ops.cast()(ops.arange(left, left + width, 1)(), "int64")
spatial_pos_embed = ops.index_select(dim=1)(spatial_pos_embed, top_indices)
spatial_pos_embed = ops.index_select(dim=2)(spatial_pos_embed, left_indices)

figured out a workaround for dynamic_slice not supporting intvar, nice. needed it for jointattention as well. should probably make arange able to directly return the required type though, its partly implemented already
in the first case it also needed a workaround for dim name deduplication, arange codegen needs the named dim to calculate the symbolic value but the input dim name for height/width is deduplicated because output uses those dims, so i just name the input dims with the output name
couple other issues fixed too

<aitemplate.backend.cuda.builder_cmake> Executing msbuild "tmp/stable-diffusion-3/build/stable-diffusion-3.sln" -m /property:Configuration=Release
<aitemplate.compiler.compiler> compiled the final .so file 

>>101610757
You'd still be able to do quality memes if this was local but since it's slow and online and limited you can't really experiment.

File: 09443-340408789.png (1.19 MB, 1024x1024)

1.19 MB PNG

you can make your own songs on
>https://suno.com/

File: file.jpg (346 KB, 1792x1024)

346 KB JPG

>>101610469
datasets go here https://huggingface.co/bigdata-pw
if the datasets get likes i might upload the loras sooner rather than later lol

What model, LORAs, etc. do you guys recommend for fantasy artworks? Making some character portraits for a TTRPG campaign. 12 GB VRAM.

>>101610991
>login, internet 2.0 accounts only no email field, or phone number
cool but no way fag.

>>101611035
autism

>>101611041
sure, still not doxxing myself to have a machine write a song, ill do it myself with my guitar. until you realize why i'm not doing this.

[Composition and Style
>Replace the blue background with a vast cosmic void, complete with twinkling stars, colorful nebulae, galaxies, and stardust trails.
>Retain the logo's original shape and center it in the composition.
>Apply a 3D metallic effect to the logo with sharp edges and a kinetic outline.
Lighting and Colors
>Illuminate the logo with ethereal starlight reflections on its metallic surface.
>Ensure the cosmic background is visible through any areas where the blue background was replaced.
>Use a palette of vibrant green, metallic silver for the logo, and a diverse range of colors for the cosmic background.
Quality and Details
>Generate a high-resolution image with detailed textures and smooth gradients.
>Focus on completely replacing the blue background with the cosmic environment, creating a seamless integration of the logo into the scene.
Constraints
>Do not preserve any blue areas from the original image.
>Do not alter the logo's shape or add text.
>Do not include additional objects in the scene]

This prompt is good or not? I get nothing remotely usable using StableDiffusion, SDXL-turbo, controlnet with cannySDXL, and FreeU
And the thing is I want something abstract

>>101611051
we know why

>>101611062
dont you have a war to start with iran or something glowie? I said im not using your service if it requires a login and cant use throwaways, period. cry more.

>>101611115
https://voca.ro/1jIWAdkQFrkK

>>101611154
thanks for proving you're a glowie with your memetic hazard tier poem or whatever, if I have to ask another person on the internet today to kindly fuck off im going to find you irl.

>>101611165
internet screw ball

File: 00178-31368371.jpg (1.27 MB, 2576x1944)

1.27 MB JPG

https://voca.ro/11dVolUjLmJK
finna invade eye ran

i think trani is shit (as a human)

>>101611344
you are valuable and you matter friend

File: 00218-2784113712.jpg (983 KB, 1560x2328)

983 KB JPG

carrying, or burying, the thread

File: E2N42Br.png (331 KB, 962x665)

331 KB PNG

are those normal speeds for a gtx 1070?
things i did for optimization:
>--medvram-sdxl --xformers command line arguments
>disabled image preview
>disabled system fallback in nvidia control panel
>disabled hardware accelerated gpu scheduling
>selected xformers in cross attention optimization in A1111 optimization settings
its still feels too slow for a single image with nothing enabled.
will moving to another webui give me better performance or is it simply the peak of what a gtx1070 is capable of?

# du -hs models/
2.3T    models/

how big is your models folder

File: _0618d49d-caa6-41d1-832c-(...).jpg (255 KB, 1024x1024)

255 KB JPG

could someone help me by making a class selection screen that looks good? I've been using bing ai and don't like how it looks.
I've been using these prompts : class selection screen, portrait, archer, knight, mage, inventor, no text, color theory, weapon in one hand, no extra weapons, faceless, helmets

File: file.jpg (236 KB, 1024x1024)

236 KB JPG

>>101612402

>>101612402
well dont try to do the whole thing at once. generate the characters separately and generate the portrait UI, copy and paste the portrait UI multiple times in photoshop or some image editor.

