[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 deleted.
Discussion of free and open source text-to-image models

Previous /ldg/ bread : >>101727444

>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
>>
Nigger
>>
File: eeehhh.jpg (1.11 MB, 3072x3072)
1.11 MB
1.11 MB JPG
I suck at configuring this stuff, but I see why people like the technology.
>>
File: FLUX_00016_.png (1.31 MB, 896x1152)
1.31 MB
1.31 MB PNG
>>
File: ComfyUI_0006.jpg (1.86 MB, 1400x1800)
1.86 MB
1.86 MB JPG
is it possible to inpaint with flux? or img2img
>>
Anons is there a way to face-swap without using those fake paid apps? There used to be a website for that but it's gone as far as I'm aware

And also, how do you even deepfake? Is there a place to learn before it gets nuked by the governments?

Thanks
>>
File: FD_00571_.png (1.88 MB, 1024x1536)
1.88 MB
1.88 MB PNG
>>101729521
img2img works, haven't tried inpainting because inpainting in comfy is dogshit
>>
>>101729524
Depends. What do you want to make?
>>
>>101729556
Just want to face-swap some random Instagram thots, if that's easily done nowadays some deepfakes would be great too, the coom would never stop coming lmao
>>
File: FD_00581_.png (2.13 MB, 1024x1536)
2.13 MB
2.13 MB PNG
>>101729524
Try this
https://huggingface.co/spaces/TencentARC/PhotoMaker-V2
>>
>>101729601
Looks promising thanks
>>
File: ComfyUI_00201_.png (832 KB, 1024x1024)
832 KB
832 KB PNG
>>
>>101729647
catbot for this one?
>>
File: ComfyUI_00437_.png (1.25 MB, 1024x1024)
1.25 MB
1.25 MB PNG
>>
File: ComfyUI_00202_.png (906 KB, 1024x1024)
906 KB
906 KB PNG
>>101729667
enjoy https://files.catbox.moe/tfw9t0.png
>>
File: ComfyUI_01525_.png (1.93 MB, 1024x1016)
1.93 MB
1.93 MB PNG
>>
How come the random seed in ComfyUI's flux workflow example always reset to the original value even though you changed the number? It doesn't do this with SD/SDXL workflows.
>>
>>
>>101729711
It randomizes after the first run for some reason.
>>
File: 1706340708837270.png (1.22 MB, 1024x1024)
1.22 MB
1.22 MB PNG
>>
>>101729728
If it changes when you moved the mouse then prob custom nodes incompability.
>>
File: FD_00598_.png (2.25 MB, 1024x1536)
2.25 MB
2.25 MB PNG
>>101729688
GTA is always to unrealistic, why is there a toilet?
>>
File: fs_0905.jpg (102 KB, 960x960)
102 KB
102 KB JPG
>>
File: ComfyUI_00204_.png (865 KB, 1024x1024)
865 KB
865 KB PNG
>>
File: fs_0912.jpg (96 KB, 1120x800)
96 KB
96 KB JPG
>>
>>101729623
>>101729573
look up reactor for a simple face swap
for deep fakes you need to learn how to use stable diffusion + loras or even train loras
>>
>>
>>101729379
Can local models do text now? Are they compatible with A1111?
>>
File: 2024-08-04_00357_.png (1.16 MB, 1024x1024)
1.16 MB
1.16 MB PNG
>>101729811
>Can local models do text now?
FLUX can
>Are they compatible with A1111?
no
>>
>>101729811
>>>/g/lmg
>Can local models do text now
yes
People will debate how well they do though.
>>
>>101729811
no compatability with A1111 yet, but looks like it will be added based on the github issue
>>
>>101729811
try flux:
https://replicate.com/black-forest-labs/flux-dev
>A1111
2 weeks
>>
File: file.png (1.23 MB, 1024x1024)
1.23 MB
1.23 MB PNG
heh
>>
File: fs_0941.jpg (88 KB, 1120x800)
88 KB
88 KB JPG
>>
>>101729837
No mario noooo!
>>
>>
>>101729852
Based gentleman.
>>
File: fs_0972.jpg (115 KB, 1120x800)
115 KB
115 KB JPG
>>
File: ComfyUI_07216_.png (1.4 MB, 1024x1024)
1.4 MB
1.4 MB PNG
>>
File: ComfyUI_00457_.png (1016 KB, 1344x768)
1016 KB
1016 KB PNG
>>
File: fs_0978.jpg (88 KB, 1120x800)
88 KB
88 KB JPG
>>
File: FD_00624_.png (1.88 MB, 1024x1024)
1.88 MB
1.88 MB PNG
>>101729923
It's only 5pm anon
>>
>>101729939
A story of hardship and perseverance. Inspiring
>>
File: 1704851545532665.png (1.41 MB, 1024x1024)
1.41 MB
1.41 MB PNG
>>
File: ComfyUI_00454_.png (1.18 MB, 1024x1024)
1.18 MB
1.18 MB PNG
>>
File: fs_0985.jpg (107 KB, 1120x800)
107 KB
107 KB JPG
>>
File: 1712927420519585.png (1.28 MB, 1024x1024)
1.28 MB
1.28 MB PNG
pixel art makes some neat stuff

