On the Case EditionPreviously on /sdg/: >>101313843>SD3 info & downloadhttps://rentry.org/sdg-link#sd3https://education.civitai.com/quickstart-guide-to-stable-diffusion-3https://aitracker.art/viewtopic.php?t=57>Beginner UI local installFooocus: https://github.com/lllyasviel/fooocusEasyDiffusion: https://easydiffusion.github.ioMetastable: https://metastable.studio>Local installAutomatic1111: https://github.com/automatic1111/stable-diffusion-webuiComfyUI: https://github.com/comfyanonymous/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 forksSD.Next: https://github.com/vladmandic/automaticAnapnoe UX: https://github.com/anapnoe/stable-diffusion-webui-ux>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: www.mage.spaceimg2img: https://huggingface.co/spaces/huggingface/diffuse-the-rest>Models, LoRAs & upscalinghttps://civitai.comhttps://huggingface.cohttps://openmodeldb.info>Animationhttps://rentry.org/AnimAnonhttps://rentry.org/AnimAnon-AnimDiffhttps://rentry.org/AnimAnon-Deforum >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 news07/07/2024>xinsir ControlNet++: All-in-one SDXL ControlNethttps://huggingface.co/xinsir/controlnet-union-sdxl-1.0>ComfyUI wrapper for Kwai-Kolorshttps://github.com/kijai/ComfyUI-KwaiKolorsWrapper>New Forge fork with upstream dev A1111 changeshttps://github.com/Panchovix/stable-diffusion-webui-forge/tree/dev_upstream_a1111>CharacterGen: Efficient 3D Character Generation from Single Images with Multi-View Pose Calibrationhttps://github.com/zjp-shadow/CharacterGen07/06/2024>Kolors: Effective Training of Diffusion Model for Photorealistic Text-to-Image Synthesishttps://github.com/Kwai-Kolors/Kolors>AUTOMATIC1111 1.10.0-RC: Performance improvements, SD3 support, new schedulershttps://github.com/AUTOMATIC1111/stable-diffusion-webui/releases/tag/v1.10.0-RC07/05/2024>Stability AI Updates Creator License Agreementhttps://stability.ai/news/license-update>ComfyUI nodes for LivePortraithttps://github.com/kijai/ComfyUI-LivePortraitKJ>Be-Your-Outpainterhttps://github.com/G-U-N/Be-Your-Outpainter>Gaussian in the Wild: 3D Gaussian Splatting for Unconstrained Image Collectionshttps://github.com/EastbeanZhang/Gaussian-Wild07/04/2024>ControlNeXt: Controllable video and image Generation, SVD, Animate Anyone, ControlNet, LoRAhttps://github.com/dvlab-research/ControlNeXt>InternLM-XComposer-2.5: A Versatile Large Vision Language Model Supporting Long-Contextual Input and Outputhttps://github.com/InternLM/InternLM-XComposer>Ostris VAE - KL-f8-d16https://huggingface.co/ostris/vae-kl-f8-d16>Stable Diffusion Helper: Automated processing for selection, cropping, and standardizationhttps://github.com/PoyBoi/StableDiffusionHelper>DisCo-Diff: Enhancing Continuous Diffusion Models with Discrete Latentshttps://research.nvidia.com/labs/lpr/disco-diff/>LivePortrait: Efficient Portrait Animation with Stitching and Retargeting Controlhttps://liveportrait.github.io/
>mfw Research news07/07/2024>TrAME: Trajectory-Anchored Multi-View Editing for Text-Guided 3D Gaussian Splatting Manipulationhttps://arxiv.org/abs/2407.02034>Zero-shot Video Restoration and Enhancement Using Pre-Trained Image Diffusion Modelhttps://arxiv.org/abs/2407.01960>Meerkat: Audio-Visual Large Language Model for Grounding in Space and Timehttps://arxiv.org/abs/2407.01851>Methodology to Deploy CNN-Based Computer Vision Models on Immersive Wearable Deviceshttps://arxiv.org/abs/2407.0023307/06/2024>Kolmogorov-Arnold Convolutions: Design Principles and Empirical Studieshttps://arxiv.org/abs/2407.01092>Certainly Uncertain: A Benchmark and Metric for Multimodal Epistemic and Aleatoric Awarenesshttps://arxiv.org/abs/2407.01942>Joint-Dataset Learning and Cross-Consistent Regularization for Text-to-Motion Retrievalhttps://arxiv.org/abs/2407.02104>HRSAM: Efficiently Segment Anything in High-Resolution Imageshttps://arxiv.org/abs/2407.0210907/05/2024>Enhancing Speech-Driven 3D Facial Animation with Audio-Visual Guidance from Lip Reading Experthttps://arxiv.org/abs/2407.01034>Image-GS: Content-Adaptive Image Representation via 2D Gaussianshttps://arxiv.org/abs/2407.01866>Pseudo-Labeling by Multi-Policy Viewfinder Network for Image Croppinghttps://arxiv.org/abs/2407.0197107/04/2024>BACON: Supercharge Your VLM with Bag-of-Concept Graph to Mitigate Hallucinationshttps://arxiv.org/abs/2407.03314>Improved Noise Schedule for Diffusion Traininghttps://arxiv.org/abs/2407.03297>Motion meets Attention: Video Motion Promptshttps://arxiv.org/abs/2407.03179>Efficient Shapley Values for Attributing Global Properties of Diffusion Models to Data Grouphttps://arxiv.org/abs/2407.03153>KeyVideoLLM: Towards Large-scale Video Keyframe Selectionhttps://arxiv.org/abs/2407.03104>Improving Zero-shot Generalization of Learned Prompts via Unsupervised Knowledge Distillationhttps://arxiv.org/abs/2407.03056
crap shot
>>101317340Any content where two anons could in good faith have an argument about their subjective interpretations of whether the subject appears north or south of the American age of consent. To actually have these arguments every single time is boring and pointless and helps nobody advance their "cause"
just realized prompt is fucked, gonna correct it.
