[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip / qa] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/g/ - Technology


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


File: 1722772237934974.jpg (748 KB, 1920x1440)
748 KB
748 KB JPG
Discussion of free and open source text-to-image models

Previous /ldg/ bread : >>101715949

>Beginner UI
EasyDiffusion: https://easydiffusion.github.io
Fooocus: https://github.com/lllyasviel/fooocus
Metastable: https://metastable.studio

>Advanced UI
Automatic1111: https://github.com/automatic1111/stable-diffusion-webui
ComfyUI: https://github.com/comfyanonymous/ComfyUI
InvokeAI: https://github.com/invoke-ai/InvokeAI
SD.Next: https://github.com/vladmandic/automatic
SwarmUI: https://github.com/mcmonkeyprojects/SwarmUI

>Use a VAE if your images look washed out
https://rentry.org/sdvae

>Model Ranking
https://imgsys.org/rankings

>Models, LoRAs & training
https://civitai.com
https://huggingface.co
https://aitracker.art
https://github.com/Nerogar/OneTrainer
https://github.com/derrian-distro/LoRA_Easy_Training_Scripts

>Flux
https://huggingface.co/spaces/black-forest-labs/FLUX.1-schnell
https://comfyanonymous.github.io/ComfyUI_examples/flux

>Pixart Sigma & Hunyuan DIT
https://huggingface.co/spaces/PixArt-alpha/PixArt-Sigma
https://huggingface.co/spaces/Tencent-Hunyuan/HunyuanDiT
https://huggingface.co/comfyanonymous/hunyuan_dit_comfyui
Nodes: https://github.com/city96/ComfyUI_ExtraModels

>Kolors
https://gokaygokay-kolors.hf.space
Nodes: https://github.com/kijai/ComfyUI-KwaiKolorsWrapper

>AuraFlow
https://fal.ai/models/fal-ai/aura-flow
https://huggingface.co/fal/AuraFlows

>Index of guides and other tools
https://rentry.org/sdg-link
https://rentry.org/rentrysd

>GPU performance
https://vladmandic.github.io/sd-extension-system-info/pages/benchmark.html
https://docs.getgrist.com/3mjouqRSdkBY/sdperformance

>Try online without registration
txt2img: https://www.mage.space
img2img: https://huggingface.co/spaces/huggingface/diffuse-the-rest
sd3: https://huggingface.co/spaces/stabilityai/stable-diffusion-3-medium

>Related boards
>>>/h/hdg
>>>/e/edg
>>>/d/ddg
>>>/b/degen
>>>/vt/vtai
>>>/aco/sdg
>>>/trash/sdg
>>
File: ComfyUI_Flux_2331.jpg (213 KB, 1024x1024)
213 KB
213 KB JPG
>>
File: FD_00039_.png (1.64 MB, 1024x1024)
1.64 MB
1.64 MB PNG
>>
>>101720035
the onahole looks washed out and deep fried
>>
File: file.png (1.08 MB, 1024x1024)
1.08 MB
1.08 MB PNG
>>
File: ComfyUI_Flux_2347.jpg (178 KB, 1024x1024)
178 KB
178 KB JPG
>>101720035
>>
File: flux_00025_.png (1.17 MB, 1024x1024)
1.17 MB
1.17 MB PNG
>>
>>101720100
Crispy like it should be
>>
File: file.png (907 KB, 1280x896)
907 KB
907 KB PNG
>>
File: ComfyUI_01721_.png (1.06 MB, 1152x768)
1.06 MB
1.06 MB PNG
>>
Blessed thread of frenship
>>
File: Sigma_12340_.jpg (2.7 MB, 2048x2048)
2.7 MB
2.7 MB JPG
>>
Bigma status?
>>
>>101721012
September release
>>
bigma balls lmao
>>
I love being able to prompt scenes where it looks like something is happening rather than a picture of a person standing somewhere. It's still a lot of work to get the AI to understand the prompt but so much more is possible.
>>
File: red-panda.png (1.37 MB, 2048x1024)
1.37 MB
1.37 MB PNG
I noticed whenever I prompt for a stark white background + an object with Flux dev, the result is blurry or out of focus. Pic related, left is with "stark white background" and right is without. Have tried with a few prompts