pixel art. Final Fantasy in game screenshot. UI and HUD visible. The main character is Hatsune Miku. The setting is a forest with a large lake.
>>
File: 1.jpg (2.35 MB, 1904x3360)
2.35 MB
2.35 MB JPG
>>
File: fs_0996.jpg (83 KB, 1120x800)
83 KB
83 KB JPG
>>
>>101729601
...What a dick.
>>
>>101729977
sex big sex big sex
>>
File: ComfyUI_00462_.png (1.38 MB, 1024x1024)
1.38 MB
1.38 MB PNG
>>
File: fs_1002.jpg (96 KB, 1120x800)
96 KB
96 KB JPG
>>
>>101729797
Thanks anon, figured it'd be the more advanced stuff
>>
>>101729973
looks nothing like a game screenshot but i quite like it anon.
>>
>>101729983
tough break...
>>
File: 1717255733725704.png (1.18 MB, 1024x1024)
1.18 MB
1.18 MB PNG
>>101730026
if you are specific with the game, it can do a lot of neat stuff. here is WoW
>>
File: ComfyUI_00011_.png (1.5 MB, 1024x1024)
1.5 MB
1.5 MB PNG
Damn, it doesn't know who Carl Brutananadilewski is
>>
can flux make bright/dark images or is it yet again forced to an average brightness like SD?
>>
>>101729961
Would
>>
File: fs_1033.jpg (117 KB, 1120x800)
117 KB
117 KB JPG
>>
>>101730031
can you try "wow character screen" ? i recall de3 was making really believable ones for that.
>>
File: fs_1039.jpg (92 KB, 1120x800)
92 KB
92 KB JPG
>>
File: ComfyUI_00470_.png (1.4 MB, 1024x1024)
1.4 MB
1.4 MB PNG
>>
File: ComfyUI_00471_.png (788 KB, 1200x768)
788 KB
788 KB PNG
>>
File: FD_00400_.png (1.06 MB, 1024x1024)
1.06 MB
1.06 MB PNG
>>101729961
I look like that
>>
File: 1693190407730156.png (1.37 MB, 1024x1024)
1.37 MB
1.37 MB PNG
>>101730091
genning, in the mean time this is another output when prompting "in the style of a super nintendo game":
>>
File: 1722836449.png (1013 KB, 1024x1024)
1013 KB
1013 KB PNG
>>
File: 2024-08-05_00079_.png (1.22 MB, 1024x1024)
1.22 MB
1.22 MB PNG
>>
File: ComfyUI_00212_.png (1010 KB, 832x1216)
1010 KB
1010 KB PNG
>>
File: ComfyUI_00480_.png (1.34 MB, 1024x1024)
1.34 MB
1.34 MB PNG
>>101730172
would get massage
>>
File: ComfyUI_00482_.png (1.39 MB, 1024x1024)
1.39 MB
1.39 MB PNG
>>
File: fs_1100.jpg (110 KB, 1120x800)
110 KB
110 KB JPG
>>
File: 2.jpg (1.12 MB, 1920x1080)
1.12 MB
1.12 MB JPG
>>
File: ComfyUI_00486_.png (1.32 MB, 1024x1024)
1.32 MB
1.32 MB PNG
>>
File deleted.
>>
File: ComfyUI_00481_.png (706 KB, 1200x768)
706 KB
706 KB PNG
It does pretty good character guides for 3D modeling. I could use this.
>>
>>101730203
Catbox?
>>
File: 1722837009.png (1.13 MB, 1024x1024)
1.13 MB
1.13 MB PNG
>>
>>101726849
That's not how any of it works.
>>
File: FD_00633_.png (1.17 MB, 1024x1024)
1.17 MB
1.17 MB PNG
Ummm actually he identifies as a cat
>>
>>101726849
Don't listen to him anon, keep looking for them
>>
>>101730247
progressive doglion
>>
File: fs_1129.jpg (110 KB, 1120x800)
110 KB
110 KB JPG
>>
File: 1719396854725326.png (1.06 MB, 1024x1024)
1.06 MB
1.06 MB PNG
>>101730091
it works:

World of Warcraft ingame screenshot, UI and HUD visible. Character select screen. The main character is Miku Hatsune, she is wearing a wizard robe and is holding a magic staff. The background is Icecrown Citadel from World of Warcraft.
>>
>>101730197
What a time to be alive
I have gtg now but I'm completely invested! good work
>>
File: 3.jpg (1.96 MB, 1792x2304)
1.96 MB
1.96 MB JPG
>>
File: fs_1145.jpg (70 KB, 1120x512)
70 KB
70 KB JPG
>>
File: 1698398387972537.png (1019 KB, 1024x1024)
1019 KB
1019 KB PNG
>>101730273
>>
File: Flux_00174_.png (1.27 MB, 1128x904)
1.27 MB
1.27 MB PNG
Anyone want some bungers?
>>
File: FLUX__00028_.png (1.74 MB, 1024x1024)
1.74 MB
1.74 MB PNG
>>101730328
i'm good
>>
File: 1706967619111156.png (443 KB, 640x480)
443 KB
443 KB PNG
>try switching to 640x480 for a more vintage pixel art game resolution and higher throughput
>the pixel art-ness actually decreases
wtf
>>
>>101730328
as long as they have the best buns
>>
https://www.reddit.com/r/StableDiffusion/comments/1ekf1mw/flux_dev_fluxguidance_node_guidance_value_tests/
So almost any CFG value is usable on Flux and it's very situational
>inb4 r*ddit
>>
File: fs_0030.jpg (89 KB, 1120x512)
89 KB
89 KB JPG
>>
File: ComfyUI_00489_.png (962 KB, 1024x1024)
962 KB
962 KB PNG
>>
File: fs_0039.jpg (63 KB, 1120x512)
63 KB
63 KB JPG
>>
File: 2024-08-04_00665_.png (1.42 MB, 1280x1280)
1.42 MB
1.42 MB PNG
>>101730338
thats not how diffusion works, it doesnt really make pixel art, it just immitates it, so more pixels to work with will make it have more room to wiggle those fake big pixels
>>
File: FD_00660_.png (1.25 MB, 1024x1024)
1.25 MB
1.25 MB PNG
Hyped for the Street Sharks remake
>>
so when do i get to prompt "1girl, huge ass, spread anus, giant stinky asshole" on flux?
>>
File: ComfyUI_00495_.png (1.45 MB, 1024x1024)
1.45 MB
1.45 MB PNG
>>101730402
>>
>>101730402
Never because that's a shit prompt.
It would be "A woman with a huge ass, spreading her ass cheeks to reveal her unwashed asshole"
>>
File: file.png (1.03 MB, 1216x832)
1.03 MB
1.03 MB PNG
>>
>>101730447
>>101730402
I've found 'Butt' seems to work better than ass. Still need to test emojis
>>
File: ComfyUI_00220_.png (1.19 MB, 1024x1024)
1.19 MB
1.19 MB PNG
>>
>>101730447
you have never used PonyXL, it's a perfectly fine prompt and it is Flux's faulty training fault for not getting it
>>
>>101730447
thank you for teaching me the proper way to generate images of big stinky shit-laced assholes
>>
>>101730447
Is that what they call boomer prompting?
>>
>>101730466
pony is based on SDXL therefore uses clip and booru tags, FLUX is using txxl, a prompt like that will (even if finetuned) just produce crap
>>
File: fs_0116.jpg (71 KB, 1120x512)
71 KB
71 KB JPG
>>
File: ComfyUI_00071_.png (448 KB, 904x1024)
448 KB
448 KB PNG
>>
File: Untitled.png (392 KB, 3820x1826)
392 KB
392 KB PNG
>>101730338
if you're using comfy then search the custom nodes for something
>>
>>101730466
You dumbass, it's got nothing to do with Pony, it's because Flux isn't trained on booru tags.
Why are coomers literal brainlets?
>>
>>101730505
Thanks, I'm using Comfy for the first time so I'm sure I'm using less than 1% of its potential right now.
>>
File: 1722839125.png (1.16 MB, 1024x1024)
1.16 MB
1.16 MB PNG
>>
>>101730483
>based on SDXL therefore uses clip and booru tags
the fuck are you talking about, do you want to rephrase that illogical sentence?
>(even if finetuned)
What is it about T5 that, even if finetuned, cannot work with booru tags?
>>101730506
you're actually retarded if you think any of "huge ass, spread anus, giant stinky asshole" requires being trained on booru tags to understand, they are fucking common English words
>>
>>101730505
Hmmm. I had no idea that top bar was even a thing.
>>
File: FD_00653_.png (1.38 MB, 1024x1024)
1.38 MB
1.38 MB PNG
>>101730537
Flux works with both booru and t5. Either is fine.
>>
File: ComfyUI_00504_.png (1.13 MB, 1208x800)
1.13 MB
1.13 MB PNG
>>
>>101730555
>works with both booru and t5
What the fuck do you think "T5" means that makes have anything to do with "booru"
your "booru" prompt goes through fucking T5 you fucking techlet fucking ignorant baboon
>>
>>101730573
Simply saying you can boomer or booru and it won't make a difference, it can understand and infer from both.
Boomer prompting is better though since you can describe the locations of things, but your old prompts will work as normal, minus the model quirks.
Point is, there's no point in arguing about this because it doesn't matter.
>>
>>101730590
That's actually nice to know and explains why I was getting anime outputs without asking for anime.
>>
>>101730590
>Simply saying you can boomer or booru