File: file.png (2.16 MB, 1373x691)

2.16 MB PNG

>>101612402
dall-e 3 just prompts a bit different
>rpg game class selection screen with classes archer, knight, mage and inventor
>rpg game class selection screen with classes archer, knight, mage and inventor, they are wearing helmets and holding a relevant weapon
>>101612521
this too, if you're trying to make a game or something you'll have to put a bit more effort in

>>101612542
it makes me so mad when I see how good dalle is at listening to prompt

>429
why do they even bother
i can always register more accounts and get new ips
the only way to stop it is either shutting down the service or publishing the data in bulk themselves so i dont have to do it for them

>>101612812
considering dalle is just a footnote now and nobody seriously uses it anymore I don't really care

>>101612812
Imagine how good it would be if it didn't have 50 layers of censorship built in that ruin even the most innocent of prompts.

Why does Burgeristan of all countries have to be the country with all the GPUs…

What's the verdict on SD3?
Shit? Usable now that people have had the time to mix it with other models?

Is the latest pony still the model to use?

>>101612812
data and reinforcement is all you need desu

if I get a second graphics card, can I crank out images twice as fast? ...or two at once?

>>101612871
>What's the verdict on SD3
still waiting for the actual release

>Is the latest pony still the model to use
sadly yes. nobody is bothering to do a big porn tune until everyone can decide on a good base model

a1111 got updated 2 days ago, says it has performance improvements, is it faster than forge?

File: 00482-TFT_1323.png (3 MB, 1536x2560)

3 MB PNG

Runescape bunny

File: file.png (403 KB, 755x561)

403 KB PNG

>>101612830
>shutting down the service
lol
hurry up and come back online, i was in the middle of automating registration
>>101613321
there's some thing that dumps models and everything from the runescape cache, would probably make a nice dataset

File: 81yrjsiUQfL.jpg (437 KB, 2560x2546)

437 KB JPG

Lurker here, with a serious question. Don't you guys get bored of all this crap after a while?
The novelty of image generation has worn off for me after like 30 minutes. Yet this general continues to exist.

>>101613564
>30 minutes
npc moment

>>101613564
Are you talking about Bing? Yeah, that gets boring quickly, but with SD you can make tons of specific fetish porn

>my brain is too small to understand that people have different interests than me
I'll assume you're trolling instead of being 20 iq

>>101610991
Udio is much ebtter

File: 01762-Adorable Quokka ori(...).jpg (93 KB, 1446x964)

93 KB JPG

Well online stuff like DALL-E 3 gets pretty tame and boring after 30, 10 mintues even but open source local stuff just has so many possibilities to explore and it's just endless, no real restriction to what you can do but also no real instructions to get what you want so it's just really fun for me and it can be pretty fun in general if you have the hardware and the patience.
>Verification not required

>>101613707 meant for >>101613564

>>101613564
its like porn fentanyl to gooner porn addicts
t.gooner porn addict

File: 00569-TFT_1330.png (3.62 MB, 1536x2560)

3.62 MB PNG

uploaded the lora for this
https://civitai.com/models/607769?modelVersionId=679378

>schizophrenic wasteland

Considering that Forge does a lot of optimizations, if I stop using Forge and switch to A1111, will I get better hands? I'm tired of fighting so hard to get good hands. I'm autistic, and I can't cope with a gen that has bad hands.

>>101613564
No I could do this shit forever. Just have a naturally creative mind bro. However I did stop since it felt like I was spending too much time on it instead of other things, and I'm willing to wait until the tech becomes even better.

ok there we go, just need to put it all together
mess with the best die like the rest desu

So I can confirm that the claim SD3 is "untrainable" is utter baseless horseshit, in the past couple of days I've trained two SD3 Loras just as tests over on TensorArt, both teaching it NSFW concepts of the sort that almost certainly would never ever appear in the base model dataset at all and could only be posted over in the /b/ "degen AI" thread. Both work. It learns the shit same way an SDXL Lora would. So what am I missing here bros? Why are people so insistent that you can't teach it stuff?

>The root cause is the sheer volume of activity: we're receiving hundreds of thousands of image uploads daily, and there are constant metric updates every second. For reference, we now get more images in a month than we did in the entirety of 2023
most of those images are trash on model pages, there's still barely any actual posts, leonardo do the the volume of the entire site every 2-3 days, maybe limit the amount of trash that can be uploaded for each model. from what ive seen here those images are never really reproduceable or accurate examples or anything anyway. and maybe just don't update the metrics every second

>>101614977
bots are farming buzz but they don't care about bots at all

File: 00264-2717956083.jpg (733 KB, 1944x2576)

733 KB JPG

sigh, trying to gen nongirl pics for a bitsu
civitai seems broke today

>>101614999
i mean, they do, they reduced the number of buzz because they're losing money. they're just incompetent so can't stop it. as an example of their incompetence, they still have sourcemaps in production, i unpacked it and put the site source on github already, granted nobody really looks at my stuff so nobody has noticed but still

>>101614851
> So what am I missing here bros? Why are people so insistent that you can't teach it stuff?
https://stability.ai/news/license-update

>>101614977
More importantly, will my RTX 4060 with an incredible 8GB be able to run it at a viable speed or I'm stuck with SDXL?

