[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]


General dedicated to local usage of free and open source text-to-image models

Previous /ldg/ bread : >>101092621

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

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

>Auto1111 forks
SD.Next: https://github.com/vladmandic/automatic
Anapnoe UX: https://github.com/anapnoe/stable-diffusion-webui-ux

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

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

>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

>Animation
https://rentry.org/AnimAnon
https://rentry.org/AnimAnon-AnimDiff
https://rentry.org/AnimAnon-Deforum

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

>View and submit GPU performance data
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

>Share image prompt info
https://rentry.org/hdgcb
https://catbox.moe

>Related boards
>>>/h/hdg
>>>/e/edg
>>>/d/ddg
>>>/b/degen
>>>/vt/vtai
>>>/aco/sdg
>>>/trash/sdg
>>
Blessed thread
>>
>>101108471
*of frenship
>>
what is local SOTA for removing backgrounds that can deal with images of dozens of MB? i dont need anything complex, only removing backgrounds of my scanned documents
>>
>>
File: UI_0004.jpg (791 KB, 1024x1536)
791 KB
791 KB JPG
>>
>>101108542
If you're still generating then layered diffusion might be it. Else probably rmbg or rembg with one of the models.
>>
Found this repo: https://github.com/DenOfEquity/Hunyuan-DiT-for-webUI
>>
>>101108647
interesting and unique gen, nice lips
>>
>>
File: UI_0007.jpg (808 KB, 1024x1536)
808 KB
808 KB JPG
>>101108726
Thanks, I might post more of her.
>>
>>101108705
Interesting, I wonder if inpainting works
>>
File: cookiesi~5.jpg (209 KB, 1304x1304)
209 KB
209 KB JPG
>>
File: 0.jpg (344 KB, 1024x1024)
344 KB
344 KB JPG
>>
How good is the anime IPAdapter? Still trying to create LoRAs for OCs for animations. So far I've tried training LoRAs on 1 or 2 images, then using them to make more images with a lot of inpainting that takes forever, then making a new LoRA with those.
>>
>>101108459
Nice img
>>
>>
>>101109596
>>
>>101109603
>>
>>101109441
Thanks
>>
hunyuan looks gooood
>>
More pixel art?
>>
I saw that two people from the pixart team have joined nvidia?
>>
File: UI_0013.jpg (793 KB, 1024x1536)
793 KB
793 KB JPG
>>
>>101110020
apparently but also continuing pixart
>>
New captioning models in taggui, I think they're worth a look:
Florence-2
Internlm-xcomposer2-vl-1_8b

https://github.com/jhc13/taggui/releases/tag/v1.29.0
>>
File: 142343565467.png (4 KB, 406x268)
4 KB
4 KB PNG
My FREINDSHIP ENDED with DIRECTML

Now ZLUDA is my BEST FRIEND

Before i was waiting up to 5 minutes with 25 steps and 10 highrez steps.

Now i can do 75 steps and 55 highrez steps in a SINGLE minute.
>>
Catbox workflow suitable for Pixarting on 16GB System Ram / 6GB VRAM (assuming GPU is recentish Nvidia, like Turing plus, e.g. a GTX 1660 Super or what have you).

Requires you to have PixArt-Sigma-XL-2-1024-MS.pth in your Comfy checkpoints folder, t5xxl_fp8_e4m3fn.safetensors in your Comfy clip folder, and sdxl.vae.safetensors in your Comfy VAE folder. Also will need the "Extra Models For ComfyUI" package from Comfy Manager installed, for a couple of the loader nodes.

https://files.catbox.moe/dkjvnk.json
>>
>>101110384
>sdxl vae
Do you find this better than pixarts vae?
>>
>>101110403
aren't they the same?
>>
>>101110403
They are literally the same thing, no observable difference exists between them lol. You can confirm this for yourself by just switching Load VAE between SDXL.vae.safetensors and the diffusion_pytorch_model.safetensors Pixart provides.
>>
>>101110271
happy for 4 anon
>>
>>101110592
or sorry that that happened
>>
>>101109240
Ipadapter blends both the style and concept of the source image with your prompt. Whether it's good or not depends on your use case.
>>
>>101109240
Using 2 ip-adapters at once is pretty powerful. there's the plus version and lite version that I sometimes use together.
>>
File: 00003-3956984557.jpg (186 KB, 1440x1200)
186 KB
186 KB JPG
>>101109641
NICE
>>
File: 04932.png (2.53 MB, 1920x1080)
2.53 MB
2.53 MB PNG
dear /ldg/,

I am GAY.

