Previous /sdg/ thread : >>101834483>Beginner UI local installEasyDiffusion: https://easydiffusion.github.ioFooocus: https://github.com/lllyasviel/fooocusMetastable: https://metastable.studio>Local installAutomatic1111: https://github.com/automatic1111/stable-diffusion-webuiComfyUI: https://github.com/comfyanonymous/ComfyUISD.Next: https://github.com/vladmandic/automaticAMD GPU: https://rentry.org/sdg-link#amd-gpuIntel GPU: https://rentry.org/sdg-link#intel-gpu>Use a VAE if your images look washed outhttps://rentry.org/sdvae>Run cloud hosted instancehttps://rentry.org/sdg-link#run-cloud-hosted-instance>Try online without registrationsd3: https://huggingface.co/spaces/stabilityai/stable-diffusion-3-mediumtxt2img: https://www.mage.spaceimg2img: https://huggingface.co/spaces/huggingface/diffuse-the-rest>Models, LoRAs & upscalinghttps://civitai.comhttps://huggingface.cohttps://aitracker.arthttps://openmodeldb.info>Black Forest Labs: Fluxhttps://huggingface.co/black-forest-labs/FLUX.1-schnellhttps://huggingface.co/spaces/black-forest-labs/FLUX.1-schnellhttps://comfyanonymous.github.io/ComfyUI_examples/flux>Index of guides and other toolshttps://rentry.org/sdg-linkhttps://rentry.org/rentrysd>View and submit GPU performance datahttps://vladmandic.github.io/sd-extension-system-info/pages/benchmark.htmlhttps://docs.getgrist.com/3mjouqRSdkBY/sdperformance>Share image prompt info4chan removes prompt info from images, share them with the following guide/site...https://rentry.org/hdgcbhttps://catbox.moe>Discord6wUwtcJsr2>Related boards>>>/h/hdg>>>/e/edg>>>/d/ddg>>>/b/degen>>>/vt/vtai>>>/aco/sdg>>>/trash/sdg
>mfw Resource news08/12/2024>Linux Foundation Welcomes the Open Model Initiative to Promote Openly Licensed AI Modelshttps://www.linuxfoundation.org/press/linux-foundation-welcomes-the-open-model-initiative-to-promote-openly-licensed-ai-models>PreciseControl: Enhancing Text-to-Image Diffusion Models with Fine-Grained Attribute Controlhttps://rishubhpar.github.io/PreciseControl.home/>ProxyCLIP: Proxy Attention Improves CLIP for Open-Vocabulary Segmentationhttps://github.com/mc-lan/ProxyCLIP>BRAT: Bonus oRthogonAl Token for Architecture Agnostic Textual Inversionhttps://github.com/jamesBaker361/tex_inv_plus>ComfyUI-RefUNet: nodes to use Reference UNetshttps://github.com/logtd/ComfyUI-RefUNet08/11/2024>Plush-for-ComfyUI: Mixture of Agents and other Agentic Systems Workflowshttps://github.com/glibsonoran/Plush-for-ComfyUI>ComfyUI-SEGAttention: Smoothed Energy Guidance Nodehttps://github.com/logtd/ComfyUI-SEGAttention>SimpleTuner v0.9.8.1 - much-improved flux traininghttps://github.com/bghira/SimpleTuner/releases/tag/v0.9.8.1>Forge - BitsandBytes Guidelines and Flux [NB4]https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/981>ComfyUI_bitsandbytes_NF4https://github.com/comfyanonymous/ComfyUI_bitsandbytes_NF408/10/2024>Forge 2.0.0 https://github.com/lllyasviel/stable-diffusion-webui-forge/commit/cfa5242>X-Labs Flux LoRA Collectionhttps://huggingface.co/XLabs-AI/flux-lora-collectionhttps://huggingface.co/Kijai/flux-loras-comfyui/tree/main/xlabs>SwarmUI 0.9.2 Beta: Flux, SAM2, UI Improvementshttps://github.com/mcmonkeyprojects/SwarmUI/releases/tag/0.9.2-Beta>ComfyUI v0.0.5 Flux improvements and Hunyuan DiT controlnetshttps://github.com/comfyanonymous/ComfyUI/releases/tag/v0.0.508/09/2024>ComfyUI-Veevee: Video2Video Framework for ComfyUIhttps://github.com/logtd/ComfyUI-Veevee>BIGstockimage2M: ~2M stock images and captions @ ~0.4MPhttps://huggingface.co/datasets/bigdata-pw/BIGstockimage2M
