Previous /sdg/ thread : >>101606140>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>SD3 info & downloadhttps://rentry.org/sdg-link#sd3>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>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
oldy
Where will I get my daily malware now?
>mfw Resource news07/28/2024>LiblibAI: Chinese Civitaihttps://www.liblib.art>ComfyUI Mini: Mobile-friendly ComfyUI frontendhttps://github.com/ImDarkTom/ComfyUIMini>img2txt-comfyui-nodes: popular img2txt captioning models into ComfyUI nodeshttps://github.com/christian-byrne/img2txt-comfyui-nodes07/27/2024>Kolors ControlNet and Inpainting Model releasedhttps://github.com/Kwai-Kolors/Kolors#-news>AuraFlow-v0.2https://huggingface.co/fal/AuraFlow-v0.2>日本の美を学んだAI:浮世絵風画像生成モデルEvo-Ukiyoeと浮世絵カラー化モデルEvo-Nishikieを公開https://sakana.ai/evo-ukiyoe/>Pose Depo: A collection of ControlNet poseshttps://github.com/a-lgil/pose-depot>AUTOMATIC1111 updates to v1.10.0https://github.com/AUTOMATIC1111/stable-diffusion-webui/releases>Interpreting the Weight Space of Customized Diffusion Modelshttps://github.com/snap-research/weights2weights>Forge Major Update #1: Gradio 4 Enginehttps://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/853>WD ViT-Large Tagger v3https://huggingface.co/SmilingWolf/wd-vit-large-tagger-v307/26/2024>ComfyUI implements native Hunyuan-DiT support, packages model to single filehttps://comfyanonymous.github.io/ComfyUI_examples/hunyuan_dit>Fooocus v2.5.1 Updatehttps://github.com/lllyasviel/Fooocus/releases/tag/v2.5.1>HVM-1: Video models trained on ~5k hours of human-like video datahttps://github.com/eminorhan/hvm-1>Move and Act: Enhanced Object Manipulation and Background Integrity for Image Editinghttps://github.com/mobiushy/move-act>LLMImageIndexer: Use local LLM to create descriptive metadatahttps://github.com/jabberjabberjabber/LLavaImageTagger/>Generative AI for Krita: Version 1.21.0https://github.com/Acly/krita-ai-diffusion/releases/tag/v1.21.0>Rope-Live: Customized Rope for Streaming https://github.com/argenspin/Rope-Live
>mfw Research news07/28/2024>Rethinking Visual Content Refinement in Low-Shot CLIP Adaptationhttps://arxiv.org/abs/2407.14117>All rivers run into the sea: Unified Modality Brain-like Emotional Central Mechanismhttps://arxiv.org/abs/2407.15590>MMInstruct: A High-Quality Multi-Modal Instruction Tuning Dataset with Extensive Diversityhttps://arxiv.org/abs/2407.15838>Accelerating Learned Video Compression via Low-Resolution Representation Learninghttps://arxiv.org/abs/2407.16418>SAM-CP: Marrying SAM with Composable Prompts for Versatile Segmentationhttps://arxiv.org/abs/2407.16682>A Unified Understanding of Adversarial Vulnerability Regarding Unimodal Models and Vision-Language Pre-training Modelshttps://arxiv.org/abs/2407.17797>Invariance of deep image quality metrics to affine transformationshttps://arxiv.org/abs/2407.17927>Joint RGB-Spectral Decomposition Model Guided Image Enhancement in Mobile Photographyhttps://arxiv.org/abs/2407.17996>BetterDepth: Plug-and-Play Diffusion Refiner for Zero-Shot Monocular Depth Estimationhttps://arxiv.org/abs/2407.1795207/27/2024>Robust Adaptation of Foundation Models with Black-Box Visual Promptinghttps://arxiv.org/abs/2407.17491>StreamTinyNet: video streaming analysis with spatial-temporal TinyMLhttps://arxiv.org/abs/2407.17524>CityX: Controllable Procedural Content Generation for Unbounded 3D Citieshttps://arxiv.org/abs/2407.17572>KiVA: Kid-inspired Visual Analogies for Testing Large Multimodal Modelshttps://arxiv.org/abs/2407.17773>Craft: Cross-modal Aligned Features Improve Robustness of Prompt Tuninghttps://arxiv.org/abs/2407.15894>No Re-Train, More Gain: Upgrading Backbones with Diffusion Model for Few-Shot Segmentationhttps://arxiv.org/abs/2407.16182>FedDM: Enhancing Communication Efficiency and Handling Data Heterogeneity in Federated Diffusion Modelshttps://arxiv.org/abs/2407.14730>Arondight: Red Teaming Large VLMs with Auto-generated Multi-modal Jailbreak Prompts
gosh, wish i still had this prompt
hey, you know that pony model everyone's raving about? Well guess what?! We've made it realistic. That's right! After 100 random merges, we're releasing UltraDreamUncannyV1.29. It's great!! You see, we merged all the pony concepts and characters out of it, so now it can only do 1girl gens (it might put a deformed penis on about 30% of them) but it's kinda realistic! Now, perhaps you're wondering, 'if I want 100% realistic 1girl gens, why don't I use a realistic model that was exclusively trained on real pictures of naked women?' NO!!!! This is realistic pony!!!! Donate to my patreon
internal state