I'm sorry to say, but i had to admit it! I just couldn't conceal my love for cock. I'm sure it has nothing to do with anything. So, enjoy!
>>
File: 1697697162462071.jpg (1.58 MB, 3024x1728)
1.58 MB
1.58 MB JPG
>>
File: 04835.png (3.36 MB, 1920x1080)
3.36 MB
3.36 MB PNG
little known fact: there are more turds in the Ganges than you have souls to contend my status as Local God.
>>
>>101111349
as long as you're not a troon that's all right kek
>>
>>
>>
>>101111480
>>101111494
fake religion.
you are gay
>>
File: long dick general.jpg (2.56 MB, 3264x3264)
2.56 MB
2.56 MB JPG
>>101108459
>>
>>101111503
lord heavens, such a thing is not allowed in our bible group, anon
>>
father, i might have sinned
>>
>>
>>101111761
held up by prayer and double sided tape
>>
File: tmpn3o0gsff.png (1.03 MB, 1344x768)
1.03 MB
1.03 MB PNG
>>
File: 00034.jpg (415 KB, 2304x3072)
415 KB
415 KB JPG
>>
File: 00035.jpg (853 KB, 2304x3072)
853 KB
853 KB JPG
>>
>>101111580
>he made the collage
>>
File: 00036.jpg (680 KB, 2304x3072)
680 KB
680 KB JPG
>>
File: 1697080161400484.jpg (1.59 MB, 3024x1728)
1.59 MB
1.59 MB JPG
>>
File: 00142.png (2.27 MB, 1536x1536)
2.27 MB
2.27 MB PNG
Somehow inpaint with the same prompt works better than adetailer on the same image here.
>>
File: 1692801997576505.jpg (1.4 MB, 3024x1728)
1.4 MB
1.4 MB JPG
>>
>>101111761
the boobies and fabric often dont match up perfectly, sadge
>>
>>101113022
awesome cuteness, prompt?
>>
File: tmp6zui6bpc.png (1.95 MB, 1680x960)
1.95 MB
1.95 MB PNG
>>
>>
The people who make Juggernaut are officially doing a PixArt finetune as their next large project
>>
File: tmpm_m4aze0.png (1.85 MB, 1680x960)
1.85 MB
1.85 MB PNG
>>101113662
Not a fun of juggernaut, but more hands on deck are always welcome. Wonder how differently their style will develop.
>>
>>
Can I run Hunyuan DIT on a 3060 12GB? What about lora training?
>>
File: 1716273289598186.jpg (938 KB, 3024x1728)
938 KB
938 KB JPG
>>101113842
cool style
>>
>>101113951
Sorry also wanted to confirm - do English prompts work or do you guys google translate Chinese? If the latter, can anyone show what their prompts look like in English before translation?
>>
>>101113951
I'm not that familiar with it, but their github says minimum of 11 VRAM. Anons mentioned a distilled, less demanding version on the way SoonTM. It also seems to be bilingual indeed.
>>
Anyone got ideas on how to go about an artstyle similar to picrel, and the likes of ~WW1 propaganda? Be they prompts or loras for a Pony model.
>>
File: 1701451600549709.jpg (1.39 MB, 3024x1728)
1.39 MB
1.39 MB JPG
>>
>>101114040
Thanks anon! Thought I'd seen some comments of others using it with less but I might be confusing it with something else, I'm not up to date on the alt models. I'll have to wait for the lower req version.
>>
>>101114321
There's also Pixart, a much smaller and less vram hungry model, that's apparently is easier to train than SD 1.5 was. Just as Hunyuan is about to get a smaller model, team behind Pixart is apparently going to work with Nvidia on a bigger, more complex one.

