Previous /sdg/ thread : >>100122230>Beginner UI local installFooocus: https://github.com/lllyasviel/fooocusEasyDiffusion: https://easydiffusion.github.io>Local installAutomatic1111: https://github.com/automatic1111/stable-diffusion-webuiComfyUI (Node-based): https://rentry.org/comfyuiAMD 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>Auto1111 forksForge: https://github.com/lllyasviel/stable-diffusion-webui-forgeAnapnoe UX: https://github.com/anapnoe/stable-diffusion-webui-uxVladmandic: https://github.com/vladmandic/automatic>Run cloud hosted instancehttps://rentry.org/sdg-link#run-cloud-hosted-instance>Try online without registrationtxt2img: https://www.mage.spaceimg2img: https://huggingface.co/spaces/huggingface/diffuse-the-restInpainting: https://huggingface.co/spaces/fffiloni/stable-diffusion-inpaintingpixart: https://huggingface.co/spaces/PixArt-alpha/PixArt-Sigma>Models, LoRAs & embeddingshttps://civitai.comhttps://huggingface.cohttps://rentry.org/embeddings>Animationhttps://rentry.org/AnimAnonhttps://rentry.org/AnimAnon-AnimDiffhttps://rentry.org/AnimAnon-Deforum >SDXL info & downloadhttps://rentry.org/sdg-link#sdxl>Index of guides and other toolshttps://codeberg.org/tekakutli/neuralnomiconhttps://rentry.org/sdg-linkhttps://rentry.org/rentrysd>View and submit GPU performance datahttps://docs.getgrist.com/3mjouqRSdkBY/sdperformancehttps://vladmandic.github.io/sd-extension-system-info/pages/benchmark.html>Share image prompt info4chan removes prompt info from images, share them with the following guide/site...https://rentry.org/hdgcbhttps://catbox.moe>Related boards>>>/h/hdg>>>/e/edg>>>/d/ddg>>>/b/degen>>>/vt/vtai>>>/aco/sdg>>>/trash/sdgOfficial: discord.gg/stablediffusion
>>100127849
>>100127919
>>100127937
>>100127983
>>100127998
How do you preserve the colors of the original image when creating images via ControlNet (assuming this image was even created via ControlNet)?
>>100128261That was probably an img2img with controlnet added, so the original colors were already there.
UKbros, are you going to jail?
>>100128470Nice animation!That looks pretty fluid, anon! :]
>>100128538what bs did they come up with this time?do they now need a loicense for image gen?
>>100128261This is also damn cool.
>>100128560They made deepfaking women a punishable crime.
>>100128551Thanks, I'm very lazy. It's pure prompt and animation model only. I will put more effort in when the new animation models come out hopefully before the end of the month.I've seen some boob-horrors in my gens i can tell you.I like your's it's very natural looking :)
>>100126765You can use AI mirror on android, theres something called magic brush(its inpaint) then you brush around your face and type astrounaut, in space and it works pretty good, theres also a filter called realistic where it takes your image and uses your prompt to change it, this is a pic of gandolfini i used in the realistic filter as an astrounaut
>>100128660That's amazing for just proomting with animdiff! :DOhh? Is there something you know of that comes out soon? Sounds exciting!LOL sometimes it has a mind of it's own hahaha!Thanks so much! Tbh I made a 3D model, animated it, then ran it through img2img haha
>>100128801Err, vid2vid not img2img hahaha
Good night, anons! See you tomorrow :]
>>100128945Take it easy dude!
Have a cute Mugi holding an exploitable sign.
>>100128960Er, fixed her mouth. For some reason it likes generating a fucking pimple below her mouth sometimes.
>>100128603time to make femboys who identify as a man with he/him as their pronouns
Good morning Sigma bros
>>100128960Is that your avatar?
>>100129049No. I don't avatarfag. I just gen various cute anime girls that I like, often holding signs I can make memes out of.
>>100128801with SD3 should come a new animation model. Pretty sure it was quoted as such by stability on X some time ago.
>>100128979unreal beauty standards :(
I find SDXL challenging. If your prompt is perfect, you get nice results. If you mess-up even one word, it becomes crap.Is this also your experience.>>100129043good morning.Were you able to install locally
>>100129114Haha and I have acne myself so that makes me a hypocrite!