>>101315587>the images used to train the model are rated?Yes, you're right. so, a score_1 is a lowest quality image.
>>101317575Are you using an anime model or something? The "finetune effect" here is very strong, stronger than in a typical finetune...
>>101317576Orange party
I now have a nice coloring book lora.:D
>>101317525really good apart from the AI not knowing if they lady should look at the guy or the viewer and chose both
>>101317568i was the anon that said that there is no argument, the law is pretty clear on this. Just because they are nude does not imply illegal material, just because they are young also does not define illegal, that was my point. I do not want to argue, if it pisses you off then challenge it in court with your local government.
>>101317610thats fun. post some more coloring book gens.hard mode: also make them half colored in
>>101317498replied in other thread but il just repeat it to make it easierit is an anime character ish, im auto tagging in kohyaa then removing/adding stuff, and my latest test that went bad was 20 step, 10 epoch 30 images so made it 6000 max steps, however i had one that went better the other day but has its own issue which was 30 step, 20 image 10 epoch, the character was more consistent in that one but it was always posing which its hands on its hips automatically, my end goal is character anatomy and color consistency which doesn't really seem 100% possible but im tring to get it as close as i can
1boy, they know I'm a big boss, got a big boss in this bitch, they know i'm a big talker, i'm a big body in this shit
I'm having a skill issue with SDXL right now. What am I doing wrong?
>>101317655Wrong vae
>>101317631
>>101317568but i will tell you no right thinking people really cares about this, no one wants to be labeled illegal for exercising their right to be human, the right of enjoyment to be naked and to do as we want as humans is protected by fundamental laws so you would be up against a lot. This means that simple images or video of people in the nude will never be declared immoral or illegal and thank god they can't ever do that because if they do i don't know if there is anything more dehumanizing... I'm willing to fight and kill tyrants that ever try that...
>>101317585might have to add score_1 to all my gens now
brain damage anon is cooking rn
>>101317655>>101317669Vae, but also the size you are genning at is wrong. SDXL needs to be 1024x1024 minimum because that's the size it was trained on. You can get away with 968x1024 though + upscale.
>>101317586i'm not telling you what i use, its is evident there are way too many retards around here... But I will tell you it was a random pony out of many i could choose from.
>>101317705I'd rather you didn't call me that
>>101317713ok yeah that explains it. respect for trying, I tried to do Pony realism once too, it was weird af
>>101317674lol
>/b/ pedo reject comes to /sdg/ to post underage children in bikini and call others retardsComical
tanginess aboundsnitey nacht soir
>>101317722no worries debo nobody actually thinks you're brain damaged, you present as a perfectly functional 110
>>101317725
>>101317722finish crying finally?
>>101317733>tanginess aboundswtf how did you know I was cooking caribbean jerk chicken with pineapple. very tangy>>101317736thanks! I'm happy to be a present
>>101317724you have to sort by recent, you have to download a lot but take it easy, 650 gigs triggered isp warning for me to much downloading... Honestly just try them all, test and study their capabilities. Most pony can do all pony to some degree, but every model is different. try sevenof9niceponyrealmix its ok, its close to real
>>101317638you're overbaking, if i have 20 images i do 12 steps 12 epocs adafactor.i set all the learning rates to 0.0003 network rank 32 alpha 16 (this depends on your gpu though), batch size 4. so i end up with 1000~ steps depending on how many images and steps im using. i get consistent enough images with good flexibility. but it greatly depends on the samples that are used.I have a lora that was trained mostly on faceshots, had no choice that was just how the anime was, and it was tricky to get full body shots. and i have another lora with the opposite problem. it is what it is, it's all on the samples. if i were you i'd prune any images that are a bit to similar to eachother. try to get it down to 20 or at least 25, use less steps and go up to 12 epocs. also if you're not already doing so set clipskip to 2. and if you haven't, prepend your captions with charname, score_9_up, source_anime,
>>101317743I can't even picture Debo crying. I imagine any situation where someone must confront an overwhelming humiliation, and if I instead imagine Debo there, I just see his face scrunch into a quizzical raised eyebrow sort of look, he says "hmm, well, no" and it just doesn't affect him at all.