I also recall someone claiming to run it will less, so maybe there's something we don't know about. Lurk around, someone might be able to elaborate more on the matter. Maybe >>101109596 can share some insight?
>>
>>101114442
Honestly I was a bit worried that the Nvidia buy out might mean a switch in direction for pixart, but if the current model is friendly to us vramlets I might as well check it out. Thanks again!
>>
>>101114503
If you do end up wanting to try out Pixart, for all I know you can do so with either ComfuUI with that Comfy Nodes addition, or through SD.Next branch of Auto1111. StableSwarm might also be able to do it, since it's based on Comfy.
>>
you can tell me your sin's anon
>>
File: tmp48g50co4.jpg (1.04 MB, 2352x1344)
1.04 MB
1.04 MB JPG
>>101114627
>you can tell me your sin's anon
I've downloaded more loras than I can chew through. Your meanwhile is generating too much of the same. Come on, gimme something for the next collage.
>>
File: 1709859128273090.jpg (1.06 MB, 3024x1728)
1.06 MB
1.06 MB JPG
>>101114627
based nun prompter
>>
File: tmp4h90x4q9.png (107 KB, 499x500)
107 KB
107 KB PNG
>>
>>101114627
I've been watching Outlaw Star instead doing work. I pee in the shower.
>>
is the guy who did the "manga painting" last thread still around? >>101095831 is there a specific style prompt to get that style? it's exactly what I've always wanted but sdxl never took to it well, even with loras
>>
File: ComfyUI_00127_.jpg (1.08 MB, 2147x2147)
1.08 MB
1.08 MB JPG
>>
>>101114794
>I pee in the shower
doesn't the hot water just steam up the piss and make your shower/bathroom/you smell like piss? why in the ever loving fuck would you do this... true jeet behavior, enough to make sloppamixes on Civit proud
>>
File: 116775249263548834-SD.png (2.23 MB, 984x1456)
2.23 MB
2.23 MB PNG
hello /ldg/
how we doing this fine weekend?
>>
File: 1710327304635885.jpg (1.36 MB, 3024x1728)
1.36 MB
1.36 MB JPG
>>
>>101115014
Cold, cloudy and wet around where I live. Perfect conditions to stay inside and gen.
>>
File: 116775249263548837-SD.png (2.34 MB, 1088x1400)
2.34 MB
2.34 MB PNG
>>101115068
May the warmth of your GPU keep you cozy
>>
>>101115111
I work on a laptop, so I unironically love to warm up my hands on it.
>>
File: 116775249263548840-SD.png (2.07 MB, 1016x1400)
2.07 MB
2.07 MB PNG
>>101115142
anthing to keep the cold away!
>>
File: 1700835716844385.jpg (1.32 MB, 3024x1728)
1.32 MB
1.32 MB JPG
>>
File: 116775249263548844-SD.png (2.04 MB, 968x1400)
2.04 MB
2.04 MB PNG
>>101115260
>>
I've been out of this stuff for a while. Is SD3 any good? I heard it was censored
>>
>>101115594
SD3 is horrible, and the licence is so dogshit civitai has banned the model kek
>>
File: 0.jpg (414 KB, 1024x1024)
414 KB
414 KB JPG
>>
>>101115594
This is the thread for non-SAI based models. You're looking for /sdg/
>>
>>101115716
/ldg/ is for all local image generation models
>>
>>101115716
It's for any locally generated works using any model or workflow.
>>
>>101115716
what he said >>101115730

also what non-SAI-based models even exist for local image generation? virtually everything is based on SD 1.5 or SDXL
>>
>>101115730
>>101115741
>>101115751
Uh, no. It's focused on the non-cucked models, PixArt and Hunyuan etc. /sdg/ is focused on the SAI models, XL, Pony, etc.