File: file.png (12 KB, 367x303)

12 KB PNG

they could reduce volume by 10% by enforcing metadata on image uploads

File: 1715138780740309.jpg (78 KB, 1024x1024)

78 KB JPG

>pc heavily freezes near image generating completion or when loading new loras
what could cause this?

File: file.png (1.52 MB, 1920x1040)

1.52 MB PNG

350m from ~162 days and they have no site issues lol
the images are going straight to r2 how is it even an issue

>>101606140
What's the best way to merge checkpoints if I want a sfw model to have nsfw capabilities?
Is there a certain way or order to merge models to improve the results of what I'm trying to do?
Like weight merging really small amounts (0.5 - 1) of several other nsfw models and then do a final merge of 50% with my original model, or use my original sfw model as a start point, then merge small amounts of tons of nsfw models to it.
Maybe I should merge loras into it? Like, if I merge several nsfw loras into it, will it know how to generate nsfw stuff?
How do I use the "add difference" properly?
Any tips?

File: file.jpg (295 KB, 1024x1792)

295 KB JPG

i sleep
reminder to hmu if you have any dataset requests

>>101615476
what's his power level?

File: 01361.jpg (613 KB, 2496x3648)

613 KB JPG

Evening /sdg/! Hope the GENs have been goooood!

>>101615386
I guess prompt injection straight into blocks might be a little faster for mapping the NSFW bits of the models

File: わいよ（意味不明）.jpg (262 KB, 1536x1536)

262 KB JPG

Bicycles sure are too detailed for models to draw correctly, huh

>>101615656
Is there any tutorial or doc I can check on how to do that?

File: 01355.jpg (569 KB, 2496x3648)

569 KB JPG

>>101615715
That's me and motorcycles as well. Or even trying to make the damn thing a future/cyberpunk motorcycle and it doesn't budge. Majority of my bike GENs are always scrapped.

Yours came out not TOO bad though!

>>101615766
Ah, I know that pain, too. I made a few motorcycle gens too but damn, the details are fucked.

File: 01317.jpg (500 KB, 1872x2736)

500 KB JPG

>>101615787
Yeah, it's why I never try to do anything with vehicles. OR well I do it VERY rarely.

Like this didn't come out bad. But bikes? Fuck off, lolol

File: 038UnOAKEl4pnRtFnldh--1--(...).jpg (365 KB, 2048x2048)

365 KB JPG

hlky gj
"Terminal Icon, Git , App Icon, RGB, ASCII"
https://suno.com/song/1f7e662f-f740-4607-b2ef-7f450e4bb003
>VXST

File: ge7J5nlE1IDnOP8YUaKS--1--zcgeo.jpg (129 KB, 1024x1024)

129 KB JPG

>(weight:1) dead animal guts intestine dog bear cat dog bear cat dog bear cat dog bear cat dog bear cat dog bear cat
>TNST
nat test pass

File: cXlheqc2WOYXNUYnTW7b--1--(...).jpg (602 KB, 2048x2048)

602 KB JPG

https://suno.com/song/638576ff-985c-4729-9051-a678ac0ce418
>STTA
sa taiga tunes

Where is debo?

>>101615189
Memory leaks?

>>101616142
he got brain damage from me posting suno and hasn't been back all weekend, I hope he's recovered soon enough
>2T4A
did anyone else experience a rise in the price of tictacs locally?

>>101616168
Was he not posting yesterday?

>>101616183
I was looking for him so I could post about my electroblog like it asked me to but I didn't see him posting when I was here
>Y0WT
yo

>>101616155
call your local plumber

>>101615062
So you're saying they're lying on purpose about technical things because they care about the license a ton for some reason?

>>101616230
I guess this doesn't matter anyways cause now I can just post Loras on CivitAI (at least the ones within their TOS) and be like "yeah it works fine, you're retarded for thinking it wouldn't" since it was unbanned there

File: Adorable Quokka, Cheese.jpg (98 KB, 1446x964)

98 KB JPG

Filling

>>101616232
>>101616232
>>101616232

File: Felt puppet cheese Adorab(...).jpg (78 KB, 1446x964)

78 KB JPG

That cheese looks oddly suspicious to a lemon