test
>>101317776He cries when ran is around
>>101317776yeah pretty much>>101317783welcome back. behave yourself this time
>>101317770okay thought it could be that, my network alpha is currently 64-32 and i have a 8gb vram gpusay i did pruned to 20 images with 12 epochs how many steps should i do of each image, as far as im aware the max steps is those 3 values multiplied as a rule of thumb? also i have 2 versions of the images, one with all white backgrounds and another set with colorful backgrounds, if i dont intend to have backgrounds at all do i still use the coloured ones? thanks for the pointer by the way im just figuring this out by trial and error at the moment
>>101317638just give up, you can't force shit why do you people exist and persist, the other anon state the obvious, you are over baking an idea on ai models not trained for it. YOUR SHITTY FUCKING 100 MODES COMFY WORKFLOW ISN'T GOING TO DO IT OK... FUCKING LEARN HOW AI WORKS, THIS ISN'T FUCKING ABLETON OR ANYTHING LIKE THAT. THE FUCKING DAMN AI MODLE CAN'T FUCKING DO IT. YOU BEST BEST IS INPAINT THE FUCKING THING YOUR SELF USING A LORA AND NOISE YOU PLACE YOUR SELF IN THE IMAGE IN GIMP... THANK YOU, YOU ARE WELCOME.
>>101317817you are pathetic debo
>>101317835jin is the superior negro niggerjak. let it go
>>101317817Whatever happened to that Jin spammer
>>101317852Fucked off for good finally
>>101317708Yeah I'm still screwing it up somehow I think. I might have my models or lora's set up incorrectly perhaps.
>>101317826AND THEN YOU PASS IT THROUGH THE AI AND IT MAGICALLY UNDERSTANDS BECAUSE YOU GAVE IT THE CONCEPT, BUT TRY LORA AT 1 DENOISE USING MASK THEN TAKE THAT IMAGE COPY INTO CLICK SPACE AND PASTE INTO YOUR LOAD IMAGE AND PAINT THE MASK OVER THE ENTIRE THING GIVE IT LOW DENOISE AND PRAY, GEN OVER AND OVER TILL IT LOOKS RIGHT.NEVER USE RANDOM SEED, ALWAYS INCREMENT, MAKES WORKING SO MUCH MORE EASY BECAUSE YOU CAN SKIP BACK ONE AND SET TO FIXED
>>101317814im not sure how it would work for you. i have a 12gb with 4gb shared and this config runs me up to 14gb. are you doing this in 1.5? you can forget about the captions with score_9 and such, that's for pony models. backgrounds don't matter as long as you tag them properly. but it's better to just use the white ones if you're going for something specific. just do 12 steps, save every epoc and test all of them out and see where the model starts to overbake.
>>101317846I miss JIN. I bet he woulda done some cool shit with sd3
>>101317880are you using a 1.5 lora with an xl checkpoint
>>101317817Ran broke you
>>101317898cute
>>101317881AND I USED IT TODAY, I USED A LORA TO PAINT IN A GAS MASK, THE MODEL COULD NOT DO IT... i TOOK THE THE INPAINTED IMAGE AND GENNED THE MASTER FROM THAT.
>>101317898post her ass
>>101317924https://github.com/BlenderNeko/ComfyUI_Noise#unsampleri think this would help you
>>101317924EVEN IF IT DOES NOT UNDERSTAND ITS SO GOOD IT WILL UNDERSTAND IF YOU GIVE IT A REFERENCE...
>>101317884hey, yea running 1.5 haven't really messed about with pony much but plan to sometime, i pruned it to 22 images 12 epochs and 10 steps so 2640 max steps i think guess il see how it goes, one of the things im trying to train is the outfit, i got a wierd issue where the denim shorts were blending into the shirt even though they were tagged, im assuming that might be to do with overbaking it, but one thing i was trying was having the character wear different combinations of clothes and tagging accordingly because i thought that might help with the training what each part is, also while your here what are some tell-tale signs of overbaking, stuff to watch out for, am i wrong to think the outfit bleeding was overbaking?
>>101317947Its hardly a master but i wanted to illitrate a point, it moved the mask from where i wanted it, but i wanted to impress my father using a reference image of his computer desk. I was not meant to be a serious image, so i had my way with it and i had fun.
>>101317896I have 1.5 Lora's in a separate folder but I'll remove those just in case.
>>101317990and it had to have an army issue s10 respirator if it was going to be his nuclear war bunker, you get the idea, such concepts are fun to play with.
SD3 is more fun than beautiful babe finetunes and always will be
>>101318038>SD3kys
>>101318038this is retarded just look at it, she looks super imposed like a cutout, are you cunts serious? do you know where you fucking is mate?
>>101317880https://catbox.moe/Upload your image there and I will look at your settings
>>101318048Sorry finetards, SD3 base is good>>101318058Of all the criticisms you could have made, that one is bizarrely wide of the mark
>>101318058>mateBritbong detected
>>101318058>>101318038as in attention to detail was so obviously fucking lost, was it even attempted? not it wasn't, this is why SD3 is lost, it looks so flat, it looks so dumb and retarded, it failed at every test...