A few days ago I posted here complaining about running out of RAM while trying to run SDXL with LORA and anons recommended me to try comfyui .With comfiui I could actually run SDXL with with Lora and controlnet thank you but when I tried to generate a picture with 1200*640 resolution, one lora and two controlnet it took one minute to generate and gave me an error like >ran out of memory while running VAE, retrying with tiled VAEwhat further optimisations can I try? How are you generating 2048*2048?My system is RX6900XT with 16Gb VRAM and 16gb system RAM. And apparently the RAM is just barely enough, it uses swap when running controlnets + LORA.Does it store the model in the ram? If so would having more RAM and enabling --medvram help?
>>100129140>Were you able to install locallyYeah, how do you install locally? Is this a thing?
>>100129192Try Forge.
>>100129145>I have acne myself so that makes me a hypocrite!I have Alzheimer's so that makes me ... uh ... I forget.
>>100129192>what further optimisations can I trytiled vae like the message suggests?
>>100129207did you read the sticky? what ui you trying to use? automatic1111 and comfyui are both easy to install locally
>>100129262>tiled vae like the message suggests?It did that itself. However my machine barely handled that, I had my cursor freezing on desktop, swap utilization and 20 steps generation took over 60 seconds. Technically it works but just barely.
>>100129281>did you read the sticky?Doh! Will do. I Sigma legit stable diffusion or something else like dall-e ?
Is switching from Auto1111 to Forge worth it?
>>100129207>>100129281I'm referring to Pixart Sigma
Herehttps://youtu.be/9FmmpY1wdSoWatch
>>100129140>If you mess-up even one word, it becomes crap.yeah, the elements i add to the prompt, the more sensitive it gets to the wrong word or mispellings etc
>>100129366For me, with 12 GB VRAM, it was very worth it.It's noticeably faster and I can do higher resolution gens without running out of memory/VRAM.
>>100129366for XL its mandatory, no clue for 1.5 though
>>100129378>I'm referring to Pixart SigmaOneline only? I get throttled real quick.
Are controlNets really neccessary, aren't they just coving up bad prompting?Never used them but reading about it atm.What can you do with a control net that you can't with a good prompt?
>>100129435>What can you do with a control net that you can't with a good prompt?get a exact pose
>>100129043
>>100129435With a controlnet you can take a picture of a city during the day it and turn it into an cyberpunk anime anime picture of the same exact city ant night with minimal deviation, or you can turn a 3d render into a background. You can photoshop an anime girl into a photo of a car and make it all anime.Basically only controlnet can enable you to get EXACTLY what you want.
>>100129145>acneget a stack of clean white hand towels. put a fresh one on your pillow every night. wash them regularly.
>>100129452Use IP-adapter-face-ID to reproduce likeness without embedding or lora.
>>100129537????????????????
>>100129570https://www.youtube.com/watch?v=KHm5Q5TfNvE
>>100129366gud
>>100128551Is this vid2vid?
>>100129396>>100129420NTA but I got 12gb ram also and A1111 is faster for me.Forge has better vram management but gen wise A1111 is faster for me by about 6-7 seconds. Peeps tell me forge is faster but Im sure as fuck not seeing it.
>>100129598now show me the exact pose matching
>>100129140Local is best. There's a diffusers version and also a checkpoint that works in comfyui. Need about 24GB to be in the 1024 club and you can also fully fine tune with that much ram
>>100129631copyright this. $$$
>>100129625It's all in the vid.>>100129631>24GBthat's too much vram
>>100129366Damn what model is this?
>>100129452>>100129498I watched a S Detweiler yt video, understand now, thanks anons.
>>100129642Copyright AI output hahahaha
>>100129685I'll make up for the both of us
>>100129977>Copyright AI output hahahahaLie. Say you did de-novo, a muse sang straight into your cerbral cortex and you drew it with your Pilot V7 Rollerball . Make their lawyers prove otherwise.
Textual data is hard to manipulate with programs.
>>100130023It's not a rare pepe. I can make thousands>>100130036Not with large language models, ever since chatgpt upped the game.
>>100129752Animerge
Who are the great light and shadow artists?
https://hyper-sd.github.io/
https://www.youtube.com/watch?v=yrBke4JsjlQ
>>100130237I don't get it.
>>100125398>>100125675did you mean as a desktop background or somethin? heres a 16:9 :P
I'm trying to cook lora. It works with prodigy, rank 16, lr 1, constant lr schedule, no warm up etc.changed lora rank from 16 to 32 and loss went to NaNpic at the bottom of loss before the disaster, earlier checkpoint was severely undercoocedShuld I go home and change to AdamW or something or fiddle with other parameters?