That's the entire reason we needed two generals.
>>
>>101115751
Read the very top of this thread. Slowly and carefully. Come back with any questions.
>>
>>101115758
wrong
everyone using local models is welcome here SAI based or not
we just dont like drama an shills gatekeeping
>>
>>101115730
>>101115741
>>101115751
Please read the thread OP and head back to /sdg/. Probably my own stupid fault for biting and replying to the bait here, but oh well
>>
File: ComfyUI_00140_.jpg (1016 KB, 1996x1996)
1016 KB
1016 KB JPG
>>101114867
>>
>>101115758
>non-cucked models, PixArt and Hunyuan etc
what do you mean? PixArt and Hunyuan are very cucked, you'd think a chink model would have no censorship at all?
>>
>>101115786
>Pixart Sigma & Hunyuan DIT
>listed immediately below a whole bunch of frontends
>literally no other models are listed
>I'm supposed to assume these are models and not more frontends
retard
>>
>>101115808
They are both open source and have nothing to do with China. Code isn't ethnic, it's fucking code you concern trolling piece of human garbage
>>
>>101115803
Go call your lawyers. Everyone using local is welcome. Please post something interesting. Drama belongs in /sdg and /ic.
>>
>>101115839
First, go and read the thread OP then come back and suck my dick in front of the thread, okay?
>>
>>101115833
you fucking retarded nigger, I was talking about the model on itself, who gives a fuck about the code? SD3 already has "code" as it can be already be used on ComfyUI and A1111, not the same can be said about the chink models
>>
>>101115859
Fucking absolute cockmongling piece of sub-80 IQ fucking garbage
How can a fucking set of numbers be associated with a country? Learn a single fucking thing about the space you're posting in before you open your retarded as fuck feminine ass mouth
>>
Schizo anon needs to up his game. Maybe some more practice on /sdg would help.
>>
File: 116775249263548863-SD.png (2.21 MB, 1128x1400)
2.21 MB
2.21 MB PNG
>>101115464
>>101115594
Shit model
>>101115692
1ball
>>101115716
not this shit again
>>
>>101115869
shut the fuck up you sub zero IQ piece of shit, nothing you say makes sense
>>
I don't give a fuck what you say, PixArt and Hunyuan are straight Chinese spyware
>>
>>101115888 (checked)
A'ight man, fuck 'em
>>
You're getting mad over the skin color of the person who coded your for loops, brothers. SAI is a fucking joke and its models belong elsewhere. PixArt is the new model in town and is getting finetunes from all of the big names.
>>
>>101115888
>Chinese spyware
What are they going to do? Hijack the google IP detector van that takes pictures of your house?
>>
File: ComfyUI_00145_.jpg (664 KB, 1024x1024)
664 KB
664 KB JPG
itt
>>
>>101115971
>Loud Diffusion General
>>
File: UI_0001.jpg (1.15 MB, 1024x1536)
1.15 MB
1.15 MB JPG
>>
File: 00009-157592537.jpg (421 KB, 1536x2560)
421 KB
421 KB JPG
>>101116049
apart from the building looking things on top of the cliff, looks quite real to me
>>
Good afternoon
>>
File: UI_0002.jpg (1.12 MB, 1024x1536)
1.12 MB
1.12 MB JPG
>>101116072
I agree with you. No matter what model I use it still has that Math structures like https://www.nasa.gov/technology/goddard-tech/nasa-turns-to-ai-to-design-mission-hardware/
>>
File: 116775249263548854-SD.png (2.41 MB, 1176x1448)
2.41 MB
2.41 MB PNG
>>101115877
>>
File: UI_0003.jpg (1.22 MB, 1024x1536)
1.22 MB
1.22 MB JPG
>>101116072
>>
File: tmpmzmlg9ob.png (120 KB, 400x400)
120 KB
120 KB PNG
Please ignore blatant shitposts trying to ragebait you into arguing with someone that doesn't even want to have a genuine discussion. If they can't keep it civil, chances are they're not here to argue in good faith to begin with.
>>
>>101116171
I think it's schizo anon, he's harassing /sdg too.
>>
File: UI_0005.jpg (1.05 MB, 1024x1536)
1.05 MB
1.05 MB JPG
>>101116158
I like this one. But the dress on >>101115877
is better
>>
>>101115464
what is the recipe for this?
>>
>>101116182
I don't know, and I don't care. The less attention is paid to shit's worth posts, the better. There's plenty more worthwhile things to do, as well as better posts to be read and be written. Take care folks.
>>
File: PixArt_00017_.png (1.55 MB, 1024x1024)
1.55 MB
1.55 MB PNG
>>101116171
There is no point in talking to no-gens. They also don't know what they are talking about and don't want to listen to what you have to say anyway.
Plus a reminder to anyone on the internet. No one in history of it ever said. "Oh, you make a valid point. I wasn't aware of that. You just changed my mind on the subject."
>>
>>101116250
>https://www.nasa.gov/technology/goddard-tech/nasa-turns-to-ai-to-design-mission-hardware/
Good point, hadn't thought of that.
>>
>>101116250
No genning isn't the issue, but the content or lack thereof in a post.
>>
File: UI_0007.jpg (884 KB, 1024x1536)
884 KB
884 KB JPG
>>101116281
Image just adds an extra to the point you're trying to make. But that's just my honest opinion.
>>
File: UI_0008.jpg (873 KB, 1024x1536)
873 KB
873 KB JPG
>>101116301
>>
File: 116775249263548856-SD.png (2.27 MB, 1168x1448)
2.27 MB
2.27 MB PNG
>>101116158

>>101116216
agree

>>101116227
1 korean female

>>101116301
ginger, green eyes next pls
>>
File: UI_0012.jpg (1.07 MB, 1024x1536)
1.07 MB
1.07 MB JPG
>>101116336
But I just switched back to architecture.
>>
File: UI_0014.jpg (934 KB, 1024x1536)
934 KB
934 KB JPG
>>101116336
>>
>>101116270
>https://www.nasa.gov/technology/goddard-tech/nasa-turns-to-ai-to-design-mission-hardware/
They have a fun way of writing nothing.
What i take away is that they have revolutionized FEM until it doesnt work because AI lol.
>>
File: 116775249263548859-SD.png (2.64 MB, 1176x1448)
2.64 MB
2.64 MB PNG
>>101116336
>>101116400
noice!
>>
>>101116418
Few years back I've read how it all started with a cast/parts where it would use half of material but would remain as strong or sometimes stronger than the fully fledged part. Making it more efficient. There hasn't been anything manufactured like that since and the only recent crap I've seen was 3d printing houses that use that type of building.
>>
File: UI_0017.jpg (1.09 MB, 1024x1536)
1.09 MB
1.09 MB JPG
>>101116336
>>
File: 00032-48643116.png (2.15 MB, 1024x1536)
2.15 MB
2.15 MB PNG
>>101116433
Nice!
>>
>>101116336
>1 korean female
I tried to attach one to a comfyui node and it got upset. can you share catbox so I can see how it's done?
>>
>>
File: OIL_01.jpg (533 KB, 1200x1200)
533 KB
533 KB JPG
>>101116418
>>101116464
Also imagine having to machine that crap, for space it makes sense but most everything else, too expensive i think.
>>
File: 116775249263548733-SD.png (2.51 MB, 1120x1384)
2.51 MB
2.51 MB PNG
>>101116433


>>101116483
I like this better!