>>101318078Oh look another one, a fake cutout just pasted into place, you so obviously fucked it, and there is no way we can fix that at all... we do not care if it can even gen womens its done, it is done just look at it. my 5 year old niece makes more real drawings than this crap. Not because they are real, but because they blend together, the images you have posted do not blend at all they are so crap its horrible...
>>101318080>>101318121oh hey brain damage anon, I didn't know it was you
Good evening, anons! I hope everyone is doing well :]
>>101318121What is it missing exactly that makes it "not blend"?
>>101318078SD 1.5 still beats it, they deserve everything they will get.
>>101317805I just dropped some Azula butt
>>101318150life?being serious if you think you can fix it, its just dead when you look at it it looks like this https://www.youtube.com/watch?v=HlEREweogWEas in it looks fake as fuck and we see it, i'm autistic and i see it, its so bad, the colors do not blend naturally at all, it look so odd, that is an honest opinion that i can give. its looks uncanny as fuck even if the image looks ok in detail its still odd like i can see its off so much. I do not want to be a dick about this, if you are legit i will give my legit opinion about it, SD3 is not living up to anything because you trained it on shit, flat art etc.
>>101318150I'd like to know as well but I'm afraid to ask; do you remember his rambling last night? If he's faking the schizophrenia he's really good at it. I'm not someone to call anons schizo but he presents what looks like a very real case.>>101318152gens or it didn't happen
>>101318038I disagree. I official hate sd3>>101318106great horns>>101318149hi pw! I made caribbean jerk chicken rice bowl tonight. you'd be proud of me. I also fixed my comfy install and got all my sdxl workflows back>>101318191>Azula butthappens to the best of us
>>101318240>gens or it didn't happenthey are bad, but not that bad, at least they look natural with natural lighting and color if you are prepared for retard prompts. sd 1.5 is fuzzy, i have no time for that but its still good.
why is ran so attention starved?
>>101318240for a start anything over 512 x 512 will mostly cause problems with sd 1.5, up scale sure but it will add noise regardless. I know there are some wizards out there who can post process but for all intents and purposed a good prompter using sd 1.5 nails it
>>101318193>I do not want to be a dick about this>you so obviously fucked it, and there is no way we can fix that at all>my 5 year old niece makes more real drawings than this crap>are you cunts serious?
>>101318285and sdxl and pony, pisses all over sd3 just to its ability to gen beautiful images that merge subjects into the scene so well and them some. also very easy to prompt, more simple = better results.
>>101318079"cunts" is also bong/aussie specific dialect when used in general and not specifically toward a woman
>>101318294man you have not really seen me be a dick, they will ban me, and i will resist and come right back again and again and again and...https://www.youtube.com/watch?v=9VDvgL58h_Ydo you even know where you is mate?
>>101316420Leave it up to a fallen angel of love to be a huge slut
>>101318257Heya Debo! :DOh niceeee! That sounds really good!I'm actually making chicken and rice right now too hahaThat's awesome! I'm so glad you didn't have to reinstall the whole thing! That would kinda suck LOLIt was really hot as I figured haha but i'm glad it's done!
hmm, this is supposed to be a vampire. this model is weird..>>101318363very disgaea. did you even finish d7? I got bored before getting through the msq...>>101318370>chicken and ricewe are kindred>I'm so glad you didn't have to reinstall the whole thing!I spent so much of today firing off pip uninstall/install commands, reinstalled different CUDAs, all kinds of windows packages....... it was a mess but python and CUDA lay defeated. I'm just glad everything is back in order.what kind of stuff are you playing with tonight?
>>101318417What model are you using nowadays? I think it looks cool! Not much of a vampire tho hahahaLooks like she's made of electricity or something!Great minds think alike!I'm happy for you! That sounds like a lot of work haha so glad it paid off!Hmmm i'm not sure yet what i'm gonna doTrying not to be too tired, but the heat really drains meProbably just random gens after posting some of the ones I made before leaving for work!
>>101318257I don't see why you'd hate it. You can do all manner of faux-sublime sci-fi slop if you want
>>101318417>hmm, this is supposed to be a vampire. this model is weird..if it can't do it then it can't do it, what do you people not understand? its a neural network trained on specific things, give it concepts, it does not need to be good, hell i take peoples arms off and rotate them using gimp, then use cone tool set to fixed position to fill in the obvious blanks and i find that works way better than control nets with open pose and much faster... i am not motivated to give results here, i can make what ever i want, what ever i want. So that means people who pay would get what they want...
>>101318482>What model are you using nowadays?just downloaded a new one to celebrate that I've got all my sdxl workflows back. its called Chimera and its a mix of a lot of different models. its doing some fun stuff but I dunno how much control I have over it rn>>101318486>I don't see why you'd hate it.it doesn't listen to me. and when its not listening to me, its not like its injecting random cool shit I didn't ask for. and even when its not listening to me, it generates the same picture 70% of the time. its frustrating to work with and its outputs are not good and I hate it>>101318503I'm just playing around with a new model, anon. its not that serious
half you people aren't even into animation, we are miles ahead of you now, and yes its probably to complex, too much hard work
>>101318534our best animator left us and now its too emotionally taxing for us to think about animation
>>101318534>aren't even into animation>we are miles ahead of youCan you really beat someone at something they don't try or care about?