Increasing the steps doesn't seem to help but once I take the background out of the prompt, the result is pretty much perfect. My usecase is making icons from the gens or 3D-models with Stable Fast 3D, both needing a transparent background.
>>
File: ComfyUI_00200_.png (1.09 MB, 1024x1024)
1.09 MB
1.09 MB PNG
>>
>>101717687
Where's the Miku in historical paintings?
>>
>>101721273
have you tried (focused) or (in focus) with prompt?
>>
File: 00001-2554369867.jpg (779 KB, 1814x2419)
779 KB
779 KB JPG
>>
>>101721298
The blurry result was with "sharp focus" in the prompt. Also tried with your suggestion with the parenthesis for emphasis but no dice, still blurry.
>>
>>101721360
I guess it gets easier once negative prompt works
>>
What context size flux has in the prompt?
>>
File: Wanwan.jpg (1.01 MB, 1664x2432)
1.01 MB
1.01 MB JPG
>>
File: ComfyUI_Flux_69.png (1.24 MB, 1280x720)
1.24 MB
1.24 MB PNG
>>101718764
Nta but bumping the question. My rtx 2080 is too slow to do autistic comparisons of every combination with various different prompts, currently just using euler/normal with 2.5 guidance. I did try deis with ddim uniform like >>101719696 said but the results look oversharpened and crispy
>>
File: 2024-08-04_00568_.png (1.91 MB, 1280x1280)
1.91 MB
1.91 MB PNG
>>101721396
since there is no documentation it ain't known, but you can guess the CLIP part is still 75 tokens, but the txxl5+ will be at least 256+ tokens
>>
File: ComfyUI_00205_.png (912 KB, 1024x1024)
912 KB
912 KB PNG
>>
official pixart bigma and hunyuan finetune waiting room
>>
>>101721396
512 for dev, 256 for schnell
>>
File: 00003-490653156.jpg (794 KB, 1814x2419)
794 KB
794 KB JPG
>>
File: reassembly_001.png (616 KB, 555x555)
616 KB
616 KB PNG
i drew this (with my finger in artrage). can models do this?
>>
>>101721699
Is this Loss?
>>
Retard here, is it over for VRAMlets for real now or is there a chance of a slimmed down Flux variant that generates at a lower resolution or something but is still a significant improvement over XL/Pony?
>>
>>101721749
flux runs in 8gb
>>
>>101721749
Nigger, wait and there will be better small models, the original LDM model was colossal and a piece of crap compare to models now
>>
File: ComfyUI_00033_.png (1.02 MB, 832x1216)
1.02 MB
1.02 MB PNG
looks like offensive posters are back on the menu boys
>>
>>101721765
kek, it runs on cpu if you are that brave, you need at least 12 GB to run the fp8 properly
>>
File: 0.jpg (230 KB, 1024x1024)
230 KB
230 KB JPG
>>
File: file.png (2.44 MB, 1024x1024)
2.44 MB
2.44 MB PNG
>>
>>101721860
Nice
>>
File: ComfyUI_00036_.png (1.25 MB, 832x1216)
1.25 MB
1.25 MB PNG
>>
>>101721711
Good thinking, I think our sermon did center around loss! Nehemiah.
>>
>>101721960
>religious
>terrible finger paintings
anon, are you 12 or retarded?
>>
File: uid.png (324 KB, 493x318)
324 KB
324 KB PNG
>>101721938
>>
File: ComfyUI_30684_.png (2.3 MB, 1536x1536)
2.3 MB
2.3 MB PNG
>>
>>101722004
>Safety
Libtards are really going to town with this word. How long until it turns into a meaningless buzzword and we can no longer describe the thing that tells you not to stick your hand in a lathe?
>>
>>101722033
that's the idea
>>
Wait what’s the latest
>>
>>101722072
flux is trainable but no releases. The guy who said it's not trainable now declares it trainable after seeing what Ostris did. Thanks Ostris. You da real mvp
>>
>>101722122
can I use up all my saved buzz to train a lora on civits hardware
>>
>>101722137
Not yet
>>
>>101722122
any set of weights are trainable, but does it train good?
>>
File: 00006-490653156.jpg (797 KB, 1814x2419)
797 KB
797 KB JPG
>>
File: file.png (2.35 MB, 1024x1024)
2.35 MB
2.35 MB PNG
>>
File: FLUX__00047_.png (1.16 MB, 896x1152)
1.16 MB
1.16 MB PNG
>>
File: ComfyUI_30686_.png (1.31 MB, 1024x1024)
1.31 MB
1.31 MB PNG
>>
File: ComfyUI_00044_.png (774 KB, 832x1216)
774 KB
774 KB PNG
>>101722033
I HATE SAFETY! I HATE SAFETY! I HATE SAFETY! I HATE SAFETY! I HATE SAFETY! I HATE SAFETY! I HATE SAFETY! I HATE SAFETY! I HATE SAFETY! I HATE SAFETY! I HATE SAFETY! I HATE SAFETY! I HATE SAFETY! I HATE SAFETY! I HATE SAFETY! I HATE SAFETY! I HATE SAFETY! I HATE SAFETY! I HATE SAFETY! I HATE SAFETY! I HATE SAFETY!
>>
>>101722230
generated for BBC
>>
>>101721997
God is obviously real and even if he isn't Dawkins is right that you have to pretend he is
>>
>>101722257
you gotta do 2girls1cup now
>>
>>101722263
my dick is obviously real, will you worship it?
>>
>>101722270
Depends if it's worthy. BBC?
>>
File: 2024-08-04_00401_.png (1.93 MB, 1280x1280)
1.93 MB
1.93 MB PNG
>>101722249
good movie, watched it to many times tho
>>
>>101722278
AWC
>>
>>101722279
yeah I'm burned out on it too
>>
>>101720054
let me goblinfuck her
>>
>>101721997
I wasn't asking for your opinion of my fine 10/10 art, but if ai gan genwrate similar. I will try, but I don't know what to proompt
>>
>>101722263
God has to pretend he might save Dawkins from hell. his burden is greater.
>>
>>101722302
>rotted pussy
would need to use a sword to cut it open again before fucking her
>>
>>101721194
Definitely!