>>101116497
its a very simple ponyrealism gen anon

 score_9, score_8_up, score_7_up, 1girl, asian, close-up, extremely detailed, eyeshadow, parted lips, split bangs, long hair, semi-rimless eyewear, looking at viewer 
>>
File: UI_0023.jpg (840 KB, 1024x1536)
840 KB
840 KB JPG
>>101116570
The starting molds probably hell of expensive and if there are any flaws the part will be more fragile.
>>
File: 116775249263548703-SD.png (2.3 MB, 1120x1384)
2.3 MB
2.3 MB PNG
>>101116573

>>101116489
thanks anon, I like your girl too!
>>
File: UI_0025.jpg (917 KB, 1024x1536)
917 KB
917 KB JPG
>>101116573
checks out.
>>
File: 00001-4152982454.jpg (253 KB, 1536x2304)
253 KB
253 KB JPG
>>
>>101116573
>ponyrealism
thanks, that's unexpected
so far most things I got from ponyrealism were grotesque uncanny valley 2.8D doll-like creatures. will try more.
>>
File: 00097-1816332804.png (1.88 MB, 1024x1536)
1.88 MB
1.88 MB PNG
>>101116573
Are you adding the noise as an extra? How would it look without?
>>
>>
File: UI_0026.jpg (921 KB, 1024x1536)
921 KB
921 KB JPG
>>101116637
With my negs added.
>>
File: UI_0027.jpg (904 KB, 1024x1536)
904 KB
904 KB JPG
>>101116637
>>101116673
>>101116655
without adding noise
>>
>>101116655
Have you made a switch to Comfy yet?
>>
File: 116775249263548862-SD.png (2.34 MB, 976x1448)
2.34 MB
2.34 MB PNG
>>101116620

>>101116637
>>101116673
its all RNGenerating hehe

>>101116647
pony is the shizzz

>>101116655
yea, I use haku-img extension for some quick noise
>>
File: 00030-2556714334.png (1.86 MB, 1024x1536)
1.86 MB
1.86 MB PNG
>>101116737
I tried to run it today, didn't like the gens :( so hopped back to a1111
>>
>>101116758

>>101116655
no noise version
>>
File: 00003-4152982454.jpg (339 KB, 1536x2304)
339 KB
339 KB JPG
>>
>>101116786
>>101116693
Thanks both look nice (with and without)
>>
File: OIL_02.jpg (459 KB, 1100x1100)
459 KB
459 KB JPG
>>
File: ComfyUI_09686_.png (1.71 MB, 1080x1920)
1.71 MB
1.71 MB PNG
>>
File: ComfyUI_temp_irque_00050_.png (2.02 MB, 1120x1440)
2.02 MB
2.02 MB PNG
>>
>>
File: 116775249263548866-SD.png (1.57 MB, 1464x784)
1.57 MB
1.57 MB PNG
>>101116758
>>101116786
switching up the aspect ratios
>>
>>101116851
>>101116871
are you done spamming sdg fucktard? is it our turn to suffer?
>>
>>101116851
>>101116871
Stay in /sdg/ debo. Dont make me get the pastebin
>>
>>
File: ComfyUI_temp_irque_00057_.png (1.98 MB, 1120x1440)
1.98 MB
1.98 MB PNG
>>101116886
>>101116892
wtf? stop following me lol
>>
File: 00004-1813854976.jpg (327 KB, 1536x2304)
327 KB
327 KB JPG
>>
>>101116892
It's Pride Month, let her have her little celebration.
>>
>>
File: Vix_0001.jpg (892 KB, 1024x1536)
892 KB
892 KB JPG
>>101116952
When is Envy Month?
>>
>>
File: 116775249263548868-SD.png (1.67 MB, 1000x1000)
1.67 MB
1.67 MB PNG
>>101116873

>>101116893
>>101116921
>>101116978
>>101117001
what the shit is a debo?

>>101116985
smaller boobage then perfection!
>>
>1girl posters moved to /ldg/
you guys ruin everything you touch
>>
File: OIL_03.jpg (476 KB, 1100x1100)
476 KB
476 KB JPG
>>101117050
>>
>>101117050
nogen
>>
File: 00037.jpg (1.92 MB, 2304x3456)
1.92 MB
1.92 MB JPG
>>
File: UI_0038.jpg (845 KB, 1024x1536)
845 KB
845 KB JPG
>>101117028
Messing about with various prompts to get a feel for negs
>>
>>101116985
Can you do smaller boobs with high heels?
>>
>>101117164
She is wearing high heels.
>>
File: OIL_05.jpg (452 KB, 1200x1200)
452 KB
452 KB JPG
>>
File: ComfyUI_temp_irque_00070_.png (2.08 MB, 1120x1440)
2.08 MB
2.08 MB PNG
>>
File: 116775249263548873-SD.png (2.49 MB, 1040x1592)
2.49 MB
2.49 MB PNG
>>101117028