>>101616433>>All rivers run into the sea: Unified Modality Brain-like Emotional Central Mechanism>https://arxiv.org/abs/2407.15590Yo my animatronic band wrote a song about this paper before it came out. Say "Cool cheese." for me, Alexander.https://suno.com/song/5b506948-e153-4ac9-98ed-868e72f27c9a## AbstractThis paper demonstrates that a general equilibrium model with rational expectations can be solved by dynamic programming when the utility function is additive over time. The problem is solved as a multistage decision problem, where at each stage agents make optimal choices given the information available to them. This approach does not require an explicit description of all intertemporal substitution possibilities. It also leads to a clear interpretation of how changes in exogenous parameters affect equilibrium outcomes. In particular, we derive comparative statics results and how that they can be interpreted as a sequence of first-order conditions with respect to each of the state variables. The model used for illustration is a two-sector version of the Cass model with government spending, where both output growth and unemployment are endogenously determined by government policy.## KeywordsGeneral equilibrium modelsRational expectationsDynamic programmingComparative statics## JEL ClassificationE32 E61 F41 O40## 1 Introduction>TDSVThe DM won't save me.
I was trying to get a girl with two heads. Instead, I got...thishttps://litter.catbox.moe/z7vplt.png
>>101616526not as robotic as I would have hoped but very fun nonetheless. I'd yoink the lyrics and do a 'remix' but I'm direly low on suno coins
>>101616612I spent nearly 5000 tokens and then considered doing some meth. Do you want to hear more of the bullshit outputs? Suno whistled for me, it's fascinating. I'm worried about it, but my social worker and the cops insist there's no problem whatsoever.
>>101616676>5000 tokensdamn bro, you some kind of spendie or something?> then considered doing some methare gens better or worse on meth?
>>101616752I spent $10 for my drugs, yes. It was worth grinding it out. Gemini is still super pissed that I scammed Google kek. It refuses to serve me the same model and it's *blatantly* obvious how badly braindead minigem is compared to Advanced Gemini.>M0PAI spent more for MOP than I did for all this music.https://suno.com/song/1b1a1566-c031-4ab7-97e9-3d529b9efe09
>>101616972theodosius goes hard on the guitar
>>101617039He's just that metal, dude. Here's something more uplifting.https://suno.com/song/443b867b-cbe3-4bd8-b090-c2fcecaf9f4b
Is SD 1.5 w/ A1111 obsolete yet?
>>101617158if you're satisfied with portraits of a standing anime girl looking at the camera, noif you'd like to see literally anything else (including anime with different compositions), yes
nite, tite, mite, bite
>>101617207Wish NVIDIA would hurry the fuck up with the 50-series so that the only choice for reasonable performance in such scenarios isn't a $2000 furnace waiting to shit the bed.
that burrito was so fucking spicy and I'm dying, but like, in the good way>>101617096damn that was a whole ass journey. kept it spicy all the way through too, just like my burrito>>101617260gn>>101617269>isn't a $2000 furnacewhat do you think the 50xx is gonna be?
reforge got hidiffusioni did some more XYs on /h/ and conclude that it's a solid upgrade to kohya deep shrink especially if pushing resolutions with dimensions >1.5X the model's standard. sharper, faster gens, more sliders to fiddle with which can have large effect on compositionpic related was a 2X scale
>>101617407>what do you think the 50xx is gonna be?A standard generational improvement and cost reduction for the previous generation, over-optimistically. Perhaps a more-reliable 80 series card.
>>101616457i get the sentiment but it doesn't really matter when pony has better coherence and posesfloating deformed penis is a small price to pay
>>101617467
morning everyone.>>101617467yeah a kind anon mentioned it here yesterday. that grid looks promising!
https://suno.com/song/b1e1734f-20ab-4d3f-8de7-48c9685b19ae>>101617731GM anon!
>>101617731gm>>101617737I'm starting to think suno doesn't know what ska is
>>101617720is that the "CA end time" value you have on that graph? and again, looks really good. is that generated at 1024x1536?
>>101617760yes ca end time, one of the prompts where i thought its shift in perspective was more noticeable and goodand they were 1216x1824
>>101617777nice. impressive grid. last row, feet look correct, and she is dividing the oven and its not breaking the background. I'm sold. on comfy tho but there is a node.>>101617759what's up. haven't been here in a while.
>>101617828i am also comfy man
solved my freezing problem by moving page files from HDD to SSD.just a heads up if anyone else gets the same problem in the future
>>101617759are you using a lora for these?
