Classy EditionPreviously on /sdg/: >>100115740>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 forksAnapnoe 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-inpainting >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/sdg>>>/d/ddg>>>/b/degen>>>/vt/vtai>>>/aco/sdg>>>/trash/sdg Official: discord.gg/stablediffusion
>mfw Resource news04/21/2024>FlashFace Inference Code Releasedhttps://github.com/ali-vilab/FlashFace>ComfyUI MagickWand: Proper implementation of ImageMagickhttps://github.com/Fannovel16/ComfyUI-MagickWand>Moving Object Segmentation: All You Need Is SAM (and Flow)https://www.robots.ox.ac.uk/~vgg/research/flowsam/>Image Effect Scheduler Node Set for ComfyUIhttps://github.com/hannahunter88/anodes/>ComfyUI-Tripo: Generate 3D models using the Tripo APIhttps://github.com/VAST-AI-Research/ComfyUI-Tripo>Conditional Prototype Rectification Prompt Learninghttps://arxiv.org/abs/2404.0987204/20/2024>Basic Stable Diffusion API GUIhttps://github.com/ThioJoe/BasicStabilityAPI-GUI/>IPAdapter Advanced Weighting support added to sd-webui-controlnethttps://github.com/Mikubill/sd-webui-controlnet/discussions/277004/19/2024>Customizing Text-to-Image Diffusion with Camera Viewpoint Controlhttps://customdiffusion360.github.io/>StyleBooth: Image Style Editing with Multimodal Instructionhttps://ali-vilab.github.io/stylebooth-page/>Sketch-guided Image Inpainting with Partial Discrete Diffusion Process https://github.com/vl2g/Sketch-Inpainting>ComfyUI ImageMagick: Image processing powered by ImageMagickhttps://github.com/jtydhr88/ComfyUI-ImageMagick04/18/2024>Meta has releases meta.ai, Multimodal AI including image generationhttps://www.meta.ai/>Stability AI lays off roughly 10 percent of its workforcehttps://www.theverge.com/2024/4/18/24133996/stability-ai-lay-off-emad-mostaque>Stability API nodes for ComfyUIhttps://github.com/Stability-AI/ComfyUI-SAI_API>Dynamic Typography: Bringing Text to Life via Video Diffusion Priorhttps://animate-your-word.github.io/demo/>InFusion: Inpainting 3D Gaussians via Learning Depth Completion from Diffusion Priorhttps://johanan528.github.io/Infusion/>Factorized Diffusion: Perceptual Illusions by Noise Decompositionhttps://dangeng.github.io/factorized_diffusion/
>>100122230ty for the bake quokka
>mfw Research news04/21/2024>Prompt-Driven Feature Diffusion for Open-World Semi-Supervised Learninghttps://arxiv.org/abs/2404.11795>MultiPhys: Multi-Person Physics-aware 3D Motion Estimationhttps://www.iri.upc.edu/people/nugrinovic/multiphys/>ProTA: Probabilistic Token Aggregation for Text-Video Retrievalhttps://arxiv.org/abs/2404.12216>BLINK: Multimodal Large Language Models Can See but Not Perceivehttps://arxiv.org/abs/2404.12390>Generating Human Interaction Motions in Scenes with Text Controlhttps://arxiv.org/abs/2404.10685>Dual Modalities of Text: Visual and Textual Generative Pre-traininghttps://arxiv.org/abs/2404.10710>DreamScape: 3D Scene Creation via Gaussian Splatting joint Correlation Modelinghttps://arxiv.org/abs/2404.09227>AesExpert: Towards Multi-modality Foundation Model for Image Aesthetics Perceptionhttps://arxiv.org/abs/2404.0962404/20/2024>Align Your Steps: Optimizing Sampling Schedules in Diffusion Models https://research.nvidia.com/labs/toronto-ai/AlignYourSteps/>Visual Prompting for Generalized Few-shot Segmentation: A Multi-scale Approachhttps://arxiv.org/abs/2404.11732>Partial Large Kernel CNNs for Efficient Super-Resolutionhttps://arxiv.org/abs/2404.11848>From Image to Video, what do we need in multimodal LLMs?https://arxiv.org/abs/2404.11865>GhostNetV3: Exploring the Training Strategies for Compact Modelshttps://arxiv.org/abs/2404.11202>ANCHOR: LLM-driven News Subject Conditioning for Text-to-Image Synthesishttps://arxiv.org/abs/2404.10141>StyleCity: Large-Scale 3D Urban Scenes Stylization with Vision-and-Text Reference via Progressive Optimizationhttps://arxiv.org/abs/2404.10681>Weight Copy and Low-Rank Adaptation for Few-Shot Distillation of Vision Transformershttps://arxiv.org/abs/2404.09326>Exploring Text-to-Motion Generation with Human Preferencehttps://arxiv.org/abs/2404.09445
thanks baker
Pic related is an example of what the ELLA research found if anyone hasn't seen it: https://ella-diffusion.github.io/It's my understanding they actually developed an additional text interpretation layer that can be put on top of SDXL and SD 1.5 in order to improve prompt understanding. I believe someone already developed ComfyUI support for it. I couldn't find anything about A1111 or Forge support though.