>>100130446What's your network alpha/dim?Did you go over?also:> "optimizer_args": "decouple=True weight_decay=0.01 betas=0.9,0.99 d_coef=0.8 use_bias_correction=True "I use these for prodigy with cosine
>>100130690I tried to make my first quick lora but it seems that kohya_ss was always the answer.
>>100130862Slap this in prompt, see what happens:Salar de UyuniBREAKMarble CavesBREAKWadi RumBREAKDevils TowerBREAKRichat StructureBREAKZhangye DanxiaBREAKDeadvleiBREAKTianziBREAKDanakilBREAKSocotraAntelope CanyonBREAKChocolate HillsBREAKCave of the Crystals
>>100130799Yea I'm just running off of kohya_ss. I've had this work with for sd1.5, sdxl, and ponyxl.
>>100128979It is probably trying to gen lip detail, but doesn't get it right. It might if you ask for "detailed lips" or just "lips".
>>100130894maybe after this batch, i dont know what half of those words are in reference to though lol
>>100130963weird places on earth
>>100130894Does that incantation summon an ancient evil?
>>100131009>Does that incantation summon an ancient evil?Try it, see what happens. Maybe.
>>100130894hard mode, change the BREAK by AND
>>100130446You could do adamw8 but imo prodigy is overall most rational to work with so I'd change its configs instead.
I found what score_9 is. But what is source_anime, BREAK, AND?
>>100131057have you tried reading the pony page on civitai and the wiki for A1111?
>>100130957
>>100131154Only "What is score_9" page, and it was on model page. Thanks.
>>100131030
>>100131239excellent. moar.
diffusion models will never be capable of producing picrel without a lora, face the limitations of your tech chuds
>5 hour old thread>not even half way to bump limitit's over ai is killtime to pick up a pencil
>>100131273>time to pick up a pencilAnd feed your pix to img2img. Good idea.
>>100131214>>100131180cute and nice reflection.Goood to see you
>>100131293that's basically impossible to do.
>>100131362kek, heres a pity (you)
>>100131219
best AI image editor I can use? I wanna remove a girl's panties from a drawing to make it have a bare ass.
>>100131560hello officer, could you please provide your name and badge number
Late morning!>>100130380Ah, I meant as part of a prompt, for the background. See picrel.>>100130894Interesting places...
>>100131019I knew it!
This guy said "show me your real art"lol
>>100131672That actually looks like an ancient evil.
Has /sdg/ been healed finally?
>>100131720Healed from what?
>>100131620I did not say a fucking kid you braindead retard.
>>100131733I just searched and found only this thread.
>>100131485adorable
>>100131714Nice
I just started using sdxl after 1.5, help me out anons.Pony vs Autismmix(with pony?) or somthing else?If I use sizes lower than 1024, will it have worse composition or just blurry/worse faces and etc? I saw that sdxl works well only with very specific sizes, does that mean this exact sizes or just using 2^n combinations is fine?Also, any tips on improving generation speed would be nice.
>>100131826You can probably get away with 1024 x 768 with pony/autism
>>100131738of course not, officer, you just want to make non-consensual nudes of adult women
just in time for fall, i wouldn't be surprised if designer/dalle3 introduces a version suitable for schoolrooms with quicker but less refined generations and enhanced safety.
>>100131926fall?
>>100131720yeah anons started ignoring d*b* again so its peaceful here
which one of you fucks was it?
>>100132040Someone from /b/ I bet
>>100131961new school year
>>100131926>enhanced safetyWill it only have one output?
I miss the colder weather
sd3 fucking whenno im not going to pay for blurred out 1girls
Forge/cumfy when? https://github.com/garibida/ReNoise-Inversion
Morning anons>>100130140Thanks for the pasta anon :)
>>100132383not sure what it is, but some fuckers certainly got a fetish for this, I bet there is a market and you could make a lot of money if you sell this kind of shit
pip install torch-directmlERROR: Could not find a version that satisfies the requirement torch-directml (from versions: none)ERROR: No matching distribution found for torch-directmlwhat do
pip install torch-directmlERROR: Could not find a version that satisfies the requirement torch-directml (from versions: none)ERROR: No matching distribution found for torch-directml
>>100128979>>100129087>>100129145nice. mind if i use it?
>>100132488>torch-directmlhttps://github.com/microsoft/DirectML/issues/544
>>100132509fucking lazy motherfuckers. python 3.12 is the stable version and they're not even on 3.11
>>100132474>not sure what it isthey're goo girls, anon. Idk about monetizing though, seems like too much effort in an already saturated environment for it to be worthwhile. but I guess if someone's getting paid for it, may as well be me. Without an established following though you're just floating in obscurity, one of a thousand other fiverr/patreon sluts begging for work
>>100132430pretty sure noise inversion has been a thing in comfy for a while
>>100132539auto also had it really early on, but it was crappy
>>100132532What model is this? I like the anime screenshot style.