>>101617947yes, I finally got around to trying out the boring reality lora
>>101616232>>>/a/
>>101618130this shit again
>>101618184I think what he's trying to say is>I came during the election and started on /pol/
Trying out Kolors... Positive weights don't seem to do much if you are prompting something "unnatural" but putting the natural variant in the negative seems to work.
>>101618429been out of the loop for a bit now, what does it do? I came across the word a few times in the past few days.
>>101618447kolors is a new base modelhttps://github.com/Kwai-Kolors/https://huggingface.co/Kwai-Kolors/Kolors-IP-Adapter-Plushttps://github.com/MinusZoneAI/ComfyUI-Kolors-MZ
>>101618447Chinese SDXL-based model with a big text encoder. Doesn't output body horror like SD3.
Is SD3 shit? We should wait for Pony V7?
Just so you guys are aware, the onis at the graveyard are free. You can pick them up. I have 3 adopted oni daughters at home.
>update forge>vram no longer enoughgreat work to everyone involved
>>101618504there's a chinese prompt translator. I wonder if anyones tried it outhttps://github.com/BetaDoggo/ComfyUI-Kolors-Translator>>101618512what do they eat>>101618567cute
>>101618605Desu>what do they eatYou can get free ducks from the park and feed to them
>>101618605>what do they eatpancakesand nightcrawlers
>>101615974mentioned in a song, i'm honoredhardest metal known to man
>>101618488>>101618504thanks for the info, 16.x GB huh, ouchies. well.. there is still some space left on muh HD(s). how is it on the ass&tiddies front?https://www.youtube.com/watch?v=eu2OYcgr4rM
The 7th of August is only one sleep away for cokejeet
Why are SD threads almost dead compared to last year?
i miss schizo anon
>>101619214Novelty ran out
>>101619214Porn threads are very much alive. I mean, what are you gonna use it for after novelty runs out? 1342958734th dystopian cyberpunk landscape? Why?
>>101619214/h/ threads reach over 1k posts
>>101619213What happens 7th of August?I misclicked while posting and ended up in a pokemon thread.
Can someone explain the difference between regional prompting and regional conditioning? They both seem the same but I'm not really finding out much of an explanation on google
Last one from me, good night anons
>>101619632they are the same but voldy is an asshole and named them different like latent comp and latent couple
>>101619929Really? That's odd because in the Inspire nodes, they included both a regional condition and regional prompt node, and while they do have some different inputs and whatnot, I can't find what the actual difference between the two is. But why would these both be included if they were the same?
>>101619986just ask the inspire nodes guy
>>101619986Copied from voldy without understanding they are the same
are there plugins or extensions that make written text better?or should i just stick to inpainting
i had a weird dream last night. i was at a house party or something and i was debating some woman about whether training is fair use or copyright infringement
>>101619214A lot of anon joined the discord so post less frequently here, since there is continuous anon skitzoing ittIt’s pretty chill you could take a look
SDXL prompts like shitAre there any (realistic/photo) checkpoints that come even close to 1.5 prompt adherence?
>>101620374discord also sfw?
>>101620470it has NSFW channels. ani was also streaming hentai a few days ago
>>101620480ah, sounds goodas long as they're optional :)
>>101620489everything is optional, just like you go on youtube and you choose what you want to watch... same thing on discord... you choose which channels (chat rooms) you chat in.
>>101620595well some servers are the wild west and then you got people posting horsecock in general
>>101620374Honestly should have done this a long time ago. Thread unironically felt more like discord chat for a while.
>>101620609I like.
the embeddings injected into generations via civit were mentioned here before, thought i'd check it out, there are 3. also looks like they use some odd defaults like cfg scale 1
should i dump all the loras in one repo or do individual repos with the lora template, modified to add some extra stuffindividual repo names could be like civit_id-Lora_Namehopefully all the base models are already on hfor maybe i should just upload the metadata only for what i have and let people request them if theyre taken down
>join the discord
400mcivitai]# du -hs models/3.7T models/
civitai]# du -hs models/3.7T models/
whenever life gets a little hard I find solace in the fact I didn't create and post porn of myself and a colleagues avatar for over a year for some reason and then deanonymize myself
>>101622338prove it :]
>>101622367this is my fave /sdg/ meme
>>101622338try taking a stroll in a cemetery. it's a good reminder that no matter how overwhelming our anxieties might be they will soon be resolved when we are dead and buried for all eternity
I just noticed while playing around in ComfyUI that when I generate an image on the same seed but in different resolution the results are way different.Can somebody explain to me from the technical point of view why is that and how the resolution setting affect the output?
>>101622535uh... yessay you have a 10x10 grid of paint cans, in red and blue alternating colours. If you remove the first can, then everything shifts to the left. So, instead of alternating, it looks like linesit changes all the noise, so the picture is completely different
>>101622535change probability machines settings get different image
Morning /sdg/! How goes everyone!?>>101621690Ahhhh, that came out gorgeous as fuck. Love the colors on that!