>mfw Research news08/12/2024>VITA: Towards Open-Source Interactive Omni Multimodal LLMhttps://vita-home.github.io/>Multi-Garment Customized Model Generationhttps://arxiv.org/abs/2408.05206>EasyInv: Toward Fast and Better DDIM Inversionhttps://arxiv.org/abs/2408.05159>Instruction Tuning-free Visual Token Complement for Multimodal LLMshttps://arxiv.org/abs/2408.05019>DreamCouple: Exploring High Quality Text-to-3D Generation Via Rectified Flowhttps://arxiv.org/abs/2408.05008>DAFT-GAN: Dual Affine Transformation Generative Adversarial Network for Text-Guided Image Inpaintinghttps://arxiv.org/abs/2408.04962>LLaVA-VSD: Large Language-and-Vision Assistant for Visual Spatial Descriptionhttps://arxiv.org/abs/2408.04957>Model Debiasing by Learnable Data Augmentationhttps://arxiv.org/abs/2408.04955>Avoid Wasted Annotation Costs in Open-set Active Learning with Pre-trained Vision-Language Modelhttps://arxiv.org/abs/2408.04917>Surveying the Landscape of Image Captioning Evaluation: A Comprehensive Taxonomy and Novel Ensemble Methodhttps://arxiv.org/abs/2408.04909>Data-Driven Pixel Control: Challenges and Prospectshttps://arxiv.org/abs/2408.04767>Prompt and Prejudicehttps://arxiv.org/abs/2408.04671>Mitigating Hallucinations in Large Vision-Language Models (LVLMs) via Language-Contrastive Decoding (LCD)https://arxiv.org/abs/2408.04664>KEEP: Kalman-Inspired FEaturE Propagation for Video Face Super-Resolutionhttps://jnjaby.github.io/projects/KEEP08/11/2024>JambaTalk: Speech-Driven 3D Talking Head Generation Based on Hybrid Transformer-Mamba Language Modelhttps://arxiv.org/abs/2408.01627>SAT3D: Image-driven Semantic Attribute Transfer in 3Dhttps://arxiv.org/abs/2408.01664>Working with Color: How Color Quantization Can Aid Researchers of Problematic Informationhttps://arxiv.org/abs/2408.03424>Decoding Visual Sentiment of Political Imageryhttps://arxiv.org/abs/2408.04103
starting with higher base resolutions really lets Flux shine, their website says it supports up to 2MP, so 1408x1408 is a good starting point. I'd adjust in 128-256 pixels on either dimension to get your preferred aspect ratio. Not much of a speed hit with nf4 too
>mfw debo
>>101853891>starting with higher base resolutionsI was starting experimenting with this a bit at the end of yesterday too. I'm curious how high you can push it before it gets weird but pushing above 1mp doesn't cause any noticable issues>Not much of a speed hit with nf4 tooI noticed this as well. I was expecting an exponential slowdown but speeds were almost the same at the resolutions I was playing with. another thing I'll be interested to test the upper bounds of
Chubby Pikachu vibes
brrrr
quiotla best artist
>>101854290I'm blue debo de badu biI'm blue debo de badu bi
'on brand' well alright
>>101854326i'm not debo, you're debo
>>101853740>t. bought a $1500 gpu for the sole purpose of generating tranime shit
>>101854415>he cares what other people do with their money
sry for mass reply>>101854397thats just how the song goesI particularly like this gen; very captivating>>101854415based. what gpu did you go with?>>101854441feels like there's a story here....>>101854474hard mode: gen her with her MEKA too. that'd be fun but idk if you're using a lora for this
>>101854514>feels like there's a story here....just a small guy with a big shield, trying to get home safely
**** still spamming his name and trash here everyday
>>101854514it is a bit difficult yes, cant find the right prompt yet
Is the party on the other thread?
>>101855526this is the schizo general yes
me, on the right
:/
>>101855526Other thread (/ldg/) is 10 times more active than this one and is better for discussions.This general is reserved for keeping debo contained.
>>101855723and what's preventing this big bad debo posting on the the other thread?
>>101855526the party is in the linked discord in OP
>>101855742you won't post in /ldg/ because this place is your home, now stay here like a good boy
>>101855807I'm coming over to ldg, make some space for me
very schizophrenic
>>101854415I think we should have anime OPs every time, specifically to piss this loser off.
>ded schizo general
>everyone moved to /ldg/ to get away from debo
>>101856292debo is on the other thread too
>>101856322as long as he isn't posting there regularly with his gens it's fine
gn people
Is Layer Diffusion compatible with other UIs than Forge and Comfy?