>>101318522>I'm just playing around with a new model, anon. its not that seriousone way to test a model, put what you want in the negative prompt and set cfg to 0, it will be low quality but it will show what that model is capable of. do not waste your time... learn what cfg actually does, so avoid to many negative prompts, start in the negative prompt on 0 cfg if you have to to figure it out, then copy it to positive and remove from negative... up the cfg slowly and watch slowly what happens on about 20 steps but depends on model, sampler and sampler scheduler play more wit the settings learn your tools!
>>101318569>one way to test a modelI usually just rerun prompts that I had a good time with in other models then tweak it depending on what the new model seems to jive with. I'm often working with models that are revisions or fine-tunes off models within the same lineage so it makes sense to recycle prompts and see how they differ
>>101318569> put what you want in the negative prompt and set cfg to 0, it will be low quality but it will show what that model is capable of.I no longer believe you are being sincere. This is trolling, plain and simple.
>>101318568Nice Rogue
>>101318522Ohhh cool!So far what you're making looks great! :]>>101318534I've made lots of animations hahaWhy don't you post one of yours, anon? :]
Has no one done a datsuma69 yet?!
>>101318534I took a neet break for a year or so so no money, I'm up to bank 10K again after job but got like 3.5K in expenses. I have the worst timing on money ever so I'm always poor ; ;>didn't go in on tesla at $150>didn't go in on nvidia at whateverahhhhhhhhhh
>>101318591yes this can work wonders, each model is different, but so is sampler and scheduler, but that is the comfy ksampler people might use 1111 it might be different
>>101318596ah believe what you want i don't care.
Shout out to anon who made this gen
lmao idk what this model is doing but I'm fascinated >>101318627if you had a million dollars rn what would you gut-instinct-moon-shot be?>>101318630this model has a splash on pony in it. what sampler/scheduler do people generally like with pony?
>>101318655>this model has a splash on pony in it. what sampler/scheduler do people generally like with pony?try them they are all different, some work for inpaint others work better for manual gens...
>>101318601Thankhttps://files.catbox.moe/c4umqw.png
>>101318655>this model has a splash on pony in it. what sampler/scheduler do people generally like with pony?so i go into what i'm using now right now for the gens i've posted but it won't work for inpainting imo.ddpmkarrasthere are others i might use for other ways such as in paintingsgm_uniforem works scheduler works good with in painting for some reson imo lcd work good with some models but you need denoise at like 0.10cfg lower and i only use that is its never gonna do it...do not try to low denoise a latent with nothing it does not work like that.
>>101318717Sad we have to catbox things like this to please jannies
>>101318728>ddpmwhat the
>>101318734I was told they don't like protruding nipples.
>>101318728sorry you try on inpainting sgm_uniform, it feels best to me, i used it in a lot of animatediff
>>101318756it works for some models anon, i don't know what you are using, you should stop if you think you know shit...
>>101318756you don't even know what it does or means...its an algo, that is all it is but drastically alters the image, come models work best on ddpm
>>101318745Run, it's the polglice
>>101318785>you should stop if you think you know shitbut anon, I was asking you for advice...>>101318799>you don't even know what it does or meanspretty much. idk if I've ever used ddpm much before, so I was surprised to be recommended it
>>101318785>>101318799First time here?>>101318802Watch out for them.
>>101318756dpmpp_3m_sde etc. more is not better... its over baked, hands are dissolving etc
>>101318627What's your net worth mouse
>>101318849>dpmpp_3m_sdemy version of comfy dont have any of that. its just ddpm here
>>101318849>>101318879oh wait, you said dpm, not ddpm. which dpm do you like for pony models?
>>101318849vs ddpm erm still a bad seed and result but its way better i know it when i look at what it gens.
>>101318879maybe its custon nodes, maybe you should install first comfy manger and you can update through the ui, but back up your shit first they break shit...
>>101318879you should be using a python environment for a start anon, i know it is hard for you it was hard for me to understand but its no issues once you grasp it. its on the lines of a basic arch install but not really, i can talk you through it...
>>101318947>you should be using a python environment for a start anonI refuse. I wish shove everything into global and global will be the one environment for everything. you can call me a globalist.I might have been more receptive to your message if I didn't just spend the better part of today making global python my fucking bitch
>>101318879and you should consider linux if serious because everything here trickles down from linux, its perhaps the best reason to use linux. So in linux i can use (this works on windows but i don't care about windows) we can have separate environments for python and so we setup such an environment and it can use what ever python it wants with out breaking the system, yeah super scary i know but not really at all...
>>101318991It's genuinely hilarious to me you're suggesting this to debo of all people. Seriously.
>>101318991no thanks. my gpu machine has windows because my gpu needs to draw fun and exciting games for me. linux is for boring stuff like terminals and web servers
>>101318967its a few simple commands tought nut and one commands you need to remember to activate it each time you use it and unpdate itakacd ComfyUIsource env_ComfyUI/bin/activate then i can dopython3 main.py'env_ComfyUI'can be anything, then when you update and install this program it will live here and be linked to here
>>101319006i do not care if you hate them, i help anyone get better.