>>100122263>>100122249Yw anons
>creates most degenerate image model for furries >>w-why didn't oai or nai hire me???Why's he so retarded?
>>100122303Some months ago someone posted a python script tagger that uses llava along with a local LLM (the comments suggested to use vicuna-13b which is what I still use) to create captions for images. I have used for a while (modifying the prompt to give it advice about the dataset) to good effect. It is definitely better than any natural language tagger I have used before that. I can only imagine a better language model could do a really good job.
>>100122361Wow, hilarious if true. Can't wait for another model to replace PonyXL.
>>100122375That's exciting to hear and would be really interesting to see! Do you have a link to it handy?
>>100122378I love when AI can replicate mundane pictures so well.
>>100122361furfags, not even once
>File too large (file: 4.01MB, max: 4 MB)
>>100122431crop it you fucking tranoidjust like you cropped your cock off
>>100122390Unfortunately I do not have a link to the original. I have mine and will share it, but I erased the lengthy tutorial comments at the beginning. This is the script:https://pastebin.com/JKNhcGttThe section "CAPTION_PROMPT_TEMPLATE" is what you modify based on the image contents. As you can see I was tagging images for a musical instrument lora so when I ran it through a 'sousaphone' folder I modified the prompt to say "each image contains a sousaphone" to nudge it a bit. This helps a lot with certain concepts it might get confused by.The way I run it is, first start llama.cpp with the following settings:> .\server.exe -m ".\vicuna-13b-v1.5-16k.Q5_K_M.gguf" --mmproj .\mmproj-model-f16.gguf --host 127.0.0.1 --port 8080 --n-gpu-layers 100"mmproj-model-f16.gguf" is the llava model.Then run the python script as:> python .\caption.py --dataset "/path/to/data" --completion-api "http://127.0.0.1:8080/completion"It will create two folders, one with a plaintext caption for each image and another that is a list of keywords, similar to a booru tagset (but not obeying booru tags precisely). Sometimes it gets confused for an image, it will spazz out, retry several times and eventually give up (for that image but proceed through the set). You will see the errors in the log when that happens and will need to manually caption that one.
>>100122361This would be Trani if he were honest about his current employment status
>>100122361its really sad that this nigger removes artist tags, doesn't accomplish anything other than gimp the model, now he will try to ship a model with some pre-built styles to compensate (spoiler: it won't serve any purpose and people will still need 10000 loras)
>>100122484This is an extremely helpful post and I really appreciate it! Thank you kindly!
ESG approved sniffs
>>100122510Np, hope it helps. It is not perfect however and it would be wise to at least give a quick lookover to each caption. Despite the prompt telling it not to provide subjective speculation like "unique and entertaining visual" it will still sometimes put those kinds of words into the caption.
>>100122472you get jpegs instead. I'm a man btw>>100122484>>100122536ty anon this was a nice share
>>100122492Next he's unironically going to remove NSFW in a bid for YCombinatior cash.
trani is shit
>>100122698Well I am legit glad and thankful he gave us the model. It takes a lot of time, tedium and money to make a fine tune like that. I hope we continue to get them. It sounds like he is personally moving in another direction so we may have to wait for the next person to step up to the plate.