>debo is crying everyone left
>debo id deboing deboone debt
>>101856438gn adventurer
>>101856668thread schizo
>>101856888checked
CatJak won so hard its not even funny anymore
shit thread
you're a pathetic piece of shit debo
who do we hold accountable for killing /sdg/?
>>101857383mike pence
one shit shizo can destroy anonymous boardthis is why people use discord
>>101854032How do you get to your house though?>>101857383Maybe the thread splitters?
Damn I finally tried inpaint sketch... it really fixes up stuff I like from dall-e nicely. Here's "before" with extra space added for drawing.
hes crying
Here's "after" - not exactly the same, but I like it
>>101857684you can stay here, maybe even consider joining the discord?
>>101857651what is inpaint sketch?
the thread is dead because of flux gen times. Nothing to do with debo or ldgsimilar thread slowdown happened when SDXL released.
/sdg/ let's sing when?
>>101857904with nf4, my flux gen times are ~1min on my 4070. thats including dynamic thresholding + adaptive guider which slows it down by roughly 25%
>>101857983I just don't get it because I had decent gen times yesterday and today it's fucked. Maybe it has something to do with updating comfyui and dependencies
i want to egg debo but with bricks
>>101858015I was having that problem a lot over the weekend w/ gen times flying around wildly. I did so much random stuff that I dunno what the specific silver bullet was but I know comfy made some changes on sunday to address some memory management issues with the nf4 library. idk if he's changed anything since
>>101858015it's probably because of the latest commit from comfy that messed around with memory tweaks
>>101858091I had the problem before that commit, and updating didn't fix
/sdg/ is dead
>>101857866See the image on the left? I just drew some black lines where I wanted the rest of the hair drawn, and the result is on the right. I'm letting it redraw the whole thing, but you can choose not to.
>>101857864I'll check it out, but I'm pretty bad at staying logged into discord.
>>101858262Dont worry! We would love to have you, it is slow-reply mode anyway in the discord, nice and chill
im back. still not impressed with flux
133m
>>101858116what does all your performance charts look like? is your cuda tapped at 100%? disk read or ram fluctuating a lot? maybe you can see something bottlenecking.>>101858262lol, would not recommend
>>101858329Yet another coffee earned!
Its been 2 years bros . How many of you got first access to beta sd 1.4 beta ?
>>101858446
grocery store nuisance
>>101858446holy shit its been two years. I didn't get into it until sept, personally. anyone know when sdg's birthday is? wasn't it in august?
>pls respond
>>101858501I was ledditor back than kek.>picrel sd3 large
>>101858402apple juice and belvedere desu
Is there any way to make my 1.5 renders look better? Currently I prompt normally and then use the Ultimate SD Upscale with the tile controlnet. It produces pretty decent results, but I'm wondering if there's an even better way to do it. Yes I can run SDXL, but I have a lot of styles on 1.5 that I like and haven't fou
>>101858892better in which way?
>>101858911I guess just a better workflow. This works, but I'm not sure if that's the proper workflow for upscaling these days. I kind of fucked off for about 6 months and came back to prompting. A lot has changed since then.Another thing I've been doing is scaling my 1.5 gens up and using Canny with SDXL to generate new images based on it. That also kind of works.
>>101859202you forgot the glock
>>101859252i had little hope of a coherent gun result, but I'll try, or a variation
>>101859065too baked still? if I drop down the guidance to 2 I get some weird blob results
>all cores maxedi need a spark cluster desu and a cluster for mongodb>>101859280any gun will do, or you can give her tattoos, she's a baddie, that's her vibe
Are you kinda autistic about your prompts?I separate physical attributes, clothing, and actions onto different lines
>>101859623same otherwise I can't keep a handle on what's in there
safety first
>>101860061she look like she poopin
>>101860088heh
i prompted myyyy way
>>101860088
I'm the Jim Morrison of /sdg/ you wouldn't probably understand my gens.
autistic prompting habits? nosir
>>101860225she got that lip bite you know thats a rough one
Any tips for getting a good image? I cant seem to get a good image. First time trying Stable Diffusion.
>>101861055try a negative embedding, or a couple. look on the civit page for that model, and look at some pics, and look at what negatives they use
>>101861055click the hires fix button. that enables a second pass that upscales and details the image
>>101861087>>101861093Any advice on getting more consistently good outputs? Any changes to settings or stuff I can do to make sure I avoid these more?