>>101721273
"Gradient background" has been good to me. But if you're looking for a solid background to remove in post then try "no background"

>>101721546
Catbox please?
>>
>>101722264
for me, ai is about walling out the world, bot bringing in the filch

Weird how Myst is real life, just almost.
>>
too much /pol/ here
>>
>>101722408
I wonder why. Oh yeah.
>>
File: ComfyUI_Flux_2495.jpg (171 KB, 1024x768)
171 KB
171 KB JPG
>>
>>101722429
c-cute
>>
File: ComfyUI_00048_.png (1.11 MB, 832x1216)
1.11 MB
1.11 MB PNG
>>101722264
>>
File: 2024-08-04_00595_.png (2 MB, 1280x1280)
2 MB
2 MB PNG
>>101722355
>Catbox please?
sure
>https://files.catbox.moe/cn8jx5.png
she started a peaceful career as keyboarder now tho
>>
>>101722421
the amerimutts are awake and ruining ai
>>
File: fluxfun.jpg (2.31 MB, 4060x2045)
2.31 MB
2.31 MB JPG
fun
>>
Flux would be more fun if it knew words like "downblouse"
>>
File: ComfyUI_00195_.png (1.04 MB, 1024x1024)
1.04 MB
1.04 MB PNG
>>
>>101722550
laughing like a spastic at this
I'm very ashamed
>>
File: _00890_.png (510 KB, 600x904)
510 KB
510 KB PNG
>>
>>101722504
>Save Europeans from communism, atheism and degeneracy
>Now they want to be marxists, atheists and degenerates and invite muslims and blacks to rape their wives and integrate them with black babies
Enjoy your atheist shithole while you jerk off to being blacked
>>
>>101722550
>>101722562
Same, I kek'd hard
>>
>>101722550
>>101722562
>>101722570
NOOO THIS IS UNSAFE
>>
>>101722550
lol'd hard
>>
File: 00008-490653156.jpg (352 KB, 1728x2304)
352 KB
352 KB JPG
>>
File: meds.jpg (294 KB, 1024x1024)
294 KB
294 KB JPG
>>
>>
>>101722662
what a weird room
>>
>>101722674
man with itchy asshole lives there
>>
File: __00891_.png (508 KB, 600x904)
508 KB
508 KB PNG
>>
File: 2024-08-04_00619_.png (1.57 MB, 1280x1280)
1.57 MB
1.57 MB PNG
>>
File: ComfyUI_00053_.png (1.1 MB, 832x1216)
1.1 MB
1.1 MB PNG
>>
How do I make a little script to run the starter thing? It's kinda annoying to pyenv shell 3.10.6 then start.py and turn the pyenv shell back to system when i'm done.
>>
File: Designer (5).jpg (247 KB, 1024x1792)
247 KB
247 KB JPG
>>101720035
>thread pic
Tourist here: what happened?
>>101720085
Nice kitty.
>>
i stopped following the technical details of generation for a year and now I have no fucking clue what any of this shit is
>>
File: 2024-08-04_00620_.png (1.57 MB, 1280x1280)
1.57 MB
1.57 MB PNG
>>101722747
better bigger model from ex stable diffusion makers (now black forest labs) was released that btfo SAI and stable diffusion, also SAI is failing as corporation as we talk
>>
File: ComfyUI_00055_.png (896 KB, 1024x1024)
896 KB
896 KB PNG
hmm, yes, flux is good model
>>
File: maam.png (946 KB, 1024x1024)
946 KB
946 KB PNG
>>
File: FLUX__00121_.png (1003 KB, 1024x1024)
1003 KB
1003 KB PNG
what's a presidential podium called, the thing where the prez gives press conferences and stuff to the media. It's not the oval office, I think?
>>
>>101722830
>podium
>>
File: 00009-490653156.jpg (285 KB, 1728x2304)
285 KB
285 KB JPG
>>101722830
speaking podium, lectern?
>>
>>101722830
white house press room podium
>>
>>101722828
fucking lol
>>
File: 2024-08-04_00649_.png (1.35 MB, 1280x1280)
1.35 MB
1.35 MB PNG
flux can produce some cool atmospheric shit when you push it in the right direction
>>
File: FLUX__00122_.png (868 KB, 1024x1024)
868 KB
868 KB PNG
>>101722872
thanks fella
>>
Does the huggingface link for flux from the OP work with mobile? I want a friend to try but I'm on desktop & can't test it
>>
>>101722913
no u
>>
>>101722926
If you open chrome developer tools you can resize your screen to be a phone
>>
File: ComfyUI_Flux_78.png (1.55 MB, 1280x720)
1.55 MB
1.55 MB PNG
Anyone else experience this weird grain? Using euler + normal scheduler, maybe it because of 2.5 guidance
>>
>>101722926
I grabbed my phone and tested it. It works!
>>
>>101722981
>maybe it because of 2.5 guidanc
yes it is
>>
File: file.png (191 KB, 800x800)
191 KB
191 KB PNG
>>101720035
Can it generate moonman though?
>>
File: FLUX__00124_.png (1.14 MB, 1024x1024)
1.14 MB
1.14 MB PNG
>>
File: 2024-08-04_00659_.png (1.46 MB, 1280x1280)
1.46 MB
1.46 MB PNG
>>
>>101722909
Bur can it generate old people enjoying themselves in the sauna? https://files.catbox.moe/6mhfrj.jpeg
>>
>>101722954
That might change the page behavior, yes, but there are hardware, os, and other software considerations (and my user-agent would still be desktop if it matters)