>>101117125
this looks great!
I really like the eyes on this!
what are you prompting for the makeup?
>>
File: 01410-467284974.jpg (333 KB, 1075x1613)
333 KB
333 KB JPG
>>
File: 00103-3969384546.png (2.24 MB, 1024x1536)
2.24 MB
2.24 MB PNG
>>
File: UI_0040.jpg (874 KB, 1024x1536)
874 KB
874 KB JPG
>>101117164
Sure, I'll reduce the size
>>
File: ComfyUI_temp_irque_00071_.png (2.06 MB, 1120x1440)
2.06 MB
2.06 MB PNG
>>
>>101117274
nice lara, what model is that, any lora?
>>
File: UI_0041.jpg (862 KB, 1024x1536)
862 KB
862 KB JPG
>>101117164
>>
File: OIL_06.jpg (501 KB, 1300x1107)
501 KB
501 KB JPG
>>
File: UI_0042.jpg (916 KB, 1024x1536)
916 KB
916 KB JPG
>>101117217
Not a full prompt but that's where she gets the make up from.
defined jawline and dark cheeks showcase her defiant and strong personality. Her striking, smoky eye makeup adds a sense of mystique to her image, and her delicate skin appears grainy and real, showcasing the intricate jewelry on her face.
>>
Sunday
>>
>>101116873
wtf this is ponyrealism I just deleted that because I thought it was shit, I guess I'm the one that was with shit prompts lol
>>
File: UI_0043.jpg (921 KB, 1024x1536)
921 KB
921 KB JPG
>>101117358
Full prompt back at you.
 score_9, score_8_up, score_7_up, Payton_Presslee, close-up, extremely detailed, eyeshadow, parted lips, split bangs, long hair, semi-rimless eyewear, looking at viewer, defined jawline and dark cheeks showcase her defiant and strong personality. Her striking, smoky eye makeup adds a sense of mystique to her image, and her delicate skin appears grainy and real, showcasing the intricate jewelry on her face.
>>
>>101117395
good gen, but
>glasses to negatives
>glasses to positives
pick 1
>>
>>
File: UI_0044.jpg (1.13 MB, 1024x1536)
1.13 MB
1.13 MB JPG
>>101117415
I like that style of glasses
>>
File: ComfyUI_temp_irque_00076_.png (2.07 MB, 1120x1440)
2.07 MB
2.07 MB PNG
>>101117290
that was leosamhelloworld
>>
File: UI_0045.jpg (1.11 MB, 1024x1536)
1.11 MB
1.11 MB JPG
>>101117395
>>101117415
and my neg prompt is just
score_6, score_5, score_4, pubic hair, pony, muscular, censored, furry, child, kid, chibi, holding, patreon logo, watermark, text, initials, pony_source, furry_source,
>>
anyone know any good sdxl regularisation images for realistic woman face+full body training?
>>
>>101117320
Whale, whale, whale.
>>
File: 116775249263548226-SD.png (1.75 MB, 896x1152)
1.75 MB
1.75 MB PNG
>>101117217

>>101117358
wow ok, Ive never gone full promptsmithing, guess I have to now!
>>101117395
thanks!

>>101117370
Pony is good, some nitpicking is there, but its worth the effort

>>101117446
beautiful style
>>
File: PA_00018_.png (1.31 MB, 1280x768)
1.31 MB
1.31 MB PNG
>>101117491
PixArt for Face training.
>>
File: OIL_07.jpg (498 KB, 1300x1107)
498 KB
498 KB JPG
>>101117492
Trash, trash, trash
>>
File: PA_00032_.png (1.41 MB, 1280x768)
1.41 MB
1.41 MB PNG
>>101117504
How do you get that style of dress every time?
>>
>>101117529
slit dress, plunging neckline usually works
>>
File: ComfyUI_temp_irque_00080_.png (1.9 MB, 1120x1440)
1.9 MB
1.9 MB PNG
>>
File: grid-0407.jpg (394 KB, 1536x2304)
394 KB
394 KB JPG
>>
File: ComfyUI_temp_irque_00081_.png (1.88 MB, 1120x1440)
1.88 MB
1.88 MB PNG
>>
File: UI_0050.jpg (876 KB, 1024x1536)
876 KB
876 KB JPG
>>101117539
That just gives her a lab coat
>>
File: UI_0051.jpg (887 KB, 1024x1536)
887 KB
887 KB JPG
>>101117623
>>
File: 116775249263548877-SD.png (3.37 MB, 1608x1592)
3.37 MB
3.37 MB PNG
>>101117504