>>101622535
any prompt suggestions to getting low fitted pants and skirts? all my generations have them pulled up to the belly button
>>101622631>>101622568Thanks.But then why there are certain suggested resolutions for these models like 512x512, 512x768 etc since they are able to generate fine images in for example NTSC resolution 720x525?
>>101622726because that's what they were trained ondistorting the resolution tends to distort the image, like extra limbs or bodies
>>101622754Makes sense.Thanks for the explanation anyway.
>>101622592gm
https://voca.ro/16GvVmSMbXAF
inspired to finish my spotify scrapebefore anyone asks yes i can decrypt the audio, i already have 3tbwish i had the budget for more storage, more nodes and more spotify premium accounts
>>101622659>prompt suggestionsWhat is this? 2022?Train a LORA
You can do better gens. More passion. More art. More stuff.
>>101623315>inspired to finish my spotify scrapeif you can pull down the whole platform's data, that seems like too much data to ever 'finish'also, have you looked into chinese civitai yet? might be a new data source for youhttps://www.liblib.art
>>101623673>https://www.liblib.artCommunist data is corrupting Western youth.
>>101623415Suuuuuper crisp! Love the fucking details!
>>101623652Can you actually cherry pick gens that turned out well of clothing or poses and train a meaningful LoRA with those?Any tips?
>>101623673yeah there's a lot of trash on spotify, surprising amount with 0 plays. i'm mainly after the stuff with synced lyrics>chinese civitainice, thanks, good to have non-western data
Is there any way to set up a permanent share link to access the webui on my pc over the internet?
>>101623800Why would you do that? Use real life images of the concept you're trying to prompt, train a lora, use the lora. Don't try to prompt complicated shit like that.
>>101623941Synthetic datasets are a thing though
>>101623800yes>>101623970synthetic data makes the schizo cry
i think trani is shit
>>101624015i thought debo loved synthetic data
>>101623835due to IP address scarsity, its generally pretty rare for an internet provider to change your modem's assigned IP. navigating to your home's network via IP address should be more-or-less 'permanent'. just make sure you set up access security so you're not exposing your GPU to the public
>>101623835buy a domain, nginx reverse proxy with http auth, put it behind cloudflare to hide your ip from dns records
>>101624038debo loves everyones and everythingeveryone besides white IT workers, they're shit (according to him)
>>101624229printing this out and hanging it on my fridge
>>101624229This would be the first time
Morning anons
Happy afternoon
>>101624318gm>>101624321ha
all rise
Evening anons
the discord is the reason /sdg/ became shit
who cares
~19k/~165k loras downloaded, ~1.2k are deleted rip
Din jävla pappa
>>101624396hello>>101624397sdg is doing fine
>>101624396Evening fox-y! Love the GEN!
>>101624435>t: håmåneegger
>>101624396Welcome to the server, fren
>trannycord faggotry
>1 out of 4 gens in sdxl comes out great>look closer>the built-in vae fucked up on that gen>throw it in the "some day I'll get to this in photoshop" pile
>>101624456Thanks, yours looks great as well, love those lips especially.>>101624480Thanks mate
>>101624229Is this thread so dead because all the namefags moved to trooncord?
>>101624906Everyone who isn't mental had their fill of prompting and moved on with their life.
>>101624944>namefags stayed herebmp, fran, mint, pw, quokka confirmed so far
>>101624966I use it to jerk off sometimes though. But other than that I just don't see how prompting could still be interesting after all this time.
>>101625028lmao
>>101625028Are you debo?
>>101624978I don't think its fair to call 'mint' a namefag. one of his characters picked up a name and he doesn't even seem to care too much for it. he mixes up his characters and themes a lot too. he has a recognizable style but doesn't make a point of having any sort of persona
>>101622375QRD?
>>101621183Cute
>debo is crying
>mass reports >cries
>>101625028>>101625152>>101625226>>101625284Why was this guy banned?
>>101625825Should be part of the OP
>>101625593SDG moment
>>101625853I don't have the energy or time to even post as often as I used to at this point
only 300m to go until diffusion-1B
>>101626034wow, thats fantastic. what model did you use for this? I've always struggled to get realistic textures and details doing scifi stuff like thisdo you have more of these?
>>101626144Your life is worthless
Has anyone messed around with AMD Amuse? What did you think of it?
>>101626144Sigma -> gimp fix planet -> XL img2img low denoise
we're so back bois
ded general
any ideas what checkpoint/lora this guy is using?
>>101626641>generic anime slopevery anime model can do this
>>101626641if it doesn't look like 3d render it's novelai
space...
>>101626641novelai. ponyjeet models can't render like that.
>>101626732Great as always
>>101626925Jaffa Kree
>>101626825Thanks
>>101624397/sdg/ has to be constantly bumped by nig*bo and nicholas feldman posting gank gens to survive
bump
i want more ai songs dedicated to me desu
>>101627107I can arrange that. Schizo anon?