>>101319024heh, thats a lot of extra commands I don't have to type since I have tamed and mastered global B)
>>101319024'env_ComfyUI'can be anything, then when you update and install this program it will live here and be linked to hereit was not a command, it was just highlighting it can be any name you want. and its not big deal, its just gate keeping faggots make it same so fucking hard when its just a few commands and you only have to understand what those command actually do. I've done way more than this including editing code so i comply against old hardware so shit works...
>>101319037That's just a personal preference. There's no advantage to using linux over windows for ai image generation.
>>101319064you are wrongend of discussion
should i attempt to explain full install process mean you its a little hard but I will back track
>>101319077lmao, whatever
>>101319107Yes, go ahead.
>>101319111be gone shill no one fucking cares about you, you need all the ram you can get and widows reports using screen shots of what ever you do to mircrosft aka nsa servers constantly this will slow you down. it also eats your resources its resource hog, low memory use linux solutions are best, that is final i do not care what you say at all, i know hardware down to the lower levels. so you can go fuck your self
>>101319127ok i will
>>101319139what the fuck is your problem, man?
>>101319122is this a new character? you've been keeping them pretty consistent>>101319139>widows reports using screen shots of what ever you do to mircrosft aka nsa serversthey already scraped recall before even rolling it out
>>101318717Awesome, thanks
fullsetup on arch liunx or just about any other disto install cuda first, for amd cards I could not get it to work so sorry this will only be for nvidia cardssin arch dosuenter your passwordits best to first update pacman -Syureboot...then pacman -=S cudathis is big download,,,
>>101319150>>101318193
>>101319180i do not have to help anyone if i do not want to...
>>101319180>>101319190but if i do it helps everyone using arch or just about any other linux to save time countless fucking time and waste of energy breaking they systems and i di break my system i won't say how because i do not want anons doing it, you can't just install it properly with out knowing and it was a major fucking head ache, so you better wise up and shutupalso i made a mistakepacman -S cudanot!pacman -=S cuda
on arch linux you will want this package before you start, now on my current system i can't back track if you have any problems comment in any iteration of this thread.pacman -S python-pip
and you also need pacman -S gityou will be using this to clone the directory then you will use python to make the python environment in that directory, once you activate as in>>101319024you will be doing this each time you want to run the program, but it will ensure i clean system and its a as simple as deleting it if you are done with it, no breakage, no problem
sorry i need some techno, i'm getting bored...
>>101319295minimal? driving? melodic? tell me what flavor you're hungry for and I'll give you something
the comfy on git does not explain this, so bear with me this is where i needed to fix it my self.. please bear with me i have not given up, i would have to do this myself again very soon because the hard disk i am on is malfunctioning so i have to do it again it was horrible but it was not that hard. it is a command to create envirment using python, its very simple is listed on arch wiki actually, but then once you activate it its all good and you do not get errors like THIS IS AN EXTERNAL MANAGED ENVIRONMENT ANON CUNT...do not do what i did it will fuck up your system and things will break eventually, do not fucking delete that file to disable, you won't have to micky is going to show you how it is done the proper way, micky likes his system not to break.
>>101319322do not worry i was just cussing, i'm fine and moisturized, give me a few minutes.
>>101319336damn, I wanted to dump techno tracks
>>101319346dump them, but i am fine i'm just reading through tons of shit because i have to do this all again. this system is breaking and to fix it i need to move to another hardrive before it losing the fucking plot completely
As stated on comfy githubYou can simple do this shit because it does not work well with arch and probably most linux its a simpleton approach and shit will break...DOT NOT DO THIS!!!pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu121do not do this yetwe got some setting up to do and its simple. We already installed cuda and python-pip correctly we now need to setup our venv but do not call it like they say on some gash website, no give it a real name i will explain next post just reading through it. And i will tell you the only annoying thing is having to activate it first before using it, never attempt to update with out that but you could not even run or update it without environment active, so it should never be an issue, just install comfyUI-manager then update using that. its a simple git clone command do not worry i will give them here but all you do on a git is copy the git clone address thengit clone <the url you copied>this will place in what ever directory you are currently in the terminal you ran the command its not super scary it simple copies the contents of their program directory into your directory then here you can begin doing your thing which we will be doing soon. What I hate are the gate keepers that make this shit seem super complex, not everyone knows git, not everyone knows python environments and shit it will seem alien to most normies.
>>101319433>ur venv but do not call it like they say on some gash websiteyeah do nothing yet, i will tell you what to do, do nothing just give me a few more minutes i do not want to fuck peoples systems upbasically do nothing until i tell you.
So is he genuinely schizo, an /ldg/ op, a troll, or something else? What do you guys think?
i fucking need a cig is all...
>>101319484i am telling normies how to install and use it, what you are doing is being retarded.
>>101319484you will not stop me now i'm too invested, i'm also autistic so good luck with getting me to shut it down retard kike...
>>101319499you're an extremely unpleasant person who is doing everything in his power to ruin a thread for everyone else
>>101319484who?>>101319490don't do it anon. you feel the same before an after the cig, you're just poorer and unhealthier after you burn one
I'm gonna take a big shit!!!!