You didn't say that, did you? You said something stupid like "booru and T5" because you know fuck all
>it can understand and infer from both
Did you not see the "1girl, etc" attempt above? Flux doesn't know sexual shit. It cannot make use of those words.
>>
File: ComfyUI_00510_.png (723 KB, 1208x800)
723 KB
723 KB PNG
>>
>>101730612
I did say that, I am using t5 and boomer prompt interchangeably. It's clear and obvious with moderate reading comprehension.
It's the most retarded thing you are trying to argue, I was actually agreeing with you but the fact that you are being a faggot about it means I am going to disagree with you out of principle now.
>>
File: 2024-08-05_00156_.png (1.49 MB, 1024x1024)
1.49 MB
1.49 MB PNG
>>
>>101730641
>I am using t5 and boomer prompt interchangeably. It's clear and obvious with moderate reading comprehension.
You're retarded, please kill yourself and then learn how this shit works and what things are called.
>>
File: 2024-08-05_00157_.png (1.37 MB, 1024x1024)
1.37 MB
1.37 MB PNG
>>
>>101730203
Holy shit
>>
>>101730647
I call all games with permadeath roguelikes.
>>
Does GPU get bottlenecked by CPU when doing image generation? I know that for rendering/gaming it causes problems but if I'm just using VRAM does the CPU matter at all?
>>
File: file.png (2.44 MB, 1216x832)
2.44 MB
2.44 MB PNG
>>
>>101730663
Short answer is yes. But it's not as big of an impact as you think.
>>
File: 2024-08-05_00159_.png (1.38 MB, 1024x1024)
1.38 MB
1.38 MB PNG
>>101730663
not really, overall CPU/system performance is kinda important for pre and post processing, loading models, saving files etc. but once the model is loaded and the diffusion is running its all GPU .. just hve a semi recent CPU with enough RAM (32GB+) and you are fine
>>
File: 1722840250.png (1.28 MB, 1024x1024)
1.28 MB
1.28 MB PNG
>>
File: 1722312966391102.png (1.27 MB, 1024x1024)
1.27 MB
1.27 MB PNG
>>
File: ComfyUI_00729_.png (1.05 MB, 1216x832)
1.05 MB
1.05 MB PNG
>>
>>101730681
>>101730669
Thank you. I'm running 32GB of ram with a 3090 (24GB vram) and I'm hoping that will be enough for image and simple-ish local chatbot stuff.
>>
File: 2024-08-05_00165_.png (1.1 MB, 1024x1024)
1.1 MB
1.1 MB PNG
>>101730752
that will be fine for both use cases
>>
File: FD_00668_.png (991 KB, 1024x1024)
991 KB
991 KB PNG
>>101730752
I am running a 4080 on an i5 10400k 32gb ram. I gen fine on fp8
>>
File: ComfyUI_00225_.png (1.05 MB, 1344x768)
1.05 MB
1.05 MB PNG
>>
this is big - negative prompting in flux
https://www.reddit.com/r/StableDiffusion/comments/1ekgiw6/heres_a_hack_to_make_flux_better_at_prompt/
>>
>>101730787
I haven't used negative prompts since 1.5. Dropped them in XL. Only picked them up again in pony because he is a faggot
>>
>>101730741
Yes, do it
>>
File: Soyjak.jpg (15 KB, 250x311)
15 KB
15 KB JPG
>>101730787
>this is big - negative prompting in flux
>>
>>101730779
Adolf Hitler, he was a great guy, a great guy
>>
>>101730767
>>101730777
Thank you both. I'm assuming DDR5 is not a necessity and that DDR4 will work fine.
Have a lovely night!
>>
>>101730796
Same, only time I can recall using them in recent memory was to stop a generation from producing monochrome results. Haven't needed them otherwise.
>>
new model can only do nazi memes? interesting art is out of the question?
>>
>>101730818
People have no imagination and it can't do porn (yet)
>>
File: FD_00422_.png (1.06 MB, 1024x1024)
1.06 MB
1.06 MB PNG
>>101730800
>nogen
>>
File: ComfyUI_00205_.png (751 KB, 1024x1024)
751 KB
751 KB PNG
>>
>>101730800
huge even
>>
>>101730818
Name a propriety model that can do Nazi memes. Or even a model that can reliably do a swastika without fucking it up.
>>
File: FD_00582_.png (1.74 MB, 1024x1024)
1.74 MB
1.74 MB PNG
>>101730818
What do you mean by interesting art?
Keep in mind the majority of people here are either coomers or techfags, and in both cases creatively bankrupt.
>>
>>101730787
Just hack on top of a hack with this model, huh?
>>
>>101730840
>Name a propriety model that can do Nazi memes
Google gemini before they removed it for generating too many black Nazis. Not even joking. Shit was hilarious
>>
File: 1722841258.png (1.1 MB, 1024x1024)
1.1 MB
1.1 MB PNG
>>
>>101729379
how much vram do you need for flux? is 16GB enough
>>
>>101730871
Yes. I'm running it (slowly) with 8.
>>
File: 2024-08-05_00164_.png (1.09 MB, 1024x1024)
1.09 MB
1.09 MB PNG
>>101730871
16GB will work with some swapping into system ram and fp8 mode on model
>>
File: 1722841443.png (975 KB, 1024x1024)
975 KB
975 KB PNG
>>101730871
You can still run it with the fp16 encoder if fp8 encoder doesn't work. You should try the Schnell model first.
>>
https://x.com/JuhanaMyllys/status/1820138628447707617