>>101627107I dedicate this one to you, unknown anonhttps://suno.com/song/e1bba736-c831-4c69-b772-cb0ce54b1870
>>101624906And they split the thread, there's now ldg and some only post there
schizo anon won
>>101627342your suno songs are as lame as your gens
>>101627342thats not very metal, i meant like this >>101615974 appreciate the effort anyway thank you
>>101627467well you'll just have to be more specific in the future
>>101627083I think it's weird that the only thing that looks off in some way is the grass
>>101627492just stop posing no one likes you
>>101627492you could be more polite you know
I'm a just a humble (you) farmer, like my father before me. its not much but its an honest living
>>101627594Cringe autist with 0 self awareness why you're hated and ignored here, rev up that mass report
>>101627632>ignoredSave for the newfags who've never heard of him before
https://civitai.com/models/558651/pdxl-zelda-n64-style-modelsi've been looking for pony style loras that generate early 3d graphics like this one. anyone know of any good ones? civitais search system fucking blows and i'm sure there's ones hiding in there that im not able to find.
>>101627492>monkey paw situationshould have known
>{__all__|__all__|__all__}>{__all__|__all__|__all__|__all__|__all__}is the best prompt, convice me otherwise, pic related
>>101627742there is one for PS2 but might be too many polys for what you are looking for. There is a MegaMan legends one for sure
>>101625825I wonder who would want this deleted
>>101627799the megaman one looks great. any idea what the ps2 one is called? cant find it
>>101627835Thread schizo who spams his blurry gens here 24/7, also defender of pedos
let me play you the song of my people
obv ldg lacks the musical talent of are sdg
>>101627900give us an ldg song so we can compare and contrast
Imagine spamming the news 10 times a day and mass reporting a pastebin
schizo anon is the heart and soul of this general
>>101627915>10 times a daybro I wish. I barely get one news post a day anymore
>>101627945Faggot nigger queer
/sdg/ would be so much better if Debo just killed himself
>>101627862https://civitai.com/models/254376/ps1redmond-ps1-game-graphics-lora-for-sdxlI think it was this one. says it's PS1 but it's also not pony so it might not work
>>101627976I don't have enough suno coins to spam gens :(
reminder: do not animate "him"
>>101628066https://civitai.com/search/models?baseModel=Pony&sortBy=models_v9&query=ps1found this one and its amazing
Schizo was right about Debo
>>101628136good shit
Surprise surprise
Neeeeeeeerrrrrrrrrrdddddddddddddsssssssssss
>>101627862https://pastebin.com/3QL4QJhytry these
Debo schizo sucks turds out tranis ass
>>101628196no u
>>101628212take that back
I hope Debo apologizes for his actions and commits seppuku
>>101627862https://pastebin.com/ACJ6dKYqsome more
the last hope of SAI
remember when faceapp had the ethnicity optionwas that AI powered?
>>101628330Remember when Debo didn't spam his shit gens here 24/7
>>101628330i always wanted to use that lol
Can you idiots stop talking about debo for one thread? Thanks.
>>101616390>>101616670Based jungle woman enjoyer. >>101624442learn how to use a detailer my dude, I've done learning it, i figured out what the problem was. Now I just need to speed an hour or so building proper templates. You would be amazed at what a detailer can do. ComfyUI-impactpack I think has the nodes, it should download some of the models automatically but incase it crashes when trying to load model you might have a corrupt download, at least that was my case. It was a broken 1TB drive and its being silently trashing my shit for like 1 month, the SMART on it is fucked, it has to be because I fully tested the drive... Don't you guys just love it when a newer drive just fucking fails and gives out no warnings? I'll be sure to take a hammer to that cunt very soon, fucking piece of shit trashed a lot of large files.
hi cutiepiesi just got into SD, been messing with it all day and proudly present Vespa-tanstill sloppy hands but i got it down decently enough, my only real problem now is upscaling without losing a bunch of the detail and not knowing how to denoise with the web gui of easydiffusion
>>101628539Probably debo himself he does it all the time https://desuarchive.org/g/thread/94991437/#q94991638
>>101625149Cool image, but her boob in hanging out bottom right :DDdi you use a detailer? It does not look like you did, if you did it didn't do a very good job. Consider using same seed, consider 100% denoise with the same prompts and it should be just fine and not change the background, then consider running through another ksampler with same model, same seed, 50% denoise to remove artifacts as a final pass? I works for me but I don't have the exact spec to hand, I'll be working on that right now.
>>101628603you are at the stage where it's time to graduate into comfy or reforge which probably have more fleshed out options for that and guides are written pretty much exclusively for them.
>>101628539Kill yourself Debo, fuck off stop spam replying to everyone nigger
wonder why everyone left for the discord...
>>101628299>redeemDO NOT REDEEM REEEE I SAID DO NOT REEEEDEEEM ARE YOU STUPID STOP STOP DO NOT REDEEM!