>>101117623
>>101117640
no way youre getting a labcoat lol

front slit dress? If your gen has the whole body I'm sure it'll work.
>>
>>101117504
Do you have a generic negative prompt you're using for all of these?
>>
File: OIL_08.jpg (476 KB, 1300x1107)
476 KB
476 KB JPG
>>
File: BigChungus_00126_.jpg (217 KB, 1024x872)
217 KB
217 KB JPG
>>101117689
>>
File: 116775249263548151-SD.png (1.29 MB, 896x1152)
1.29 MB
1.29 MB PNG
>>101117640

yeah, same defauly pony negs
unless Im forcing something, I keep it the same

>>101117781
thanks, need eyebleach now
>>
File: UI_0058.jpg (1 MB, 1024x1536)
1 MB
1 MB JPG
>>101117689
Finally, it is not very consistent as I thought it would be.
>>
File: ComfyUI_temp_fzfdu_00006_.png (2.03 MB, 1120x1440)
2.03 MB
2.03 MB PNG
>>
>not using the certified 1girl model Hunyuan
>>
File: UI_0062.jpg (794 KB, 1024x1536)
794 KB
794 KB JPG
>>101117689
>>
>>101117864
nogen
>>
>>101115464
>>101116786
can't believe how this hand came out
>>
>>101117859
>I can't type in Chinese nor do I want to.
>>
>>101117874
ayyy meant for this fag >>101117859
>>
File: UI_0063.jpg (848 KB, 1024x1536)
848 KB
848 KB JPG
>>101117891
Was about to say, damn.
>>
What would I use if I wanted to generate images similar to a base image I give it for reference?
>>
File: UI_0064.jpg (799 KB, 1024x1536)
799 KB
799 KB JPG
>>
File: UI_0065.jpg (866 KB, 1024x1536)
866 KB
866 KB JPG
>>101117927
You'll need Controlnet for poses.
Pick model for style
>>
>>101113951
>>101114040
Hunyuan can run on 6GB-
https://gist.github.com/sayakpaul/3154605f6af05b98a41081aaba5ca43e
There's also
https://github.com/DenOfEquity/Hunyuan-DiT-for-webUI?tab=readme-ov-file
Which claims to only require 8GB of VRAM.
Keep in mind the default chinese neg is important when running Hunyuan (at least for good hands, etc...), though without it you can get more rough styling depending on prompt, so it's something you want to experiment with, and that UI seems to be optimized with pixart prompts and negs on style list.

As for LoRA training reqs, not fully sure, https://github.com/Tencent/HunyuanDiT/issues/106 claims it's 24GB, but some anons here were able to train on 16GB so maybe they can chime in on that.
>>
>Sister-general for local diffusion with Chinese language models:
>>>/g/ldg

They fucking what?
>>
>>101117991
that's low, even for them
>>
File: _00234_.png (1.7 MB, 968x1400)
1.7 MB
1.7 MB PNG
whelp, I'm getting a lot of horse faces with pony realism. I must be missing something in the negative prompt to push it further from ponyism.
>>
>>101117991
lol, let them cook
>>
File: UI_0072.jpg (892 KB, 1024x1536)
892 KB
892 KB JPG
>>101118033
drop a catbox, I'll mod it for you.
>>
File: clown_world_00004.jpg (662 KB, 1080x1268)
662 KB
662 KB JPG
>>101117991
>>
>>101118153
>thanks, need eyebleach now
Wonder-full.
>>
>>101118069
bogged cheeks
>>
>>101118069
https://files.catbox.moe/u24zof.png
ty ty
>>
>>101118234
Do you have 'Manager' installed?
How complex do you want me to modify it by?
>>
>>101118294
yes, please feel free to do whatever. I can install missing nodes
>>
File: madhousemegalora.png (9 KB, 1451x50)
9 KB
9 KB PNG
aww shit here we go again
>>
>>101118352
Fact check me. I don't think it's available on windows.
>>
>>101118352
post settings
>>
>>101118314
Give me like 15 minutes or so. Going from scratch.
>>
File: prodigy.png (9 KB, 462x352)
9 KB
9 KB PNG
>>101118383
ah yeah it gives the triton error always, should have cropped it

>>101118430
prodigy + noise offset etc. pic related
>>
>>101118503
what's the learning rate on prodigy?
>>
>>101118517
>what's the learning rate on prodigy?
it's 1. You can change it with d_coef
>>
>>101117859
(if you know how to prompt it). Clearly most sdg tourists here can't prompt for shit and they are also blind, CGI faces produced by XL finetunes aren't realistic.
>>
>>101118643
oh yeah, this is so real I cant believe my eyes
damn son how you genning so real girls
>>
File: Test_0001.jpg (1.95 MB, 1664x2432)
1.95 MB
1.95 MB JPG
>>101118033
Is this what you were going for?
>>
Straight outta oven, it's fresh...
>>101118874
>>101118874
>>101118874
>>
>>101118795
that certainly looks better.
something close to >>101115464 was the goal, from >>101116573
really just want a solid starting point for pony realism
>>
>>101118881
A bit early. Bump limit is at 310, and image limit is at 150. We've hit neither.
>>
>>101118906
Adapt or die
>>
>>101118906
Let him cook
>>
>>101118881
Your collage is little bit biased, but it will do
>>
>>101118927
>>101118930
I'm just chilling. Nice to see someone else do a collage for once.
>>
>>101118935
My collages are also biased, cannot be helped.
>>
>>101118935
Sorry, there weren't too many good pictures to choose from...
>>
File: Test_0001.jpg (1.91 MB, 1664x2432)
1.91 MB
1.91 MB JPG
>>101118897
That's a different model.