My only pitch for NAI is that it is quite literally the ONLY commercial AI service that is 100% completely uncensored and has 0 restrictions on what you generate whatsoever. NAI never sees what you generate. None of it is saved on their servers. There is no way to report generations to NAI, not even on the user end (most AI services let you tattle on yourself if the generator makes smut if it wasn't asked to LMAO). This is true for their image and text generators. No other subscription AI service is like this. Every single one has restrictions. Even TEXT BASED generators! For that reason, I support NAI.
>>101861102same answer. hires fix. its pretty much a free upgrade on every picture
>>101861135y tho? are they based in some eastern european country where you jsut need to bribe te right peoole.,?
>>101861055Use a negative prompt. Just use the default NAI one to start, it's better than nothing imo.>jpeg artifacts, normal quality, worst quality, low quality, missing fingers, extra fingers, bad eyes, ugly eyes, text, watermark, signature,Also use hires fix like the other anon said.
Is there a prompt length limit for Flux?
how do i update pip? like ot tells me to python.exe -m pip install --upgrade pipbut where do I type that?
>>101861558are you joking?
>>101861558>/g/ - technology
>>101861522512 tokens supposedly>>101861558>but where do I type that?in your cmd window
>>101861616Noted the prompt length, and thanks.
the gang
>>101861717who's who
>>101861744from left to right:>cat>kat>kate>cait
anon with RDNA2 gpu who cried about shit getting broken a while back when updating comfy.looks like the culprit was mixlabs nodes. these fucking chinks somehow install nvidia stuff when trying to update or something.
(boring photo circa 2015) trick isn't working
Penis
>>101861941She mad>>101862087These were supposed to be 1980s
It’s a goddamn shame what happened to /sdg/
idk what happened with this one but its slightly kino>>101862263she looks like she's mad her fake ID didn't get her into a local punk concert
>>101854147>>101854161nice ones!>>101859065>>101859354is this flux?
>>101862099Which model is this?Zesty mix XL?
>>101862494>flux?yea
>>101862603Img's look realy good. Love them :) And thanks for the info
ran old, april 2023, prompt, much too thicc and 'mature'. will have to work on it on the morrow.nite.
>>101862721gn
>>101862603any style/aesthetic tokens you'd recommend to get something more organic like that?>>101862721gn
pillow fight
>>101862782I wanna say most of the heavy lifting is being done by the xlabs flux realism lora
>>101862936ah fack, I wanted to try that but I'm using nf4. I think there's another realism lora on civit too but its also fp8
>>10186272199% of the women you want will look at you and vomit koff, mature women will accept your hairline not little girls
Flux loras should work with dev-fp8 right? The bored reality and xlabs loras worked, but none of the other ones I've tried have.
>>101863085For example, this one definitely did nothttps://civitai.com/models/639820/kazuma-sketchbook-flux
>>101863085dev works with dev, schnell works with schnell. they aren't compatible cross-model. precision shouldn't make a difference>>101863112a guy in the comment said he pumped it to 1.2. if you have some time to burn, it'd be interesting to see an XY comparison across different lora strengths to see what its doing (if anything)
>>101863124It's been a long time since I tested a lora with x/y and I've never done just strength without also testing epochs. I'll see..
>>101863124>>101863228This was the AT-J lorahttps://civitai.com/models/641755/flux-at-j-lora
Imagine if there was a way to fuse all loras and put them all together into the model to get some cheap finetunes, would be crazy
Imagen 3 from Google is now available to play withhttps://deepmind.google/technologies/imagen-3/
dropped cfg, cranked guidance. closer to what I wanted but overfitting>>101863417dang, you can never get those gpu cycles back. at least you know for sure now. I guess people are just uploading trash while the irons hot>>101863441you can merge loras into models but you just start getting fucky weights>>101863451its been available for a week or so but you need an account
/ldg/ is making fun of us for being ded today
>>101863462that's a pedo general, don't go there
>>101863451my very first try LMAO>>101863462>us
>>101863451>>101863456Gaayyy
https://reddit.com/r/StableDiffusion/comments/1eq98ca/euler_cfg_actually_works_on_flux/What? I didn't know you could use the cfg++ samplers on flux? Everytime I tried that I got some completely fucked up outputs
>>101863482I've been using dynamic thresholding + adaptive guider. it does... something, I guess. people have been trying a bunch of different approaches to get better cfg/guidance options but idk if there's a gold standard yet. people seem to overstate how well most of this stuff works
>>101863474don't act like you're above me nigger
>>101863492>people have been trying a bunch of different approaches to get better cfg/guidance options but idk if there's a gold standard yet.yeah, for the moment it's just complete experiment time, there's even an alternative to Dynamic Thresholding that is being tested right now >>101861407 >>101862667
>>101863456I'm guessing they work, just not for me for some reason.I went ahead and tried this one too>>101863112a cute Asian woman in a musky, dark bar <lora:odkzsb_0.2:1>
good morning puter friends
I assembled my succubus wrong>>101863569gm. nice gen
>>101863589You consistently destroy flux's body structure. lol
Going to bed earlier todayLast one from me,good night anons
>>101863814gn
https://github.com/comfyanonymous/ComfyUI_bitsandbytes_NF4/issues?q=AHAHAHAHAHAHAAHAHA
>>101864015>?q=
>>101864024Fuck off fag no one likes you
>>101864080
>>101864024it shows all the issues, even closed ones
150m
>>101864239Ok and?