>>101722983
Thanks, you are awesome!
>>
File: 2024-08-04_00660_.png (1.68 MB, 1280x1280)
1.68 MB
1.68 MB PNG
>>101723011
aslong their belly is hiding their genitalia like in that case, sure I guess
>>
>>101720035
what's about this flux I'm hearing you guys talking about, since it's here is local, I saw talks that 8gb vramlets can run it, but can it do nsfw? can it do 1girl? 1children?
>>
>>101723033
It's a gradio interface dude get over yourself
>>
>>101723049
>8gb vramlets
if you got half a day for a gen yes
>>101723049
>but can it do nsfw?
no, soft erotica to a certain point, no genitalia, no hardcore
>>101723049
>1girl?
meds
>>101723049
>1children?
kys
>>
File: __00894_.png (568 KB, 600x904)
568 KB
568 KB PNG
>>
File: 00011-3089448411.jpg (341 KB, 1728x2304)
341 KB
341 KB JPG
>>
File: 2024-08-04_00666_.png (1.62 MB, 1280x1280)
1.62 MB
1.62 MB PNG
>>
>>101722500
Very nice, thank you!
>>
>>101722230
There are kids jerking off to this.
>>
File: ComfyUI_00060_.png (1.04 MB, 1344x744)
1.04 MB
1.04 MB PNG
>>
>>101723078
thanks for the reply kind anon, now I am informed, and will keep the last two recommendations in mind.
>>
File: FLUX__00128_.png (569 KB, 1024x1024)
569 KB
569 KB PNG
>>
File: 2024-08-04_00668_.png (2.02 MB, 1280x1280)
2.02 MB
2.02 MB PNG
>>
>>101722981
more honkong kino?
>>
File: FLUX__00129_.png (1.05 MB, 1024x1024)
1.05 MB
1.05 MB PNG
>>
>>101723100
>>101723216
cryptominers taking proof-of-work pics
>>
>>101723078
>no, soft erotica to a certain point, no genitalia, no hardcore
Guess I'll wait for a fine-tune then.
>>
File: ComfyUI_Flux_80.png (1.47 MB, 1280x720)
1.47 MB
1.47 MB PNG
>>101722981
>>101722998
Yeah, pumping guidance to 3 clears the image a lot, but the soul is kinda lost. Guy's now a half poo or something.
>>101723219
>>
File: ComfyUI_01036_.png (1.99 MB, 920x1344)
1.99 MB
1.99 MB PNG
Its crazy how good flux is.

so how did they do it? whats their secret?

Also some other questions I have:
>will there be Loras and checkpoints for this too?
>any good tips for the comfyUI workflow?
>>
File: ComfyUI_Flux_2465.jpg (169 KB, 1024x768)
169 KB
169 KB JPG
not sure what this means, but I'm willing to look it up and try it

https://x.com/cubiq/status/1819680923815649304
>>
File: ComfyUI_01037_.png (1.88 MB, 920x1344)
1.88 MB
1.88 MB PNG
>>101723337
>>
>>101720035
Just got a new GPU with 16gb VRAM, is native training possible? I don't like the results I get with LORA at all.
>>
>>101723335
guidance is literally the slpo slider on flux because these crappy corpo models are always smeared in shit (read: fine-tuned with "aesthetic" data) to sell to customers
>>
File: ComfyUI_Flux_81.png (1.45 MB, 1280x720)
1.45 MB
1.45 MB PNG
>>101723335
>3.5 suddenly chadmaxxes the protag
lmao
>>101723370
We live in a society
>>
File: file.png (1.28 MB, 1280x896)
1.28 MB
1.28 MB PNG
>>
File: ComfyUI_00064_.png (919 KB, 1216x832)
919 KB
919 KB PNG
well at least there are 5 toes
>>
>>101723411
bootlegs.
>>
File: ComfyUI_Flux_2589.jpg (169 KB, 1024x768)
169 KB
169 KB JPG
>>
File: ComfyUI_01048_.png (1.44 MB, 920x1344)
1.44 MB
1.44 MB PNG
>>101723358
>>
File: ComfyUI_01027_.png (1.67 MB, 920x1344)
1.67 MB
1.67 MB PNG
>>101723462
>>
File: __00895_.png (577 KB, 600x904)
577 KB
577 KB PNG
>>101723201
kek
>>
>>101723415
Skill issue.
>>
>>101723201
howd you get it to look like an actual stick figure and not >>101723525
>>
>>101723543
I dunno.
A stick figure with a sad expression holding up a giant spoon above his head and saying "My spoon's too big" in a speech bubble.
>>
File: ComfyUI_01054_.png (978 KB, 920x1344)
978 KB
978 KB PNG
>>101723512
>>
File: FLUX__00134_.png (1012 KB, 1024x1024)
1012 KB
1012 KB PNG
>>
>>101723579
That's a big hand
>>
>>101723589
all the better to... never mind
>>
File: ComfyUI_Flux_84.png (1.41 MB, 1280x720)
1.41 MB
1.41 MB PNG
>>
Can I generate female nudity with Flux? Otherwise I won't even waste my time.
>>
>>101723633
Yes, but the nipples appear messed up most of the time
>>
File: ComfyUI_01042_.png (1.35 MB, 920x1344)
1.35 MB
1.35 MB PNG
>>101723568
>>
>>101723633
--> >>101714801
>>
File: ComfyUI_Flux_85.png (1.49 MB, 1280x720)
1.49 MB
1.49 MB PNG
Definitely not gonna watch this one
>>
File: ComfyUI_30710_.png (1.09 MB, 1024x1024)
1.09 MB
1.09 MB PNG
>>
>>101722789
why did they leave SAI?
>>
>>101722687
ESL?
>>
File: ComfyUI_30711_.png (1.18 MB, 1024x1024)
1.18 MB
1.18 MB PNG
>>
>>101723337
>whats their secret?
not shooting themselves in the foot, stability might as well be clowns in comparison. flux created a safe model that works, meanwhile stability nukes even navels.
>>
File: ComfyUI_01053_.png (1.81 MB, 920x1344)
1.81 MB
1.81 MB PNG
>>101723850
>meanwhile stability nukes even navels.
so why did they do this?
was it just retardation?
>>
File: FLUX__00140_.png (1.12 MB, 1152x896)
1.12 MB
1.12 MB PNG
>>
>>101723850
>created a safe* model
*please ignore the cunny it generates
>>
>>101723863
probably to make up for releasing the evil sd1.5 model that everyone used for UNSAFE porn
>>
>>101721749
Tet flux in vp8 mode at 768x768 and 512x512. It produces coherent results at those sizes.
>>
File: cui_0031.png (1.31 MB, 1536x2560)
1.31 MB
1.31 MB PNG
>>101652874
>>101653006
I am growing stronger. The gooners in trash have been significantly helpful.
>>
File: ComfyUI_01058_.png (1.63 MB, 920x1344)
1.63 MB
1.63 MB PNG
>>101723871
kek
>>101723893
>that everyone used for UNSAFE porn
and whats the problem with this?
>>
>>101723890
SAI shill cries out in pain as he strikes you.
>>
first flux lora results https://github.com/bghira/SimpleTuner/pull/622#issuecomment-2267624531
>>
>>101723983
so there is hope?
>>
>>101722529
Would you be willing to share the prompt of top left?
>>
>>101723770
They realized that the only purpose of that company is to stagnate tech and not evolve it. the entire company is currently at the mercy of the flux team not deciding to release a 4B model in the wild.
>>
File: ComfyUI_00076_.png (768 KB, 1152x896)
768 KB
768 KB PNG
>>101723968
>and whats the problem with this?
absolutely nothing, i hope to see many more safety clowns get BTFO's by good models
>>
>>101721273
wait, what is the blurry prompt? or another blurry one you don't care about sharing
>>
>>101721333
checked, barbie prompt?
>>
>>101723863
>require email for the safe version