>>100132539>pretty sure noise inversion has been a thing in comfy for a whileif only. there's no real solid implementation, besides the included inverse sigmas node, but that's only half the process. You have to hodgepodge multiple custom nodes together to get a similar result to Multidiffusion's Noise Inversion
>>100132558it's a merge I mixed that i've been tinkering with, but Animagine 3.1 is the foundation
>>100132206fuck this dog lmaoi hope someone draw a picture of it getting killed
>>100131714catbox?
>redhead >teen more please, sir
someone requested
>>100129192Are you using DirectML?
Slow Monday.
A long overdue suggestion from a friend.This one's for you, buddy.
>>100131826autismDPO. >if I use wrong sizes will it be fuckedit will be fucked. used SDXL sizes. they have around 2m pixels and the dimensions are in increments of 64. >speedeuler 30 steps. use mixture of diffusers instead of hires fix. when you find an image you want to upscale. figure out what command args give you speed or swap to forge.
1girl
Requiem for Bauhaus
1boy
>>100133172that's still a girl you retard
>>100133293how do you know?
>>100133092>Requiem for Bauhaus
FREE SD3https://glif.app/@fab1an/glifs/clv488uy10000djtrx70u03no
a little toasty... lemme fix that.
>>100133485>sign inI refuse
where are the weights
>used to be able to prompt for over an hour>now can't even gen more than five or six images at a time without getting a BSODMy GPU's fucked, isn't it?
your mom's hiding them in her purse like your gameboy
>>100133913>he didn't tighten his tensor cores after every 10,000 gensbig oof
>>100133655but I dont(its ogre)
Im surprised this worked though>a red ball on top of a blue cube, a pink bottle on the left side
Are the guides in the OP out of date?I'm trying to learn proper prompting. I'm using A1111.
>>100134017I always suspected juicing fucked up your bone structure
>>100134034is this sd3?
>>100134048the technical aspects of prompting or just random prompting aspects? if the former just search the web about how clip works on SD, tokenization etc, if the latter just browse civitai, though you will learn the most by just randomly testing stuff anyway>>100134080supposedly yes>a green house on top of a purple aquarium, the house has a garden with blue sunflowers
What's the deal with SDXL open pose? The only available model is full precision and 5gb, why? Is there a lightweight or at least fp16 model for posing the characters?
>>100134135please do "Arnold Schwarzenegger throws a cream pie at Sylvester Stalone, 1980s action movie"
>impasto of a person doing a handstand>professional photography of a person doing a handstand
>>100134189ogre
>>100134240thought 1:1 would help but not so much
>>100134240arnold is too strong of a token. damn you arnold and your inhuman strength
Slow threads these days huh
>>100134240did you spell stallone correctly for the prompt?
>>100134251Ideogram for comparison, without and with magic prompt
>>100134259weekdays are randomly slow sometimes. but maybe spring break is over or something? who knows>>100134294lmao these are so wrong and so funny
>>100134291Now that you mention it I just copypasted the prompt, ugh .....with correct spelling
>>100134294
>>100134318ideogram with correct spelling and magic prompt
>>100134318oh shit my bad. I probably lulled oyu into a false sense of confidence by spelling schwarzenegger correctly
sexooo
>>100134340this is the closest to get the concept correct, though it still quite far off. I guess the pie throwing challenge remains for the next generation of models
>>100133913Are you using vlad's auto1111 fork? it's fucked for some reason
>>100131485Extremely cute.>>100132532This leads to muh secret prompt, best keep this away from here even if you do it.
>partially occluded facenow thats something different
>AttributeError: 'UiControlNetUnit' object has no attribute 'enable'How do I fix this?
I wonder if comfy would approve of this ear size
>>100134555would
I wonder if debo would approve of this horn size
I wonder if ani would approve of this loli size
I wonder if anon would approve of this boob size
>>100134636bigger. much bigger. BIGGER!
>>100134686its not in the cards :(
>>100134736would love to see this more overgrown and abandoned
>>100134555That asari's tentacles looking a lot like hair. She must be trying to imitate human women
>>100134834
>>100134875Did she get those from a second hand store?
>>100135058moody
>>100134294bottom right is detective monk.
almost
New>>100134744>>100134744>>100134744Thread
>>100135188