>>100122723>>100122743why do you fantasize about ani being a tranny faggot? more importantly, why do you do it all day and try to interrupt tech discussion?
>>100122698I unironically hope he does so that everyone will stop using his model.What's most interesting there is that he talks about storage. It sounds like the entire dataset for PonyXL fit on 100TB? With a good deal you could get a 10TB hard drive for around $100, so it'd probably cost around $1000 if you wanted to store the data set locally. That's good to know.
>>100122759I don't think there's anyone in the universe gayer than the spamming Ani hater given his eternal obsession with one person and homoerotic overtones in the process.
In more important news the steam deck fully works on linux now so if you have one as a gift you can make it into a monitor/hotkey machine
>>100122797>she thinks it's one person
>>100122698lmao, everyone was praising this faggot and shitting on emad for 'cucking' and he doesn't even like the main use case of his own degenerate model and does the same thing to appease VCs
hello sdg>>100122431>>100122631super cute
>>100122759ani is a proud gender dysphoric
>>100122828>makes a transsexual avatar with his eye color>open talks about his futa obsession >made a Lora to replicate the style of a artist that turns men to womenI wonder what the next cope will be from the Trani defender
>>100122786i hate trani but what is this julien stuff?
>>100122859>closeted anon fighting real hard to keep his repressed emotions bottled up
>>100122866He doxxed himself trying to flex a job offer, retard used photoshop with his own personal account.
>>100122723Kek'd
>>100122859what is with you and this eye color thing? you've been trying to make it a thing for more than month. its never gonna be a thing
>>100122906>debo running defenseSay the line again debo
julien is beans
>>100122915not everyone who thinks you're a schizo is debo, psychopath
>>100122825>super cuteyours too!
>>100122938>it's only one anon>I mean it's multiple but I'll say it anyways>Ani is not gay btw>so what if he post tranny porn of himself fucking another man he knows irl
>>100122945*licks toes*
>>100122961>licking Trani toesKek
>>100122884is he ((()))? sounds like a (((name)))
>>100123003it's not even him
>>100122945The chibi kingdom doesn't acknowledge you as princess unless your crown is made of chibis
>>100123009Yes a it's perfectly normal to use someone else's computer to *check notes*>post a edited version of your job offer>proceed to spam futa
https://www.youtube.com/watch?v=JBQwSprUSfY
>>100123028>The chibi kingdom doesn't acknowledge you as princess unless your crown is made of chibisfair lol. what pass did they all appear on? kinda impressed, it's hard to make it happen without inpaint
>>100123123Topkek, I emailed you a picture of my dick
>>100123042>t. it's perfectly normal to obsess over a literal who
City is flooded for some unknown reason in this one gen of today's batch.
her huge doughs won't knead themselves, anons
>>100123152Excuse me sis but he's the self appointed line leader :)
>>100122698It's tragic to see even in a horsefucker. This guy might leave a day job for something that will most likely fry his nerves, and yield chump change revenue against a pile of liabilities while YC gets all the useful IP for cheap and liquidates their holdings, before he's left in the dust high and dry. Hopefully the journey at least yields decent prospects for another corpo job and he doesn't take on personal financial debt.
>>100123194what is a line leader?
>>100123185nice gen
>>100123189Those buttons look a little strained.... It would be a shame if something... happened to them...
Weeb modelhttps://twitter.com/SakanaAILabs/status/1782207884170080407
>>100123216I'm thinking of a yc startup myself, what did you mean by this??
Normally I would speak against this but I'm at the apathy phase
>>100123216YC won't touch him with a borrowed 20ft pole. He'll need to start his own thing and pray to allah an angel investor or seedy VC funds come through.
>>100123245against what, tranjak?
>>100123268Name doesn't work because I don't post dicks everywhere. Stop trying to deflect the drama away.
>>100123271ok tran
>>100123271Cope Tranjac. All avatarfags are Tran- in spirit
What did you discordfags do to make this schizo so unhinged?
>>100123323why do you obsess over trans? are you gay?