>>101862978you are strangely interested in my relationship prospects. 99%? of women?! 1% seems generous but ill take it, ty.
looks like a poisoned image
>>101864584>dead butt
couldnt sleep much
>>101864992>dead schizo general no pedos (currently)
wet hair LoRa when
>>101865101there is wet hair baked into one of the rain or wet clothes loras
>>101865248nice
>>101865339ty
openai devday sounds neatanyone here applying to go?
>mfw deboing
>be stable diffusion exclusive general>have NAI gen as OP
>>101865961NAIs model is based on stable diffusion XL.
>>101865961>be autistic>get upset when something isn't literal or highly specific
>>101866248If this was LDG, he would actually have a point.But it isn't, so he's retarded and wrong as always.
>schizo general
is torch2.4.0 a requirement? when i update comfy it always tries to install it, overwriting my older version. also it installs the cuda version isntead of rocm...
>>101865094Nice one
How is flux for inpainting?
>>101867088>updates comfy, everything breaks https://vocaroo.com/16Huigfivv1r
how do i install the flux NF4 node?there are security requirements and changing the .ini does not do anything...
Where do you find cannibal and dismeberment models?
>>101867121>>101867297am I being mocked?
>>101867306cool
G'mornin Anons, have a great day!
>>101867365Maybe by someone but I >>101867121 liked it at least.
13 downloadsnicewent out and had a nice lunch desu little reward for myself for all my work so far and a pre-reward for what comes nextsoon:tm: two more weekstrust the plan
>i could easily refute every single point of the pastebin>but for some reason i wont
Any of you guys have luck creating catfish grade hot girls with the new flux model yet? I just tried it and it's still not that good. Feel like I can just sleep on AI image generation for another 6 months at the very least
>>101868158Yes
what the fuck is it with comfy's security level violations? i can't even install simple nodes now.
>>101868282>so comfortable
>>101868282aREn'T yoU fEeLiNG ComFy, AnoN?
>>101868282Fighting allegations of malware by only allowing his buddies to distribute malware
>>101868282How long before they start selling user analytics?Comfy doesn't see the tooling of the comfy org collective part of the project, using this logic as long as another team member does it, it's okay.
>>101868604>as long as another team member does it, it's okayYes. There has to be someone to shift the blame onto.
>>101868604crazy how a UI can make pedos mad af
Hey, are you the guys who have been flooding rule34 with AI slop? >:(
>>101868827it wasn't me
>>101865561>registration fee is $450spending money to travel to SF is ick
>>101868827beatings for them i say
>>101868282qrd?
>>101869349damn girl you fart with that ass
>>101869437why would she not?
>22 hours ago
>schizo containment general
my puter made a scary picture
>>101864498You're human nightmare fuel
>>101869376it's not letting me install certain nodes. for example the flux nf4 node. i tried changing the security level in the .ini file but it didn't do shit.
>>101869706It's illy's fault
this thread is reserved for keeping nigbo contained while he waits for pw to get on and start roleplaying
>>101869841what are their crimes?
>>101869841Based
>>101869858Far too many to list
>>101869703sounds cool/kool, thanks. i appreciate your obsession with mig
>>101868935L O N D O N desu
>>101863471This guy is the pedo BTW >>101869951
i miss debo
I hope your gens are meeting diversity quotas or we'll have to revoke your prompting license.
>>101870229same>>101870269flux has some pretty decent inherent diversity
>>101870339Thread schizo
174mdownloading a few m so i can get them captioned
>>101870111https://youtu.be/GM-e46xdcUo?si=OGupZk7xAStEAcNb
>>101870849AI is great but it will never be able to capture how cool the turkey gun guy is
>>101870959>>101870959>>101870959