l m a o
>>
File: ComfyUI_00228_.png (1.02 MB, 1024x1024)
1.02 MB
1.02 MB PNG
>>
>>101729822
Blessed.
>>101729827
>>101729830
Waiting two more weeks (tm) then
>>
Are they planning on releasing paper about flux and any code, or just dropping black box model and allowing community to hack it?
>>
>>101730868
>>101730923
I recognize the attempts at making it generate lewd things but I think I'll keep using Pony until Flux can do actual porn and not have to do weird prompts to make girls hold vaguely phallic objects
>>
>>101730857
What a dumb statement. The entire imagen scene is based on hacks.
>>
>>101730957
How much more porn do you need? Don't you think there is enough already on the internet? More than you could ever hope to watch in your lifetime.
>>
>>101730924
Can you explain your thought process on why you think this link is relevant to local image generation?
>>
File: ComfyUI_00230_.png (816 KB, 1024x1024)
816 KB
816 KB PNG
>>
File: Flux_00210_.png (1.83 MB, 1024x1024)
1.83 MB
1.83 MB PNG
>>101730818
> interesting art is out of the question?

"Very interesting art for very smart people"
>>
>>101730976
>How much more images do you need? Don't you think there is enough already on the internet? More than you could ever hope to see in your lifetime.
See? Sometimes you want extremely specific images that don't already exist. There aren't any real life photographs of MILF with exaggerated hentai bodies or other very specific weird fetish stuff so I have to generate those myself
>>
can flux do feet
>>
File: 1722842378.png (1.02 MB, 1024x1024)
1.02 MB
1.02 MB PNG
>>
>>101730787
It actually does .. something .. not sure if it really is what it intends to be, but it does something
>>
>>101731019
yeah, placebo
>>
>>101730983
well you see, local image generation takes 1 second while he spent 40 hours. And that is where the humor comes from.

Do you really have such a stick up your ass that you don't find this humorous? At all? Are you fucked in the head from listening to people crying all day about AI?

It's fucking funny.
>>
File: FD_00683_.png (1.31 MB, 1536x1024)
1.31 MB
1.31 MB PNG
>>101730976
You don't understand, I have a very specific fetish where I need to see a new pornographic image every few seconds of the most depraved shit in order to get half a chub. Porn is very important to my identity.
>>
>>101731035
Nta, but I thought the implication was that he was lying about actually putting in 40 hours and was actually cheating with AI generation then doing manual touch up.
>>
File: Flux_00034_.png (1 MB, 904x1104)
1 MB
1 MB PNG
>>101731014
not very well. Sometimes it can show soles but good luck getting 5 toes. Any kind of leg lift turns into hand feet.
>>
>>101731042
Ow your are Epstein .. I wonder where you went, strange you went AI porn to satify yourself
>>
>>101731048
I thought the implication was he spent 40 hours on something the character passed by in 4 seconds.
>>101731035
If I was using AI to make art for a game I would spend 40 hours in gatcha generating the perfect tree so I really don't get the joke.
>>
File: 1697274287933043.png (1.43 MB, 1024x1024)
1.43 MB
1.43 MB PNG
>>
File: ComfyUI_00233_.png (861 KB, 1024x1024)
861 KB
861 KB PNG
>>
>>101730976
That makes me wonder exactly how many images have been generated now globally and how that compares to all the images that came before it. I know some anon said he had something like a terabyte of things he genned.
>>
File: FD_00684_.png (1.72 MB, 1536x1024)
1.72 MB
1.72 MB PNG
>>101731051
literally my first attempt, what are you on about?
>>
>>101731074
So boomers stand no chance anymore right? I already walked in on my dad and his friend watching AI generated garbage on youtube. It was just random pictures with an AI voice talking about made up current events.
>>
>no collage
>no bred blessings
>mindless 1girl spam
/ldg/ has fallen. all that's missing now are the avatarfags.
>>
>>101731074
what did you prompt to get that bad camera effect? pretty cool gen
>>
>>101731083
https://www.youtube.com/watch?v=c5AxDLkqZQ8