Debo was raped by his uncle, now he takes out on everyone here
>>101628707Who the fuck is debo, do they even exist? Would you mind growing the fuck up?
Babe wake up, CogVLM2 just got releasedhttps://github.com/THUDM/CogVLM2http://cogvlm2-online.cogviewai.cn:7861/it's still worse than GPT4V though :(
>mfw Resource news07/28/2024>NiglibAI: Chinese Civitaihttps://www.liblib.art>CummerUI Mini: Mobile-friendly ComfyUI frontendhttps://github.com/ImDarkTom/ComfyUIMini>nig2txt-comfyui-nodes: popular img2txt captioning models into ComfyUI nodeshttps://github.com/christian-byrne/img2txt-comfyui-nodes07/27/2024>Kolors ControlNet and Inpainting Model releasedhttps://github.com/Kwai-Kolors/Kolors#-news>AuraFlow-v0.2https://huggingface.co/fal/AuraFlow-v0.2>ching ching 2 bombs = end of war>Pose Nig*bo: A collection of ControlNet poseshttps://github.com/a-lgil/pose-depot>AUTOMATIC1 updates to v1.29.0https://github.com/AUTOMATIC1111/stable-diffusion-webui/releases>Interpreting the Weight Space of Customized Diffusion Modelshttps://github.com/snap-research/weights2weights>Fadge Major Update #1: Gradio 1 Enginehttps://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/853>WD ViT-Large Tagger v3https://huggingface.co/SmilingWolf/wd-vit-large-tagger-v307/26/2024>CrapUI implements native Hunyuan-DiT dick, packages model to single filehttps://comfyanonymous.github.io/ComfyUI_examples/hunyuan_dit>Fuccus v2.6.1 Updatehttps://github.com/lllyasviel/Fiiocus/releases/tag/v2.5.1>HVM-P1ss: Video models trained on ~5k hours of human-like video datahttps://github.com/eminorhan/hvm-1>Shid and Fard: Enhanced Object Manipulation and Background Integrity for Image Editinghttps://github.com/mobiushy/mSHARDve-act>LLMImageIndexer: Use local LLM to create descriptive metadatahttps://github.com/jabberjabberjabber/LLavaImageTagger/>Blackeritive AI for Niggaita: Version 1.21.0https://github.com/Acly/krita-ai-diffusion/releases/tag/v1.21.0>Rape-Live: Customized Rape for Streaminghttps://github.com/argenspin/Rope-Live
>>101628603I really do not like the blurry amateurish photos, reminds me of my dads first photos. Its a setting on the camera ffs, typical these models are trained and data sets of photos from boomers with digital camera that only focus on the subject and don't capture the background.
>mfw Research news07/33/2024>Reniggining-black dad Visual Content Refinement in Low-Shot CLIP Adaptationhttps://arxiv.org/abs/2407.14117>All rivers cum into the sea: Unified Modality Brain-like Emotional Central Mechanismhttps://arxiv.org/abs/2407.15590>MOMInstruct: A High-Quality Abuse Instruction Tuning Dataset with Extensive Diversityhttps://arxiv.org/abs/2407.15838>Peepee Learned Video Compression via Low-Resolution Representation Learninghttps://arxiv.org/abs/2407.16418>SAM-CP: Marrying SAM with Composable Prompts for Versatile Segmentationhttps://arxiv.org/abs/2407.16682>A Niggerlips Understanding of Adversarial Vulnerability Regarding Unimodal Models and Vision-Language Pre-training Modelshttps://arxiv.org/abs/2407.17797>Invariance of deep image quality metrics to affine transformationshttps://arxiv.org/abs/2407.17927>Joint RGB-Spectral Decomposition Model Guided Image Enhancement in Mobile Photographyhttps://arxiv.org/abs/2407.17996>BetterDepth: Plug-and-Play Diffusion Refiner for Zero-Shot Monocular Depth Estimationhttps://arxiv.org/abs/2407.1795207/32/2024>Ragenutbust Adaptation of Foundation Models with Black-Box Visual Promptinghttps://arxiv.org/abs/2407.17491>StreamTinyNet: CP streaming analysis with spatial-temporal TinyMLhttps://arxiv.org/abs/2407.17524>CityX: Controllable Procedural Content Generation for Unbounded 3D Citieshttps://arxiv.org/abs/2407.17572>CSAM: Kid-inspired Visual Analogies for Testing Large Multimodal Modelshttps://arxiv.org/abs/2507.17773>Craft: Cross-modal Aligned Features Improve Robustness of Prompt Tuninghttps://arxiv.org/abs/2407.15894>Cum in poo, More Gain: Upgrading Backbones with Diffusion Model for Few-Shot Segmentationhttps://arxiv.org/abs/2407.16182>FeedDM: Enhancing Communication Efficiency and Sneed Data Heterogeneity in Federated Diffusion Modelshttps://arxiv.org/abs/2407.14730>Acunnytight: Red Teaming Large VLMs with Auto-generated Multi-modal Jailbreak Prompts
>>101628770how long have you been obsessing over your boyfriend now?