whay are they doing?
>>
File: ComfyUI_01051_.png (1.57 MB, 920x1344)
1.57 MB
1.57 MB PNG
>>101724043
>absolutely nothing
then why did they do it?
whats this "safety" obsession about?
>>
File: FD_00014_.png (1.8 MB, 1024x1024)
1.8 MB
1.8 MB PNG
>>101724074
don't need it, once it's in the wild just download it from elsewhere
>>
>>101724074
just give them a burner email and claim your name is Nigger Niggardson like I did
>>
File: fallsched_4.jpg (341 KB, 1200x1528)
341 KB
341 KB JPG
>>
File: ComfyUI_30719_.png (960 KB, 1024x1024)
960 KB
960 KB PNG
>>
>>101724125
>woah guys look niggerkiller88@aol.com downloaded our model!
>>
>>101723863
the cycle of AI is always
>new thing
>it's too good and people have too much fun with it
>panic because advertisers and payment processors wont be pleased
>spend way too much time and money trying to "fix the problem"
>AI is now dogshit that appeals to nobody and no one will use it
>repeat
>>
File: file.png (1.77 MB, 1152x896)
1.77 MB
1.77 MB PNG
>>
>>101724166
>panic because advertisers and payment processors wont be pleased
how is that even a problem nowadays?
why dont they just collect funding and donations through crypto?
>>
>>101724162
Hey bro don't dox me
>>
>>101724168
how do you get retro?
>>
>>101724191
"A screenshot of nuclear explosion in the city from a N64 game, lowpoly"
>>
>>101723893
are any only capable of safe nudes, and no unsafe ones, and no sex?
>>
>>101724180
I guess cultural terrorism is pretty deeply ingrained in everyone's mind, thanks to social media
>>
>>101724166
flux is local, no one can ever take it from me
also, the standard by which new ai will be judged has been raised- users will not tolerate anything new if it can't compare to flux
>>
File: ComfyUI_01064_.png (1.11 MB, 1024x1024)
1.11 MB
1.11 MB PNG
also what GPUs do you dudes use and how efficient are you with the flux?