a.i. images and voice
>>
>>101731079
>side view
now try getting the soles showing on both feet.
>>
>>101731090
i miss the good collages
>>
File: ComfyUI_00234_.png (729 KB, 1024x1024)
729 KB
729 KB PNG
>>101731095
thanks, proompt is
>A grainy photo taken at night by a Nokia camcorder in the woods. It is very dark and almost nothing is visible except the immediate surroundings due to a flashlight. A BREAKING NEWS caption streams across the bottom from a news broadcast, with text saying "LEAF FALLS FROM TREE IN THE WOODS". In smaller text, another caption below reads "ARE WE NOT ALONE IN THIS UNIVERSE?". A leaf is lying on the ground in the center of the view. A red circle has been drawn on the image to highlight the grass next to the leaf, but not the leaf itself. A red arrow pointing to this circle has been drawn as well, despite being completely unnecessary. Red text is lazily drawn next to the circle that says "LEAF IN QUESTION".
>>
File: FD_00690_.png (1.58 MB, 1536x1024)
1.58 MB
1.58 MB PNG
>>101731100
>>
>>101731083
Unironically I started to consume those generated stuff too. City Pop music, dunno with other genre.
>>
>>101731126
Well there are 5 toes even if not all of them are attached. I guess I'll concede and call it a skill issue on my part.
>>
File: FD_00691_.png (2.06 MB, 1536x1024)
2.06 MB
2.06 MB PNG
>>101731126
>>
>>101731133
seen some images of statues while a.i. voice talks about stoicism in youtube shorts, a.i. gen content is pretty common
>>
File: 1711238874501263.png (1.05 MB, 1024x1024)
1.05 MB
1.05 MB PNG
the current state of the Japanese stock market:
>>
File: FD_00694_.png (1.44 MB, 1536x1024)
1.44 MB
1.44 MB PNG
>>101731138
try "bottom of her feet" instead of soles, and "spreading her toes"
>>
https://suno.com/song/d23153a6-6b9a-450c-838a-3f2e7150f030
>>
>>101731042
How is that different from wanting to get non-pornographic images? Porn bad?
>>
File: FD_00700_.png (1.67 MB, 1536x1024)
1.67 MB
1.67 MB PNG
>>101731166
porn indeed bad.
>>
>be proompting with Flux
>13 minutes for fancy prompt in Comfy at first
>suddenly drops to 7 minutes for a few gens
>now it's 20 minutes
idgi
>>
>>101731241
I had similar results until I turned off the Cuda sysmem fallback policy in the nvidia control panel. It kept dumping everything into memory too quickly when my Vram was full. Now my gens fluctuate by +/- 15 seconds instead of 5 minutes.
>>
>>101731271
>Cuda sysmem fallback policy
I haven't updated my drivers since I got my computer in 2022 so I'm pretty sure I don't even have this setting yet, kek, thanks for alerting me
>>
File: ComfyUI_00010_.png (1.09 MB, 1024x1024)
1.09 MB
1.09 MB PNG
>>
File: ComfyUI_00238_.png (803 KB, 1024x1024)
803 KB
803 KB PNG
>>
>>101731079
Do something where leg is lifted up or sole is visible. It struggles quite a bit.
>>
File: ComfyUI_00242_.png (881 KB, 1024x1024)
881 KB
881 KB PNG
>>
>>101731126
>>101731162
Ngl, that is bad.
>>101731140
This is decent, but not what it struggles with.

Anyway, it's much closer than SD ever was and it does passable feet with some loras, so... Until then, my gens are cropped or have shoes.
>>
File: ComfyUI_00245_.png (938 KB, 1024x1024)
938 KB
938 KB PNG
>>
>>101731316
Mmm thicker
>>
File: 1693771821197047.png (1.01 MB, 1024x1024)
1.01 MB
1.01 MB PNG
it's neat what you can do with technology.
>>
>>101731763
Too attractive to be Kamala. Besides you could achieve a much better result in 2 minutes on GIMP.
>>
File: FD_00709_.png (1.21 MB, 1024x1024)
1.21 MB
1.21 MB PNG
You're all breaking the LAW!
Cease your criminal ways at ONCE
>>
>>101731832
the pro of genAI is I can make a bunch of virtual communists in the back too. could shoop people in too but it's way easier to prompt to make stuff.
>>
>>101731842
Will need to wait for a Kamala LoRA. She has been axed from the dataset.
>>
>>101731852
don't think she has been axed, she was just not important till 3 weeks ago
>>
File: FD_00713_.png (894 KB, 1024x1024)
894 KB
894 KB PNG
>>101731879
She is the vice president. It knows Mike Pence
>>
File: 1720536922940919.png (1.07 MB, 1024x1024)
1.07 MB
1.07 MB PNG
>>101731852
it's passable in some gens, flux makes AMAZING Trump's though.
>>
>>101731891
Devil's advocate literally nobody cared about Kamala until recently, not even democrats. There was probably less photos all around
>>
File: 2024-08-05_00317_.png (878 KB, 1024x1024)
878 KB
878 KB PNG
>>101730787
underrated post, this actually works and makes cfg accessible ..

pic related cfg 1.0 version of
>photography, frontal view, woman wearing bikini, flat chest, no breasts, oily skin
>>
File: 2024-08-05_00318_.png (1.54 MB, 1024x1024)
1.54 MB
1.54 MB PNG
>>101731920
and this cfg 3.0, cfg mimic 1.0
>photography, frontal view, woman wearing bikini, flat chest, no breasts, oily skin, under boob
>negatives: anime, large breasts, big boobs, censored
>>
>>101731902
Yeah I have been saying, Trump is the new Emma Watson as far as imagen goes.
>>101731912
It's not the reason. Most female celebrities have been axed from the dataset. This is how they achieve safety.
>>
>>101731920
*also had under boob, this just doesn't work .. just forgot to copy that part of the prompt