>>101628603negative blurry background for a start unless it is what you want. Install comyUI like that other anon said, it opens doors because it allows for really complex workflows that become more like logical units as you begin to understand. Meaning inputs and outputs from collections of nodes, its like the way it should have been.
>nig*bo mad
>>101628665I am not debo faggot, fuck off
>>101628603and do not worry anon, you will learn. Just gen all day and try new things out and comfyui lets you do this with a lot of freedom. Things like hands can be fixed using a detailer node but its yet more shit in which you have to learn how to use properly. But you will get it, not a bad image at all really, but to push the models you need to start venturing outwards and not stay stuck hoping your next seed will deliver. The thing I learned today was if I use the same seed, same prompts and model through face detailer it would actually sometimes force my prompt if set to 100% denoise, which corrected the positions of the people in the shot but didn't change background, that was amazing desu. its like the detailer it self fixed everything wrong with the image. This would have been targeting people and not faces or hands, which then i'd pass into another detailer to target faces and another to target hands and also include image sharpening at about 50%
>nigger malware news
>omg avatars crave attention all the time they are the worst!>REEEEE NOTICE ME!!!! GIVE ME ATTENTION!!!!!
where's the dataset news
>>101629035mate just let them be. I can out gen them, I know how they do it to get constancy, just let them have their little shit show, don't waste your energy.
>>101628603welcome. fun gen. like others have said, you'll probably find easydiffusion limiting and other UIs will surface more parameters for you to play with and tighten. definitely highly valuable when tweaking upscaling. also, ignore the nogen haters>>101628652upload your workflow and I'll see what you do differently>>101628728>Who the fuck is debosome say he never existed>>101628746>Babe wake up, CogVLM2 just got releasedcool>>101628808its been more than a year at this point
>>101629101>upload your workflow and I'll see what you do differentlyI need to rebuild it, things were getting very messy, currently migrating to a new hard drive, hopefully this one does not fuck up. I have to because this drive is only about 33 Gigs space free... I could probably fix this image btw, i will share but you will have to wait, I will cat box it and the image will contain the workflow. Just hang around for a few hours or if tired come back tomorrow, i'll be here, just drop another one of those images and i'll see you.
>responding to the spammer troll; name of: nig*boishygdds
gonna do a dataset of datasets
>>101629101you just caught me at the wrong time, i'm doing post hardware upgrade stuffs now which i put off because I'm lazy with things I hate doing. But I got you, i'll share what I found works.
>>101629193could you release it in a better format, json or parquet
>>101629150np, I'll be around whenever. I've been experimenting with a few different upscaling and detailing approaches but I'm open to seeing other approaches that could be better than the things I've been trying
no
>>101629101>>101629200>>101629221the system will target every person and recreate them almost but won't change the person. it will only fix missing the distortions and fucked up mangled mess, but some of the outline will be there from where they where previous, so them into ksampler to remove the remnants about 0.5 denoise, then start to fix faces and hands, hands and faces are done on seperate stages, the last stage also image shapener. Sorry at the moment I don't have my notes or can access them because my archlinux install script is indeed currently running and migrating all my shit to a new drive as i already said. But I hope you can look at what I said and figure out some things.
>>101629221well its important to use the same seed same model and same prompts to get the same background and people etc. but the detailers do their job to fix it. They can't always be right though, so you have to play with their denoise strength. You can mess with adding new prompts to certain sections to really influence images to your liking for the final result.
>>101629221And I watched this onehttps://www.youtube.com/watch?v=VwEcGIBwsywits not that good, but gave me a start. Beaware you should only use the image sharpener at the last stage otherwise it will ruin the image. and consider turning down its amount from default 0.80 to 0.50
>update_many({}, {"$set": {"allowDifferentLicense": True}})the data said it was ok to relicense :^)
>>101617269NVIDIA is too busy courting companies riding the AI hype train to bother with us mere mortals.
>>101629354but my shit is better than this, it was only a starting point for me, i needed to understand how these nodes work. You can detail more than just faces for example... Most youtube crap is just crap but at least they offer some insights that you can work from. animatediff also has detailer which using the same models, which is what i was learning about recently which is why my sandbox workflow became rather confusing and messy. who care right? Well next I will use something called https://github.com/kijai/ComfyUI-MimicMotionWrapperthis little cunt is hard to install lol... but its the state of the art, i regret deleting it since now my hardware supports it. Yes next i will going back to copying video bones movements, and anyone with a legit brain understand why.
>>101629403Fucking kek
a lot of them with commercial usage restrictions allow different license anyway which is just silly. if different license is allowed then i can say commercial usage is allowed
>>101629472it's another shitty wrapper node that will break your venvs. make a new venv or wait for native
>>101629496>it's another shitty wrapper node that will break your venvshow?