>>100123209>>100123151>I emailed you a picture of my dicki wouldn't have done that, he'll turn you into a furry futa as revenge
>>100123244I cannot speak against YC specifically. I mean venture capital is often a raw deal where the VCs structure for maximum leverage and privileges on IP, and any situation where you'd probably have to gut the value of your core product to appease their sensibilities (like taking the smut out of your smut generator) won't end well. If you do your diligence, think the terms are good, and are confident in the venture then don't let me dissuade you.
>>100123123I just reran it. It looks like the body detailer step causes it. Here it is before the detailersI normally do SDXL -> 1.5 -> body detailer -> head detailer -> hand detailer in the hopes of never having to inpaint. I could just lower the denoise strength as well
>>100123372I think it's funny it just thought the gems looked like little chibis. cute gens regardless!
>>100123489I have the threshold set as low as it'll go so it detects anything that remotely looks like a body. It helps with the big crowd pictures, because some of those 1st/2nd pass bodies don't even resemble people lol.I always forget to change it between pictures
Does 3.0 work on the webui?
There are 3 pictures of ani on the webcan you make a lora with that??
>>100123028 >>100123123cute chibis, also the crown is great.
>>100122378holy fuck this is amazing
>>100123626>>100122959>>100122859Alright im curious, can someone provide the lore behind this person? Curious why all the hate
what would comfy gain by locking users out of 20% ish of their vram?
>>100122230lmao you took forge off the op
>>100123731prevent OOO so that he can gloat
>>100123746OOM error
>>100123746oh yeah, users will definitely prefer to go into low vram mode instead.
>>100123722legitimate schizophrenia
>>100123765Pleasu understand cumfy needed that win against voldy
>>100123731>locking users out of 20% ish of their vram?what? I thought comfyui was the most vram efficient
>>100123770>dekmFuck off avoiding filters
>>100123722he:>wrote guides on how to animate in the early days>helps anons with SD problems>posts anime girls>got excited about getting a job making anime and posted about it>is training models to share with everyone>is friendly with anons that frequent hereclearly the antichrist
>>100122698h-horsefucker sisters.....
well okay then AI
>>100123779not at all, since he got his 3090 from SAI things slowly changed, first it kept models loaded in vram, which caused more advanced workflows to oom. now it locks you out of 20% of your vram and sends you into lowvram mode.pretty much it keeps unnecessary shit from earlier parts of the workflow loaded into the vram while running the later parts of it. comfyui is targeted toward corpo backends now, please understand they have to save the 5 seconds it takes to load all that stuff. also no such thing is present on forge
>>100123815Hmm, maybe he'll fix it when sai goes bust and the ui is a hobby again
>>100123802>Currently the no.1 attack dog for cumfy will go on multiple platforms to defend his boyfriend's honor>famously made a ass of himself threatening the creator of controlnet on github over forge>gets banned constantly for posting futa>tried to control the OP on multiple occasions>Used to blog about his depression, buying prositutes and his homosexual IRL adventures with cumfy
>>100123824not really the 20% vram lock out was the straw that broke the camel's back for me. why? for everything else there's a reason, even if it's a bad one. but for this, i have no words.
>>100123841He's in hiding because SAI is in flames and he used to brag about being employed by them
>>100123743Sorry haven't made threads on a while and I don't have the most updated version of the pasta, if we know where I can get it, I'll appreciate it.
>>100123853Copy previous OP
>>100123832scraping the bottom of the barrel for reasons to dislike someone
>>100123853ah I was hoping it was a cheeky change since the guy wont fucking update it
>>100123743good it's dead anyway
>>100123722He's harassed some devs that compete with comfy ui and threatened to report their repos because he's a retard.Plus the endless and inconsistent larps about animation jobs he doesn't have and posting gay porn a lot.
>>100123864Being a janny for comfy while begging to work at SAI is good enough of a reason to dislike him
>>100122698Kek where is the business they're aiming at? Some where a ton of venture capital shows up?