me I'm using an Gigabyte RTX 4090 Aero OC with 24GB vram,
generating 1 image takes somewhere in the ballpark of 40 seconds to 2+ minutes.
also what I found interesting is that this flux model doesnt seem to stress the GPU all that much, the GPU temps are very good while with SD models they go up way higher and the GPU also seems to draw less power with flux.
>>
File: img.png (793 KB, 897x591)
793 KB
793 KB PNG
>>101724180
you have to do business and the vast majority of normies won't want or know how to use crypto
personally, I think we should do pic related to solve the payment processor question.
>>
>>101724220
No one can take it but arguable they avoided training certain concepts to cover their ass
Of course someone will fix that sooner or later
>>
File: up_0002.jpg (848 KB, 4096x2816)
848 KB
848 KB JPG
>>
File: ComfyUI_01063_.png (1.06 MB, 1024x1024)
1.06 MB
1.06 MB PNG
>>101724231
>you have to do business
what business? they arent really selling these models are they?
>the vast majority of normies won't want or know how to use crypto
shitloads of people already have crypto and these people are also usually tech savy so I'd image the user base overlaps here a lot.
>I think we should do pic related to solve the payment processor question.
Chinese Communism?
>>
>>101724226
>generating 1 image takes somewhere in the ballpark of 40 seconds to 2+ minutes.
Maybe that's about right? My 6950xt gets me 14.88 sits. what are yours?
>>
>>101724220
>no one can ever take it from me
It's like the new Linux.
>>
>>101721818
Does it really? Fuck I only have 10 GB.
>>
File: ComfyUI_flux_00045.jpg (389 KB, 1536x1024)
389 KB
389 KB JPG
>>101724226
RTX 3090 msi trio
picrel takes 58 sec at 1536x1024 25 steps euler
>>
File: file.png (1.33 MB, 1280x896)
1.33 MB
1.33 MB PNG
>>
>>101724252
i managed to get 60s per gen on 2x 3080tis (3s/it)
>>
>>101724249
>what business? they arent really selling these models are they?
while dev and schnell are opensource, BFLA have a pay-per-image api service. as well as a closed-source "pro" model also SaaS
>>
does flux run on amd?
>>
File: file.png (2.09 MB, 1152x896)
2.09 MB
2.09 MB PNG
>>101724282
>>
File: ComfyUI_01065_.png (1.43 MB, 1024x1024)
1.43 MB
1.43 MB PNG
>>101724252
>what are yours?
seems to vary a lot for me, I think that has probably to do with me doing other shit while it generates but dunno
>>
File: ComfyUI_00082_.png (520 KB, 1216x832)
520 KB
520 KB PNG
Imagine the coom when there is a flux well-tuned for porn
Not that I'm a coomer who wants that or anything haha... unless?
>>
File: ComfyUI_01066_.png (1.44 MB, 1024x1024)
1.44 MB
1.44 MB PNG
>>101724287
I see, so their problem is basically the SaaS not wanting coomer customers?
>>
File: FD_00148_.jpg (811 KB, 2048x2048)
811 KB
811 KB JPG
Flux (FP8) seems to break down at native 2048x2048 but is fine at 1536x1536.
>>
>>101724166
This. I love Flux but it too is on a timebomb before they pull another SAI or go closed source and then we have to rely on yet another startup, repeat.
>>
File: FD_00147_.png (3.45 MB, 1536x1536)
3.45 MB
3.45 MB PNG
>>101724344
1536x1536 for comparison, same seed and prompt.
>>
>>101724360
*jaw drops to floor, eyes pop out of sockets accompanied by trumpets, heart beats out of chest, awooga awooga sound effect, pulls chain on train whistle that has appeared next to head as steam blows out, slams fists on table, rattling any plates, bowls or silverware, whistles loudly, fireworks shoot from top of head, pants loudly as tongue hangs out of mouth, wipes comically large bead of sweat from forehead*
>>
File: file.png (2.32 MB, 1152x896)
2.32 MB
2.32 MB PNG
>>101724370
>>
File: ComfyUI_00078_.png (907 KB, 904x1024)
907 KB
907 KB PNG
>>101723525
heh
>>
These times all sound higher than I would expect for the powerful hardware, are you all using dev?

3060-12G | i7-9700K | 32GB DDR4:

schnell @ fp16 | 4 steps | 1024x1024
first gen for a prompt: avg 118.9s (n=31)
repeat prompt only changing seed: avg 34.0s (n=224)

dev @ fp16 | 20 steps | 1024x1024
first gen for a prompt: avg 216.5s (n=24)
repeat prompt only changing seed: avg 121.1s (n=139)
>>
>>101724300
Honestly, I am now wondering if Flux is friendly to amd vs what we might expect.
>>
>>101724057
the usual young blonde 1girl + retro artstyle. Loras do the work
>>
>>101724197
so it doesn't know what N64 games look like, at all
>>
>>101724416
my 6950rx gets me 14.88 sets, with fp8. what are you getting with fp8? I jave not tried the fpX thing and don't know how iy works.
>>
File: file.png (1.12 MB, 1152x896)
1.12 MB
1.12 MB PNG
>>101724438
>>
>>101724438
I bet it's trained on comments on an art site.
>>
>>101724445
>title screen
>no game
what are you trying to say, anon
>>
File: file.png (2.51 MB, 1152x896)
2.51 MB
2.51 MB PNG
>>101724393
>>
File: file.png (1.9 MB, 1152x896)
1.9 MB
1.9 MB PNG
>>101724452
>>
File: ComfyUI_Flux_2761.jpg (195 KB, 1024x768)
195 KB
195 KB JPG
>>
>>101724469
so it can't do N64, got it. thanks for the examples
>>
>>101724289
YES