both same seed .. well DynamicThresholding seems to make cfg of up 3-4 accessible, and by that negative prompts .. won't uncensor the model, but it will make some things tricks possible that were not without negative prompts
>>
what should fluxguidance be? it's 3.5 by default
>>
>>101731920
>>101731931
I think the use cases are very limited and it's not worth the performance hit.
>>
>>101732168
2nd example. I think one of these is objectively better than the other, and it's not the one I used a negative prompt on.
>>
>>101731347
why does it have not one, but two cocks?
>>
File: 2024-08-05_00338_.png (2.89 MB, 1536x1536)
2.89 MB
2.89 MB PNG
>>101732168
I agree .. still its a legit way to get negatives working. It was nearly impossible to get flat chested adult women (I dont want to gen kids!) Ill keep it in my useful workflows for special cases.
>>
I dunno, maybe I just like the soft look. Cranking and normalizing the cfg makes it seem like its obviously AI.
>>
>>101732296
Wouldn't use it with no negatives in mind, then it just changes the look for .. well subjectively. I will test it with different use cases tho. And it/s drop to about half the value.
>>
>>101732391
I haven't come across a situation where I have wished for negatives to remove something for an image. If I describe the image idea well enough there's only what I request in it.
>>
so, what is the default cfg? my workflow has no cfg node. just fluxguidance, but that isnt cfg.
>>
>>101729837
Mrs. Obama, GET DOWN!
>>
File: 2024-08-05_00340_.png (3.41 MB, 1536x1536)
3.41 MB
3.41 MB PNG
>>101732439
cfg == 1 unless you use normalizing witth custom DynamicThresholding node
>>
>>101732471
>>101731920
>>101731931

WHY DO THEY ALL HAVE SHIT IN THEIR BELLY BUTTON?
>>
File: FD_00746_.png (1014 KB, 1024x1024)
1014 KB
1014 KB PNG
>>101730285
>What a time to be alive
>>
File: 2024-08-05_00354_.png (3.01 MB, 1536x1536)
3.01 MB
3.01 MB PNG
>>101732605
kek.. ask FLUX
>>
>>101732605
show your clean belly button.
I just pulled a ball of lint out of mine
>>
File: ComfyUI_Flux_3269.jpg (179 KB, 1024x768)
179 KB
179 KB JPG
>>
File: coomer-2.png (17 KB, 250x208)
17 KB
17 KB PNG
>>101730827
>>101730957
>>101731042
>>101731166
Its possible to gen using semi real pony, route the decoded image into a vae encoder and apply the flux vae and send the latent into ksampler connected to flux and use a low denoise like 0.3 and enhance that porn. But its hit and miss and anything higher than 0.3 denoise strength will destroy the image.

But when it works it looks amazingly real.
>>
File: Flux_00238_.png (1.34 MB, 1024x1024)
1.34 MB
1.34 MB PNG
>>101729425
still possible but nowhere near as horny as de3
>>
File: ComfyUI_00553_.png (2.75 MB, 1600x1024)
2.75 MB
2.75 MB PNG
>>
File: ComfyUI_00716_.png (1.34 MB, 1024x1024)
1.34 MB
1.34 MB PNG
Messing around with detection nodes, i got bored of flux for now.
>>
How to prompt a comic book cover with title?
>>
>>101732815
prompt?
>>
>>101732845
>comic book cover with title

There you go
>>
>>101729926
very creative
>>
>>101732893
kind of crappy first attempt

T5 box:
Dark Screen capture from the 1980 cult film, laser disc quality: envision a scene of a Japanese woman squat amidst an overgrown patch of wildflowers. She has red colored rope bound tightly around her hands and body restraining her suggestively. Her hands tied behind her back. Knees bound clasped together. The red colored rope is like Japanese shibari binding. She poses in a ritualistic scene. The dark soil and the overgrown wildflowers, bathed under full moonlight, frame this intimate moment perfectly. Ensure her wrists are bound with red rope.

Clip box:
Dark Screen capture from the 1980 cult film, laser disc quality
>>
File: ComfyUI_00718_.png (1.28 MB, 1024x1024)
1.28 MB
1.28 MB PNG
I need to figure out the best ordering or detailing people, as in face, hands, person. Currently i'm doing it like this person > face > hands. Plus I need to figure out the best settings for setting of the mask
>>
>>101732778
show some examples
>>
File: 2024-08-05_00387_.png (3.79 MB, 1536x1536)
3.79 MB
3.79 MB PNG
>>101732981
thx okay I see.. doulbe and tripple explain it to t5, Ill try it once my q is finished
>>
>>101732992
Imagine trying to navigate those tables.
>>
I'm gonna have to replace my 3060ti with a 3090/4090 to keep up with you fuckers, aren't I?
>>
>>101733021
yea and then you bite your ass cause the 5090 is due this winter
>>
>>
>>101732997
I can't I'd get banned.

But follow these steps, gen as normal with SDXL whatever, route the latent that comes out of ksampler into VAE decoder which is connected to a SDXL VAE model, take the resulting latent and route into a VAE encoder connected to flux VAE, then use that latent image for your flux ksampler input. Just sure to use low denoise strength, it will treat the latent as 100% masked, feed in the same prompts or use different prompts. Works best if the image is already or close to photo realism. I was messing around with it 2 days ago, I don't have the workflow to hand. However its a simple as converting SD 1.5 images into SDXL by routing and up-scaling which you won't need to do here since 1024 x 1024 will be fine for both.