>>100123802you unironically described why he's the schizo's target. ani is friendly, helpful, smart, and successful. qualities that schizo can never possess. so he seethes and devotes his life to lashing out. since he can't be happy, all he can live for is trying to make other people as miserable as he is
>it's another DEBO post
anyone who posts in SD threads are trannies so I don't see a point in talking about it
>>100123770>>100123802>>100123864>>100123895major damage control
>schizo is crying
>>100123917>implying you're doing any damagelol, get over yourself
>>100123802His:>"animations" are shit>gens and advice are garbage (lol, don't use score on pony?)>posts trannies>talked out his ass about working for mappa, then netflix, then every other animtion studio on earth >is making crap loras anyone could do>is a cunt to everyone unless he's avaterfagging, then becomes a passive aggressive piss ass instead clearly a fucking faggot
>>100123937tbqh anyone you associate with is delt damage via your association
>schizo when going without a (You) for one minute
>>100123953fucking kek
>>100123864Prompt? I've not got good mirror gens with sd yet
>a-anyone who dislikes the obnoxious Avatar clique is the same person!!!!Ironically schizophrenic
>>100123964not a gen, sorry
>just because I say the same stuff hundreds of times doesn't mean I'm the same personeither you're deeply retarded or you think we're deeply retarded
>>100123998Schizo, take your meds
>>100123942what loras did he make?
>>100123848>He's in hiding because SAI is in flames and he used to brag about being employed by themGlad I'm not the only one who noticed
>>100124017some lineart one he supposedly sold to all da anime companies for millions
can someone post a "good" tagged dataset for a person/celebrity? I want to do better.
>>100124020s-schizo, don't talk bad about my boyf- I mean comfy like that
>>100123971i think it's a case of him accusing others of what he does himself. both of them constantly pose as anon or as another avatar, so they just assume anyone who dislikes them must also be doing the same
TAKE YOUR FUCKING MEEEEEEDSSSS
>>100122698Kek do they think YCombinator gives 500k to people without even superficially one of the most viable business plans?
>>100124070you can't make me>>100124096VC gives way more for way less. you just need to be adjacent to CURRENT_BUZZWORD and you get free money
So what's the best way to tag images nowadays?
>>100124156weebs use WD (waifudiffusion) tagger. trying to figure out 3d myself.
>>100124116No you also need to give US boomer investors the idea that they'll get one of the companies that might turn into a basically monopoly over a part of the market. The one dominant ride sharing company, the one dominant producer of the cheapest most automated blood tests.YC surely also wants companies that look good for purposes of such investment given that they surely aim at making money off their early 7% particularly from getting money soon when investors jump on board - not when the 7% have earned over 500k via dividends.
>>100124116YC is VC you halfwit, and you're not getting "way more" from a pre seed stage than YC gives
>>100124156I'd use taggui and wd14 v3 [vit?] and/or llava / cogvlm,
>>100124182People say they use an LLM to tag a lot of the newer shit, but there's no fucking way I'm feeding my dataset to a server.
>>100123953
>>100124185idk, I've seen all kinds of shit companies get incubated. I don't think VCs have the level of standards you think they do. but its usually a lot more about who knows who than whether the ideas are good or not>>100124193>YC is VC you halfwitIk, I'm using them interchangeably. work on your reading comprehension if you wanna call other people halfwits
>>100124195cogvlm requires 16/32/80gb depending on quantization T_T
I’m using the same model, prompts, loras, and settings as the user images on Civit, but my outputs are completely different. What am I missing?
>>100124104that would be cooler if the feet and legs didn't contrast with the rest f the body so sharply
>>100124195>>100124237Checked them all, Llava was fit for my purposes.
>>100124039anyone?
>>100124229>I'm using them interchangeably. work on your reading comprehensionNo you're not.You said "VC gives way more for way less." implying YC and VC are separates and you're comparing them. Plus you're completely wrong about VC giving way more, dear pony fucker would be lucky to get half of what YC gives for twice the equity from a pre seed round. If he even got it, which he wouldn't with no viable business plan and his claim to fame being a beast finetune of SDXL.
>>100124253can you make the girl a slimegirl in those realistic 3D clothes?
>>100124104Btw, never got b& you weasel faggotroon
>>100124264true, too much paperface. I'd have to block merge 3D stuff in or train a lora for 3D renders>>100124290>his claim to fame being a beast finetune of SDXLto be fair, any idiot can scrape, autotag and overtune a model with multiple boorus if they got showered in money and compute too
its pretty crazy how good some anons are at imggen we are blessed they post
>>100124329I find it amusing you never feed schizo and he gets desperate for attention
>>100124243hires or detailer, usually>>100124290>implyingjust because you read it doesn't mean I was implying it. spend more time reading books and less time arguing with people on the internet. >>100124332its my pleasure
>>100124329You might've misread. I said beast, not best. No venture capitalist is gonna throw their money at ai animal porn
>>100124349Have you ever once admitted to being wrong you imbecile, or is it always deflection?