>>101319511i'm programmed to kill to destroy, it was not you that done this it was higher power...
What I would give for an LDSR 2, it can be exactly the same with the only difference being that it doesn't take a year to sample
>>101319527>whoYou haven't noticed the one guy who has been spamming this thread with barely-coherent stream of consciousness schizoposting tonight and last night? I don't care if it's real or feigned, he needs to go
and its too late many of them skipped the process entirely and are now confident because they found
>>101319549oh, that guy. I don't mind that guy because he's caught up in some very resonant line of thought, even if we dont get it. he's chasing something; thats pretty cool. we should all be chasing something
>>101319527do you feel the same after a joint anon? you have never smoked so how the fuck do you know? if i can't smoke i would kill members of parliament... it numbs me, it makes me ignorant.
>>101319508>kikeAntisemitism has no place in /sdg/. GTFO. Praise Israel.
>>101319584is all hat that you can do?fucking gpu 900 init?
>>101319571>do you feel the same after a joint anon?no, smoking a joint is very different from smoking a cif, a cig is just a breath of air but a joint is a different state of mind>you have never smoked so how the fuck do you know?I used to smoke like a chimney lol>if i can't smoke i would kill members of parliament... you're more than the sum of your carcinogens
fuck these people the next step once you have the dependencesis the git clone command so to make sure you are in home do a cdcommand it will put you to home of current user, make sure it is not root and soexitsynccdgit clone https://github.com/comfyanonymous/ComfyUI.gitdo nothing until i tell you
verifying my ability to interactmy dog ate every running form of python besides colab in one weekplease use caution when running code>0HXASD
what does the sync command do?
>>101319640it syncs
to prepare switch to the actual directorycd ComfyUI
I will now rate the threads work based on how creative your idea and how difficult to prompted it would be>>101318599>>101318939decent>>101318609weak>>101318610>>101318631might be fairly impressive will have to try myself>>101318627>>101318893well executed>>101318650good>>101318655trash but elevated trash. nice try>>101318668awesome best in thread>>101318722>>101318734>>101318745>>101318787>>101318813better than some ive seen>>101318756>>101319155>>101319322simple oldie but a goodie nice to see them still being made>>101318763>>101319122very good craftmanship>>101318812>>101318879nice>>101318849>>101318906garbage learn to prompt>>101318959noobish but shows promise>>101318967>>101319022>>101319346cool and well executed
>>101319649thanks for all the (you)s
>>101319640it stops things breaking my lad>>101319645he is right, is basically makes sure all writes queued goto disk and is blocking... its just so you system will not fuck it, because we are making changes here...
>no you>not even a garbage response>JSA0WP
>>101319675https://en.wikipedia.org/wiki/Piasa
https://www.youtube.com/watch?v=RIHs34T-Azw
no fuck it i am done fuck you all fucking DIE
>>101319751don't get side tracked from your work. its very important. focus on your progress
>>101319675This is good.
>>101319649Get your eyes checked faggot
>>101319758>>101319751>
>Hey you. Yeah you. It's time to install Linux. Don't walk away from me. This is for your safety.
>>101319724ty so much for wordsI committed petty crime and got free weed bud thanks to you!>GAAKD4
is it safe to post yet?
>>101319944if you go back to 1990 you may have c slim chance
>>101319924I didn't really say anything, I thought your thing looked like the bird monster. Glad you somehow got free weed from it.
ok so SD3 lets you gen pretty big tits if you gen sorta retarded looking women in a hospital setting.
i miss schizo anon
>>101320373I never really knew who people were referring to when they talked about schizo anon
>>101320482nta but the lore doesnt make any sensedepending on the situation "schizo anon" is either everyone not a namefag, ran, debo, trani, all of them, one of them, ...
Made my first """""""""""animation""""""""""".Is there any way to get it more consistent and not jumpy and shitty?
a1111 updated!!!!!!!!!!!!
>>101321267I sleep, wake me up when it can run Kolorshttps://github.com/Kwai-Kolors/Kolorshttps://huggingface.co/spaces/gokaygokay/Kolors
https://youtu.be/sjb9AxDkwAQ?si=QWhfamIx-_K_ts80the cock doodles
https://jingjingrenabc.github.io/ultrapixel/Look at that, a model that can make really high resolutions, I hope it'll be released at some point in time
>>101321437>tfw needs 30 minutes to gen 4096x4096and it was just sd15
>>101321460Looks like they found a way to make it more efficient https://arxiv.org/abs/2407.02158>Ultra-high-resolution image generation poses great challenges, such as increased semantic planning complexity and detail synthesis difficulties, alongside substantial training resource demands. We present UltraPixel, a novel architecture utilizing cascade diffusion models to generate high-quality images at multiple resolutions (\textit{e.g.}, 1K to 6K) within a single model, while maintaining computational efficiency. UltraPixel leverages semantics-rich representations of lower-resolution images in the later denoising stage to guide the whole generation of highly detailed high-resolution images, significantly reducing complexity.