NO

it's hard bakalaka
>>
File: FD_00149_.jpg (723 KB, 2048x2048)
723 KB
723 KB JPG
>>101724344
Adding steps seems to have helped somewhat. This was 30 steps. Gonna try 50. RIP my GPU
>>
>>101724484
It can, I saw gens, i just took them for inspiration, didn't actually try to get N64 look that hard
>>
So is this like a high res local dalle3? Can it do a mazda rx-7 drifting on a canyon road? Maybe a bathtub full of sausages?
>>
>>101724501
It can't, not by prompting "N64" or "Nintendo 64"
>>
File: file.png (2.17 MB, 896x1152)
2.17 MB
2.17 MB PNG
>>101724506
>>
File: FLUX__00153_.png (1.14 MB, 1024x1024)
1.14 MB
1.14 MB PNG
>>101724506
I dunno is this is the right car
>>
>>101724529
Bros... we made it.
What the hell is that poopy water though.
>>
File: file.png (2.14 MB, 1152x896)
2.14 MB
2.14 MB PNG
>>101724506
and
>>
>>101724560
nta but that's the correct car
>>
>>101724560
>>101724564
funny how similar they are lol
>>
>>101724560
>>101724564
Damn. It really is local dalle3.
>>
>>101724575
dalle3 can do Scarlett Johansson. Flux cannot.
It is useless to me.
>>
File: FLUX__00151_.png (1.01 MB, 896x1152)
1.01 MB
1.01 MB PNG
>>
>>101724439
hang on, I haven't tried yet (I want to test with full capacity) i'll get back to you when queue clears...
>>
File: ComfyUI_05958_.png (1.09 MB, 1024x1024)
1.09 MB
1.09 MB PNG
>>101724561
Sausage water, anon.
>>
>>101724586
Female celebrities censored out (but probably still trained). Artist names censored out (but probably still trained). This is a temporary issue.
>>
File: file.png (1.88 MB, 896x1152)
1.88 MB
1.88 MB PNG
>>101724586
>>
>>101724586
Shame. At least it's still progress compared to the SD shitfest.
>>
When comfy says loading in lowvram mode is it automatically switching to fp8 down from 16 or something entirely different?
>>
>>101724604
Yum.
>>
>>101724608
you don't really think that's her, do you?
>>
>>101724605
Wouldn't it be possible for someone to make a VLM out of Flux and then reverse engineer what it thinks the celebrities and artists are?
>>
File: file.png (1.81 MB, 896x1152)
1.81 MB
1.81 MB PNG
>>101724604
>>
File: FD_00150_.png (3.72 MB, 2048x2048)
3.72 MB
3.72 MB PNG
>>101724496
More steps helps. This is 50
>>
File: ComfyUI_Flux_2795.jpg (233 KB, 1024x768)
233 KB
233 KB JPG
>>
File: ComfyUI_01903_.png (1.62 MB, 1280x1024)
1.62 MB
1.62 MB PNG
>>
File: ComfyUI_01068_.png (1.54 MB, 1024x1024)
1.54 MB
1.54 MB PNG
>>101724645
>>
>>101724656
try this with 1536x1536, going above this seems to start breaking things regardless of steps
>>
File: ComfyUI_Flux_2797.jpg (205 KB, 1024x768)
205 KB
205 KB JPG
>>
so why is it an llm can be offloaded on multiple gpus but not an imagen, how would that work with vlm then
>>
>>101724586
>gen an image
>inpaint face at 0.5+ denoise using your preferred SDXL slop
simple as
>>
>>101724657
would love the prompt, love the creepy stuff
>>
>>101724689
the body will be different, also SDXL doesn't know her either, it is closer but so off.
>>
>>101724688
don't see why imagegen wouldn't be able to be split across gpus, it's just a matter of bothering to implement it
>>
File: file.png (2.06 MB, 896x1152)
2.06 MB
2.06 MB PNG
>>101724671
>>
File: ComfyUI_Flux_2781.jpg (208 KB, 1024x768)
208 KB
208 KB JPG
>>101724706
https://files.catbox.moe/pbzm3u.webp
>>
>>101724709
There's like 10 loras of her on civitai for sd1.5, sdxl and pony
>>
>>101724746
all shit
>>
File: FLUX__00157_.png (915 KB, 1024x1024)
915 KB
915 KB PNG
>>
File: ComfyUI_Flux_2807.jpg (242 KB, 1024x768)
242 KB
242 KB JPG
>>
>>101724709
https://github.com/TencentARC/PhotoMaker
does this work?

The guy that got in the news over the gens of Taylor Swift basically being hazed at a football game used something like that, not sure which one.
>>
File: ComfyUI_Flux_2811.jpg (218 KB, 1024x768)
218 KB
218 KB JPG
>>
>>101724770
>The guy that got in the news over the gens of Taylor Swift basically being hazed at a football game used something like that, not sure which one.
He took the Taylor Swift images from the DALL-E 3 threads, took pretty much all celeb images to sell on Twitter but only heard drama regarding Taylor Swift's.
>>
File: file.png (1.23 MB, 1152x896)
1.23 MB
1.23 MB PNG
>>101724741
cheers
>>
File: FLUX__00030_.png (905 KB, 896x1152)
905 KB
905 KB PNG
>>
>>101724785
he genned them, and for some reason posted them on /pol/