>>100124353>You might've misread. I said beast, not best.oh my bad that is the case kek
>>100124229>>100124349>worlds lamest attempt at gaslighting sad
>>100124373all the time, but only when I'm actually wrong>>100124402is she eating the slime monsters? that can't be healthy
>>100124373>having an argument with debo in bad faithHow fucking dare you
>>100124411>>100124387
>>100124411Lmao. Post a single example of this happening.
>>100124437am I supposed to be keeping a humility log? first I've heard of this. also, can you post some gens? I'm quickly getting bored of talking with nogens
>>100124237yea but you can use system RAM so it's not THAT extreme>>100124269good then
>>100124474debo ffs stop fucking replying to him
>>100124494ok
>>100124494>debo resorting to replying to himselfhe's fucking losing it
>>100124494DEBO-KUN HE'S GONNA BTFO YOU S-STOP
Newfag here, any tips on how to generate hyper realistic images? I'm using Juggernaut right now.
>>100124474Any normal, well adjusted person should be able to recall their latest act in humility.
Good evening, anons! I hope everyone is doing well :]
>>100124535>Any normal, well adjusted personAnon...
>>100124494Kek. it's funny seeing the discord cabal trying to tardwrangle window licker demon kid
>>100124532hyper realism tends to come in through the hires/upscaling steps, so you could start exploring that>>100124543hi pw, nice to see you
>>100124532go their civit page, sort images by newest, scroll until you see a good pic. copy/modify prompt and explore.
>>100124543hi pw
>>100124544Cute purple witch :]>>100124558Heyy Debo! Nice to see you as well! I hope you're having a great day!>>100124603Heya Blue fox anon!! :DGreat to see you again! Comfy gen! Haha
>>100124349thanks, kimosabe >>100124121do you have to roll a lot for the faces? i remember the author of either the lora or checkpoint saying faces were bad
>>100124532realistic modelupscaling image with GAN + secondary sampler (hires fix)prompt styler / analogous prompt
>>100122303And what about dreambooth?
>>100124677adding too much of the boring reality lora seems to mess up faces and everything in general. I can use boring reality at 100% for simpler pictures without people, but I use around 50% otherwise. at the same time faces in sds_film look like they have more artifacts and problems than other models too, especially at medium and farther distances
ty. I saw your couch gen and thought I'd share mine lol
imms eat pizza
>>100124677but yeah, still some amount of rolling at selfie distances
>>100124868That picture frame is hella cute hahaha!
How's Pixart-Σ compared to alpha?
>>100125267i'm trying to get shadow effects.. but will have to wait for another day
>>100125290sigmaficantly better
THE OBSERVER, worm home black hole, iridescence, soap bubbles, palm leaves, marbleized quartz, crystallized glass, metallic, translucent, opalescent, neon, xray, illuminati, japanese, DRUG INDUCED INTOXICATED LATENT DREAM SEQUENCE ACID TRIP,
Greetings.Anybody got a list of exclusive Pony tags? I don't mean the big spreadsheet, I mean the things like the score tags or "source_anime" and the like.>>100125358I'm gonna make this my background, possibly tomorrow.
>>100125398Good evening!! :DIt's great to see you again!Every time I tried pony it would get too lewd so I just gave up on that for now hahaha
>>100125442Hello~I haven't really had problems with unwanted Pony lewds, maybe you needed some extra negatives?(I wouldn't mind getting lewds tho)Anyways, I'm gonna be making some stuff on Animagine for some time before diving back into Pony models. I'm just wondering if there are more of those "source" tags or anything else.