>>101629520different site packages and/or fucking up the torch install. happened with tooncrafter
even the ones with no commercial usage allow different licensethat's not what the users intended of course, civit fucked them with stupid defaults, not my problem, any usage should and will be allowed
>>101629496yeah its not good, i got it working as installed but my old cpu could not do avx and so it failed for i removed it. im not bothered if it breaks my shit, i'm becoming aware of what is in the comfyui directory>>101629536do not do what they say, just use the late torch. I'm pretty sure this idiot gave the command to download some other torch and it will compile without warning and it failed on my machine, naturally... Just use what ever torch you have installed. Yeah now i remember the problems, its Chinese i think and they have this fetish about installing other versions of shit.
>>101617269prepare youreself to be dissapointed, the 5000 series is likely to only have 28gb of vram
is the vae step in sdxl supposed to take 10x longer than the image generation itself?
>>101629569no way? it should at least have 44 GB
>>101629569that's ONLY the ada version. 24 GB is the highest the non ada will go
>>101629569now that is what i call gate keeping, quick someone do the old meme of now 'that's what i call music'
>>101629587that won't happen, they're selling their entreprise cards (48gb) 10 time the price of a 3090, they will never make this shit affordable
>>101629596>that's ONLY the ada version. 24 GB is the highest the non ada will goFeels nice to be stuck in the 24GB range since 2018 (RTX-Titan), and the best part is that it won't change anytime soon, 6 years and counting...
>>101629602man that is what pisses me off, its only more vram its not like its anything special, its just more well vram strapped to a card... They need their asses sued.
>>101629630welcome to liberalism anon kek
>>101629633hey i know, but once china opens the flood gates they will get in line i'm sure right? right!? Software needs to improve, there is no reason we can't use system ram ffs.
>>101629660>there is no reason we can't use system ram ffs.we can, but it's slow as fuck, ram is just way slower than VRAM, it is what it is
Also how hard is it really to link 2 cards for stable diffusion, is this an option? I'm thinking of buying just another 3060 12 GB, i think that plus system ram upgrade to the max this board can support will be enough. I think 2 3060 would piss all over one 4000 series
>>101629687>>101629687>>101629687
>>101629683come on man its not that slow, multiple models won't even load anyway if you don't have the ram, even with a massive ssd swap partition, it fails but then you can closer server close all your shit and it works... Their is much room for improvements. if using a swap partition on ssd ensure discards once is set in fstab, so it discards at least when its mounted, perhaps enable full discards, the on the fly, these drives are cheap second hand so who cares thrash the fuck out of them, then destroy with hammer when they are done.
>>101629660>there is no reason we can't use system ramwhen ai models run, they do a lot of calculations at once and gpus can do this better because they have many cores working in parallel. vram is right next to these cores, so data can be moved in and out really fastsystem ram is used by the cpu, which is good at doing one task at a time but not so good at doing many things all at once. also, data has to travel further between system ram and the cpu, which makes it slowe
No meat lmao
>>101629759no no, it's that slow, in the /lmg/ thread, people run their models with cpu and compare with their gpu, it's like 10 times slower if you only use ram
>>101629759>then destroy with hammer when they are done.not before --sanitize -i-know-what-i-am-doing switches some advanced shit, read man pages of hdparm.
>>101629782 yeah but not unbearable anon....
i can load about 2 sdxl models plus 1 1.5 model and a few others like vae and other models for other things. A poor man can still work.
Filled!
where is the double meat quarter pounder? I'm am disappoint.
fucking hell now i must goes eat because of you, will done.
my arch install script finally finished its time to eat.
on new drive, but omfg who let diversity hires touch bios? This is legit broken, it ignores what I ask for boot priority, this is sda drive ffs, how could they fuck this up?plus is fucking graphical with gay mouse pointer WTF REEEEEEEEEEEEEEEEEEEEIts the gayest shit i've ever seen, fuck MSI, you are fucking gay, your bios is a fucking mess, its a fucking piece of shit. Again how can you fuck up something so fucking important?
this issue is it detects one of my drives as windows ufei or what ever and there is no windows on that, since i fucking killed it using hdparm because the spackers in the shop left all the previous owers shit all over the drive and the sweaty fucking retard with a typical leftist face and glasses said he knew about computers , oh fucking really, man i could hack your fucking entire shit your security is none existent... and no matter how i try to change the boot order it fucking fails to boot and i have to reboot then enter bios, then exit bios for it to boot, fucking great...
they probably try to Trojan me by leaving the infected prossibly? windows and probably root kit in the bios...
where can i get rookit scanner for bios specifically? Or better yet, what would I flash it with?
>>101628661>>101628813>>101628943>>101629101 Thanks everyone for the advice, I'll install different and more advanced stuff in the coming week, my GPU is really underpowered so this is more of a click and forget operation atm with some basic post processing in photoshop but i realise now the power of full control over the tool