Prompt:>>101117395
Neg:>>101117487
>>
>>101118937
There's three of us now!
>>
>>101118937
I'll never make it in one of those QQ
>>
File: tmp4yar2hax.png (308 KB, 608x384)
308 KB
308 KB PNG
>>101119026
Now that's what I call community effort!
>>101119032
>tfw a collage maker finally makes it into someone else's collage
>>
>>101118881
really weird choices for the collage
>>
>>101119007
wait, it's not pony realism? did I misunderstand >>101116573 ?
>>
>>101119032
photorealistic 1girls cant make it (for some reason)
>>
File: Test_0002.jpg (2.13 MB, 1664x2432)
2.13 MB
2.13 MB JPG
>>101119007
Same Settings. With this model:>>101116786
>>
>>101119064
How so? They went for a landscape theme, sure, but it's a very solid composition desu.
>>
>>101119073
Now I know the secret sauce for this. It is time to cook some Pizza for the next College party
>>
>wonder why my gens are suddenly shit
>I forgot the style including score_schizo
Every time. I'm impressed whenever I manage to pull off something half-decent without the boost in quality.
>>
So the only reason they're spamming their 1girls here is to be in collage....
>>
File: 0.jpg (498 KB, 2048x1024)
498 KB
498 KB JPG
>>
>>101119074
this model that model. might be better to say the name
>>
>>101119119
>Uses pony
lmao, like kill yourself.
>>
>>101119119
This is a PixArt/Hunyuan thread, take your pony shit to /sdg/ furry fuck
>>
>>101119130
No in us spamming 1girl you faggot!
>>
>>101119130
Ironically enough that doesn't work in my collages (most of the time). I try and prioritize creativity in whatever shape or form. The more of the same is posted, the less likely it is to appear, unless by sheer luck one of them ends up being exceptional.
>>
>>101116871
>>101116851
>>101116786
>>101116782
Wtf? You like women? What are you, trans?
>>
>>101119203
That's a negative, I'm a Bear in the woods sir.
>>
File: tmp4bz69jlb.png (201 KB, 550x550)
201 KB
201 KB PNG
You'll have to work harder for the badge of excellent quality.
>>
>>101119177
>Local Diffusion General
Where in the name did you see that it is only PA or Hunya?
>>
>>101119287
Don't bother, these are not genuine posts anon. See above.
>>
>>101119240
Shut the Fuck faggot
>>101119275
Fag
>>
>>101119200
I mean, I take issue only when it is blatant spamming with lack of creativity. It's hard to post quality with XL but it's certainly possible, that said there's not a single Pixart or Hunyuan gen in between and they clearly are not testing any DiT model.
For instance
>>101117591
Seems ok since it has more cinematic look, but the girl is still just standing there whivh is annoying. When you get used to prompting models capable of more you can immediately tell slop from a mile away.
>>
>>101119310
See>>101119240
>>
>>101119287
>>101119302
It's the schizo anon twins falseflagging. Having a good sophmoronic laugh.
>>
>>101119345
What I'm more curious about, is whether anon/s involved try to agitate both generals against one another. Pretending to be from /ldg/ whilst annoying over at /sdg/ and vice-versa.
>sophmoronic
that's a new one for me
>>
this general has no bitches!
/sdg/ is full of breedable hot twinks!
>>
File: file.png (150 KB, 256x256)
150 KB
150 KB PNG
Beautiful atmospheric perspective.
>>
>>101119450
It's a "use your imagination" kinda gen. Cooking well I see.
>>
>>101119327
>>
>>101119450
HYPE!!!!!
>>
File: ComfyUI_temp_fzfdu_00018_.png (2.51 MB, 1120x1440)
2.51 MB
2.51 MB PNG
>>
>>101114627
i've prompted copious amounts of jinx in the past 2 weeks
>>
>>101119450
That's not AI
>>
File: 00205-739674756.jpg (489 KB, 1058x1411)
489 KB
489 KB JPG
>>
File: 00202-85912475.jpg (318 KB, 1058x1411)
318 KB
318 KB JPG
>>
>>101118234

See
>>101119222
>>
File: ComfyUI_temp_fzfdu_00026_.png (2.27 MB, 1120x1440)
2.27 MB
2.27 MB PNG
>>
Hop over here, in case you missed it:
>>101118874
>>101118874
>>101118874
>>
>>101116433
cute



[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.