>>100125442for pony and merges with pony:positive: rating_safenegative: rating_explicit
>3D Model gen in the newsHow is it? Anyone try it out? What's the best for 3D models right now? I haven't been keeping up with /sdg/ for a few months, but wanna start getting into 3D Model gens
>>100125492I'm not even sure if I got the right model honestly hahaha!It's V6XL or something like thatI have a huge list of negs for lewds hahaMost models work with em>>100125524Ohhhhhh! Thanks, anon! :DI've give it another shot then haha
>>100125533instantmesh is probably the best right now. I've tried out a few but instantmesh was really impressive. I don't do a lot of 3d stuff though
>>100125552That should be the same Pony I have. Gotta wrangle with the tags sometimes, but it seems you got an answer this time.>>100125524This is exactly the kind of tags I'm looking for. I'd seen these two before in Civitai examples, but forgot about them.>>100125580Heyo. What kind of tags are you using for that kind of layout? Don't tell me it's just "comic page"...
>>100125533>Anyone try it outIt's shit
>>100125580Its a solid lead, thanks anon>>100125629Was a little skeptical when I saw it wanted an API key. Any recommended alternatives I should try?
>>100125358>>100125398Alright, I ended up trying this as a background already. Even before highres, it looks pretty nice.
>>100125661>it wanted an API keyhuh?
>>100125684>ComfyUI-Tripo: Generate 3D models using the Tripo API>using the Tripo APII guess it does say its optional on the repo tho though
fun fact: adetailer will detect feet as hands even while using a high detection threshold anyway, gn sdg
>>100125398afaik it's just rating_explicit, rating_questionable, rating_safe, and source_anime, source_furry, source_cartoon
>>100125624Seems to have worked too!! :DI just found a list of artists and picking em at random haha>>100125524Thanks again, anon! :]
Magic too dangerous for hands.>>100125752I see, thank you. I might have seen source_science or something like that on Civitai and thought there were more.>>100125861That's cool. What list did you find btw? The large spreadsheet with weirdly named tags?
>>100125750Gn anon
>>100125861hello PW! <3 I'm just about to hop off. I just missed you before I buried myself with work ;_;. anyways, have a good night!good night /sdg/!
>>100125874It's this one!>https://civitai.com/models/317578/pdv6xl-artist-tagsDownload the training data file for the list of names :]What's that one you're talking about?
>>100125891>What's that one you're talking about?https://lite.framacalc.org/4ttgzvd0rx-a6jf
>>100125890Hello, Animanon!!Goodnight, Animanon!!It's ok hahaha glad you stopped by before knocking out! I'll probably see you tomorrow, but for sure Tuesday if you're here!Sleep well! <3
>>100125874npfwiw, I switched to AutismMix from Pony and have been having much better results. https://civitai.com/models/288584
>>100125908Cool, thanks! I'll check this out! :]
Hands hazards continue.>>100125891I'll give it a try tomorrow, I'm logging off in a bit. That LoRA seems to be mostly for furry artists, but it's a lot of them, so something must be usable in normal gens.>>100125912Yeah, I've tried that model for a bit before. It's been better for standing 1person and porn more than anything else I've tried. Still need more testing tho.
>>100125958It was the first result I found hahaha! I didn't get the LoRA, but I did grab the list and changing the names seems to work pretty well haha
Dangerous magic is no joke.Time to sleep.>>100126001Gotcha. I'll give just the list a try then. But for now, good night!
>>100126100Goodnight!! Sleep well! :]Hope to see you again soon!
>>100126100>>100125890>>100125750good night.
>>100122230More experimenting with using sd for refs... God that shit frustrates me sometimes. It's amazing for autocompleting images, i get a bunch of good refs in the exact shapes i need and options for details, but it just feels like I'm wrestling with it sometimes. It can create these amazing images when I'm just experimenting freely but steering it is so hard.Next time before i even start I'm going into clip interrogator on hugging face and getting like an entire page of proompts to use
>>100126347It only steers a bit with ipadapter, controlnet or sth.
Do captions affect loss on training LORAs?I feel like everything about lora training is fucking cargo cults and voodoo magic.
Last one from me, good night anons
>>100126444good night.
>>100126417>ipadapterInteresting, might try it. Image only prompts could work. Would still need to build up a ton of refs anyway. I feel like if i just run them through clip interrogator i can pick and choose the best parts of each as well
>>100126444Goodnight, Quokkanon!! Sleep well :]
>>100122303Ella was completely neutered from the get go. It refuses to work with NSFW and celebrities. It's just a lousy research paper.I don't even think they released the code, but I'm not sure.That being said, it does work, quite well.