If you don't have tons of system ram, be sure you have plenty of swap space and expect a lot of swapping which is slow but it gets the job done.
>>
>>
>>101733012
imagine thinking the AI understands this.
>>
>>101733056
Post a catbox damn it. Gen something tamer if you're worried we'll judge your fetish, which we will.
>>
File: FD_00757_.png (1.11 MB, 1024x1024)
1.11 MB
1.11 MB PNG
>>101732845
Picrel. Literally just describe what you want to see, it's that easy.
>>101733056
You can share the catbox.
I have already done the reverse, gen images in Flux and run a pony pass to correct the nipples and shit, but not the other way.
>>
File: 2024-08-05_00399_.png (1.76 MB, 1536x1536)
1.76 MB
1.76 MB PNG
>>101732981
yaa. its some sort of rope bondage alright.. but its far from the artistic exact shibari I am looking for.. thanks anyhow, until loras for this are available ill just be creative
>>
File: Flux_00256_.png (1.87 MB, 768x1368)
1.87 MB
1.87 MB PNG
>>101732845
>>
>>101733078
can I be arsed? Well, maybe this once, but only because I'd like to test the detailer I have setup and feed the result into flux to see what happens, such as this image
>>101732992
I can also see other uses like getting a certain style or character into flux and enhancing it, so yeah. Wait out.
>>
>>101733108
get it consistent or just 1/5 times?
>>
>>101733096
>I have already done the reverse, gen images in Flux and run a pony pass to correct the nipples and shit, but not the other way.
I've done that but then it loses the quality overall, I guess you are using detector? Or are you manually inpainting? Fuck that man can't be arsed with that.
>>
>>101733126
>>
>>101733137
prompt it up
>>
>>101733130
Neither, just a low denoising pass in pony. It does lose quality but it sticks on proper nipples. Flux knows where the nipples go, unlike SD3, so it's easier to make it look less shit.
>>
File: FD_00762_.png (1.97 MB, 1024x1536)
1.97 MB
1.97 MB PNG
>>101733146
comic book cover with title "BRAINLET"
The image is a comic book illustration of a man with a glass dome in place of the top of his skull, inside is a tiny brain the size of a walnut.
On the top right is a sticker with the text "$1"
On the bottom right is the Marvel Comics logo.
>>
File: file.png (2.6 MB, 1024x1024)
2.6 MB
2.6 MB PNG
seems like flux is susceptible to overloading it with schizo symbolism in the prompt, you just need a lot more of it before the chaos begins to emerge compared to SD
>>
>>101733158
slow it down there buddy, maybe a flowchart would help
>>
File: 2024-08-05_00407_.png (3.58 MB, 1536x1536)
3.58 MB
3.58 MB PNG
>>
File: 1722856872_S1.png (3 MB, 1024x1024)
3 MB
3 MB PNG
>>101732815
>>101732981
same exact prompt on de3 for reference...
>>
>>101733158
thanks
>>
>>101733188
seems de3 def has more rope data, kek
>>
File: ComfyUI_Flux_3391.jpg (162 KB, 1024x768)
162 KB
162 KB JPG
>>
Bread

>>101733170
>>101733170
>>101733170
>>
>>101733216
MODS
>>
File: FD_00766_.png (2.21 MB, 1024x1536)
2.21 MB
2.21 MB PNG
>>101733176
https://files.catbox.moe/qggnqx.png
>>
>>101733231
nice
>>
>>101733150
its difficult, i'm getting my head messed up trying to correctly encode it, crashed twice, wish I saved the workflow, I'll figure it out...
>>
>>101733150
grrr it does not want to play, perhaps i need a normal SDXL VAE and not the model baked in? I know I have to convert the image into a latent that works with flux but some reason its not letting me.
>>
File: 1716921191142138.png (552 KB, 1312x736)
552 KB
552 KB PNG
>>
>>101733338
>>101733297
Ah I'm retarded, oh course its the clip most probably, i can't reuse the clip from SDXL, the image processes fine into a preview window once decoded with the flux vae but it fails in ksampler because I'm feeding it the positive prompt encoded from a none flux clip model
>>
File: 1720178890864360.png (531 KB, 1312x736)
531 KB
531 KB PNG
>>
How fast is flux on a 3090?
>>
>>101733668
if you move your display to another GPU or to iGPU then the whole model fits
>>
>>101730985
kek
>>
>>101733749
Why is it so hard to buy in europe ugh
>>
>>101733164
How long is your prompt? I was getting quite nice results with 2000-2500 characters from a LLM.
>>
Fuck I want to gen so much, but I've been genning nonstop for 2 days and now my deadlines are overdue.
>>
>>101732992
>PlateUP
>>
>>101734481
So with Flux I'm supposed to do long winded prose in the prompt? Which LLMs would you recommend?
>>
>>101734617
I only tried https://huggingface.co/cognitivecomputations/dolphin-2.9.3-mistral-nemo-12b-gguf/tree/main
It's uncensored so you can prompt whatever you want, Q6_K model completely fits on 3060 12GB and I get 12+ tokens/second.
This only took 33 seconds to generate https://files.catbox.moe/m9dnm2.png
And here's the flux output with that prompt https://files.catbox.moe/fnztdb.png
>>
>>101734510
What are you generating?
>>
>>101734997
You should use LM Studio or jan to get better outputs.
>>
>>101735026
Thanks, I'll try it.

>>101734999
>>101734997
>>
>>101730537
Giant stinky asshole is not even a booru tag
>>
>>101730976
What if there's no porn of the person you want porn of
>>
>OP img deleted
imagine seething so hard you mass report a SFW gen



[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.