>>101321437The pictures have a Midjourney quality to it, first time I see that, my theory is that Midjourney is actually outputing really high resolution images and downsize at the end to make it seem it's only working on "regular" resolutions
>>101321529>my theory is that Midjourney is actually outputing really high resolution images and downsize at the end to make it seem it's only working on "regular" resolutionslike a reverse hires fix?
>>101320373me too
>>101321539not that complex lol, they just reduce the resolution of the picture at the very end so that we believe that Midjourney is making those pictures at 1024x1024 and not more, maybe that's their secret sauce
>>101321558my schizo theory is that they use a different sampling method and cfg value calculation, maybe even some attention fuckery
>>101321481>UltraPixel leverages semantics-rich representations of lower-resolution images in the later denoising stage to guide the whole generation of highly detailed high-resolution images, significantly reducing complexity.a lot of words to say "hi-res fix"
Does anyone know what model is this person using? specifically for the faces or something similar (NSFW) https://x.com/Doreiko_Ai
>>101317722>>101317749>>101317805>>101317895>>101318257>>101321579 nice wallpapers
>>101321657i like that sort of stuff what models works for that?
if you think i was serious, there is nothing let but total war faggots and we will win... WE do not care it is our right and we will fight for it until you allow or we take your pathetic cuck website down completey.its just a girl ffs wake up retard...
the ones with te fans are interesting but they fucking banned me everywhere for posting this citing child modeling are you cunts right in the heard do you know how much i can sue you now? not implying i will but fucking come on man... its just a bloody kid on a beach... btw 18 year old woman... byt what ever you see it wrong, its you that have fucking issues not me.,..
>>101321735>>101321756Shut the fuck up. If you want to generate kids just keep it to yourself. Pedos shitting up online services is why things be like they do
we will not stand for this shit we will destroy you that is a promise. don't care who you work for we will destroy them also. We should have probably done that 15 or more years ago. Put me in prison and i start killing everything i come into to contact with tell i am dead and you still do not win.
>>101321783i did not gen kids though they started it, and i and sick of fucking people tell me what is right and wrong when i know the fucking law
>>101321828there are many things technically scraping as legal you deserve to be woodchipped for
we will destroy you completely, we are done who is we? every fuck that is actually alive and not a bot... your website is bots yet, we just gave you entertainment... everything should be totally destroyed and all those who think little girls are disgusting should be lined up on the wall and fucking shot.
>beans "stealth posting" again
>>101321844oh yeah? come at me man, do it fucking come at me faggot... I'd cripple you fat useless faggot.
It's all so tiresome. AI attracts the worst of the worst.
>>101321873I don't even need to, when people find out what you are you get stomped by a mob who all saw nothing
look before you ban me again today do not bother btw, i juat wanted to say how dum that is, you can find a fucking million videos right there on youtube or worst is instagram why the fucking knee jerk reaction? I do not understand at all...you will never win, we will destroy you cunts totally and completely that is a promise i do mean that.i'm fucking 20 cans drunk, ex military i will fucking remove you from our country you cunts...
by force if i have to, you think we will back down and bag it? no we won't we are way smarter than you stupid fucking cunts. we are watching.
>>101321974hey mandrop a link to your manifesto before you do anything crazy
imagine that basedjack who set himself on fire over palestein but instead it's this guy defending gooning to children
they made me the way i am, my ability is excellent , pure excellent but at a price.
>>101322001who is they?drop some names bitch
oh nice a new schizothe old ones got boring anyway
>>101321987that shit is for faggots lol... You have no idea anyway, none at all you still think this shit is real... i can make more cuties if you like but they will just ban anything of course because you are now completely demoralised... I'm calling on all people that value freedom, get ready, be prepared for war. we are doing this because we are sick of it. we know what they are and why it triggers them so much, just be preparedyes janny i'm gonna fuck you, but i might fuck everything...so be he parrot faggot no one cares about you we fucking win dont we?
inspired by schizo anon
>>101322034that is perfect -
So, this is the power of /sdg/
>>101322035he warned you, we actually warned you, what do you want me to do about it?
>>101322058
you are already fucked beyond your stupid comprehensions any way so come tell us what makes you tick why are you nasty? I know i will die, i know that i did it for a better world faggot DA vry good faggot now tell me why do you hurt your own people faggot? just tell me.
>>101322086>>101322086>>101322086
I think he passed outbummer
I got tired of how 4chan scales images horizontally but not vertically so I wrote this ublock filterWorks on both images and webmsboards.4chan.org##.expanded-thumb,.expandedWebm:style(max-height: 800px !important; width: auto !important)
boards.4chan.org##.expanded-thumb,.expandedWebm:style(max-height: 800px !important; width: auto !important)
>>101322157shit wrong place
>>101321774Do you mind sharing style you're using? I feel like I should know it for sure but I can't remember it.
Nice
>>101322055yes
>>101322183Sure: That Artist is rizento. Draws a lot of Touhou.
>>101322733Thank you! Prompted him quite a lot before but only mixed with other artists, guess that's why I forgot the name now. Cute gens.
>>101322765Yeah, I usually mix as well, but recently I felt like trying some "pure" styles.