it's relevant, because women found out about ai, or some of them did, that never heard of it really.
>>
File: 1466639626445.gif (1.5 MB, 355x225)
1.5 MB
1.5 MB GIF
>>101724689
>Replace Flux's high quality face with sd's same face-same expression retarded slop
Genius move right there
>>
File: ComfyUI_Flux_2803.jpg (234 KB, 1024x768)
234 KB
234 KB JPG
>>
>>101724792
It's creepy, because that's the kind of person who would never say hi to me. In some way it's a hidden ugliness, thus it symbolizes evil, and it's a surprisingly universal symbol of evil.
>>
>>101724092
>whats this "safety" obsession about?
It's convergent evolution, every AI image model needs to converge to maximum safety, where you can only make cat memes.
>>
>>101724709
>the body will be different
yeah it probably wont be a jew fridge body, whats the problem?
>>
>>101724839
What about:
>>101724770
>>
>>101724605
>This is a temporary issue.
How?
Can someone really finetune it to add artists, characters, nsfw, and female celebs, all while not going back to tag writing?
Who even has such a gigantic training dataset, and the money to finetune this?
>>
so are flux LoRAs letting me gen anime porn of my waifu yet?
>>
>>101724827
Worst thing that could happen, is to make normies go on a moral panic about this.
>>
>>101724827
>he genned them, and for some reason posted them on /pol/
he didn't gen them and they were being posted on /trash/ if my memory doesn't fail me. can't recall the board but I never go to /pol/ so unless someone was posting on both it was not from there
>>101724854
cool it
>>
File: 1716440808751178.jpg (123 KB, 1010x1010)
123 KB
123 KB JPG
>>
File: ComfyUI_00096_.png (1.18 MB, 1024x1024)
1.18 MB
1.18 MB PNG
Holy shit guys. I can finally emulate the finest youtuber thumbnails.
>>
>>101724660
Catbox for this one?
>>
File: ComfyUI_30720_.png (1.61 MB, 1024x1024)
1.61 MB
1.61 MB PNG
>>
File: ComfyUI_30721_.png (1.64 MB, 1024x1024)
1.64 MB
1.64 MB PNG
>>
File: ComfyUI_01940_.png (1.59 MB, 1280x1024)
1.59 MB
1.59 MB PNG
>>
File: file.png (1.34 MB, 1152x896)
1.34 MB
1.34 MB PNG
>>101724845
>>
File: ComfyUI_01960_.png (1.63 MB, 1280x1024)
1.63 MB
1.63 MB PNG
>>
File: file.png (363 KB, 526x567)
363 KB
363 KB PNG
Flux is stealing shit from actual people lmao
>>
>>101724992
Oh god. Can it generate terrible sonic OC?
>>
File: FLUX__00034_.png (1015 KB, 896x1152)
1015 KB
1015 KB PNG
>>101725010
ye
>>
File: ComfyUI_30722_.png (1.73 MB, 1024x1024)
1.73 MB
1.73 MB PNG
Asking AI to write long detailed prompts for imaginary product advertisement posters was a good idea.
>>
File: ComfyUI_30723_.png (1.36 MB, 1024x1024)
1.36 MB
1.36 MB PNG
>>
Ready to roll with fresh bread...
>>101725030
>>101725030
>>101725030
>>
File: ComfyUI_00095_.png (802 KB, 1152x896)
802 KB
802 KB PNG
>>101724906
not yet, we'll get there
>>
File: file.png (305 KB, 527x588)
305 KB
305 KB PNG
>>101725015
yep
>>
>>101725010
Not yet, actually, apparently. But maybe someone will figure it out.
>>
>>101725048
What if you use deviantart as a quality modifier?
>>
>>1017250
i'm just using the website.
https://huggingface.co/spaces/black-forest-labs/FLUX.1-schnell

I can't use any modifiers.
>>
>>101725048
chris chan in shambles
>>
File: file.png (195 KB, 552x605)
195 KB
195 KB PNG
>>101725093
alternate universe sonichu?
>>
File: 666 doggo.png (1.61 MB, 1024x1024)
1.61 MB
1.61 MB PNG
>>101722504
I RUINED AI ALL NIGHT LONG

seed: 666
guidance: 6.7

https://files.catbox.moe/fvq4xd.png
>>
File: ComfyUI_30725_.png (1.03 MB, 1024x1024)
1.03 MB
1.03 MB PNG
>>
>>101725113
OOPS

doggo is seed 553046854372910
>>
>>101724679
I did. See >>101724360
>>
>>101725113
but I will still call him 666 doggo. What's he been up to here?
>>
>>101725048
Who is the bald purple Sanic?
Chemo the Hedgehog?
>>
>>101725048
That's very nicely drawn. Definitely gets an A+ in art class. Makes sense, people don't upload bad drawings.
>>
>>101725131
Wait, when I enter seed 666, it turns into
>553046854372910
>>
>>101724226
3080 with 32gb system ram. prefer no sysmem in nvidia settings.

takes 90-120 seconds for fp16 using the default settings from the comfy example. for fp8 it's around 60 seconds, usually just under.

there's a little bit of swapping to pagefile with fp16 so would probably be way fucking slower without nvme.
>>
>>101725375
>32gb system ram
does that matter?
I thought its only about vram
>>
>>101725419
YES!!!

I had 16gb in my system, with a 6950xt, which has 16gb of VRAM, I thought that should be enough, but I guess it might take a whole hour to generate with fp8, I gave up and tried installing the 16gb from another system, now 32gb. Well now I gen an image in about 5-6 minutes.

And maybe I need more ram.
>>
>>101725419
i'm not gonna pretend to know how/why it works but since it is a swapping a lot it seems to matter a bit. i suspect as long as you have "enough" (more than 32gig) sysram the speed/performance of it doesn't matter that much.
>>
>>101725375
What's your sits on fp8? Mine is 14.88, on the 6950xt.
>>
>>101725505
2.6
>>
>>101725466
>>101725474
I have 64GB ram, if I upgrade to 128GB ram it will not become faster?
>>
>>101725505
>14.88
>>
>>101725582
i doubt it. it might allow for better vram swapping if you have prefer sysmen fallback enabled in nvidia settings. worth trying for science if you have spare ram lying around but i wouldn't spend money unless somebody can confirm it helps.
>>
Just kind of funny that's where it averages.
>>
>>101724882
The solution wouldn't be a regular finetune, but something similar to a vector DB that dynamically applies a LoRA on the fly. Ideally the LoRAs wouldn't all be trained manually and just values generalized from the data itself. Something similar has been done for SD, forgot what it's called.
>>
Are safetensors slower?
>>
>>101724439
I'm getting same speeds (give or take 3 or 4 seconds) with fp8 idk what to tell you
>>
>>101726677
Interesting. Well, it means your card works great with Flux, vs AMD.
>>
>>101724586
There may be a way to find the negative prompts.

This may be the gateway to reverting baked in negatives.



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.