Hi, this seems the best place to ask:Is there something online that lets me use a picture of myself and add?>Take selfie>As an astronaut, space, sunlight>Does it while preserving some features like beard hair and faceI'm just saying, that Meta/Google/Microsoft have Petabytes of database, nevertheless less cost for energy, infrastructure and optimisations..Is there a software that integrates various different AIs like copilot does?>Make an image of party>Add a birthday cakeWithout going through different software.Regards
>>100126765There's a comfy UI workflow for all of that, but it's not online.
>>100126775That's the issue, I have an rx570, and not the best CPU, and not that much storage to throw away, it could probably be set up in less than 30minutes.Since you seem knowledgeable, I'm also ok in just using a specific software that does this:>A selfie>Add a hat and a long beardRegards
>>100126854Nope, might I suggest taking a selfie or buy a fake hat and beard from a dollar shop?
>>100126854ah, im in the rx580 4gb club myself. i use the directml fork of automatic1111. it should work with your gpu, but its gonna be slow as shit.
>>100126964Your gens are great, anon! :]I'd say they're worth the wait! You have a cool style!
>>100127016thanks, though each takes a few minutes (i switched back to euler a recently which has helped a lot), and i just cram loras together, though the style on these is mostly just porforever.
>>100127055I get that! I used to have to wait a bit tooThrowing LoRAs together is really fun to see how it comes out!>porforeverJust looked em up, and I really like that style!
>>100126964I know it's slow as shit, but also, database size?Different models require different databases, in your case, if you want to mix anime girl + space, it would be like 4gb anime and 3gb space, that's 7gb.How much size your models have?
>>100127331the main model i use is 2GB. It's FP16 and pruned.it's also a 1.5 model, and i oom a lot even genning at these resolutions if im multitasking.>>100127176yeah, they have a lovely style. there's also some fluffydango mixed in this, but its hard to see.
>>100127391They do! The mix works well! It's really nice :]
>>100127455is there asanagi mixed in this pic
>>100127455Nah haha, i'm trying out random artists from a list I got earlier!
>>100127479>>100127506Whoops, I accidentally replied to myself LOL
>>100122230Why do my results look like shit?
>>100127701It's likely your model, anon!What model are you using?
>>100127747Either actually look at the screenshot or don't bother responding at all.
>>100127756Oh right, forgot A1 has the whole thing at the bottom hahaha! Yeah, it's the model i'd imagine!
>>100127747I just set up this one: https://github.com/AbdBarho/stable-diffusion-webui-dockerI don't see any option to select a model. I see 2 options called checkpoint and sampling method.
>>100127701Definitely your model, go on civitai and download one of the porn 1.5 ones (easy to spot because the photo will be something lewd).
>>100127802Yeah, checkpoint is gonna be your model!I just saw your proompt but oh well LOLTry this one!>https://civitai.com/models/359999?modelVersionId=402346
>>100127805>>100127826I see that the files on civitai are all *.safetensorsDo I need to convert it to ckpt?
>>100127879No, should work fine. safetensors are preferred anyway
>>100127879Nope! They'll work just fine! :]Just put em in the models folder and reload your UI!
>>100127897Okay, great. I'm downloading it now.I've also seen some optical illusion pepes like pic related using the img2img tab. Which model is suitable for getting results like this?
>>100127992Great!What GPU do you have btw? Just curious hahaUhhh honestly there are many models you can use to get that kind of effect! It's just medium to high denoise with a relevant proompt haha
>>100128016I'm using an RTX4060 with 8 GB VRAM.I tried one of the default models and it looked like crap.>https://civitai.com/models/359999?modelVersionId=402346I tried this one and the results look weird. I tried some simple prompts like "dancing naked" and it was an image of a girl with one boob coming out of her armpits.
>>100128243Ohh gotcha! Yeah the default is pretty bad hahaSo that one is an XL model! I should have mentioned that earlier!>https://huggingface.co/stabilityai/sdxl-vae/tree/mainGrab the sdxl_vae.safetensors and use that as your vae :]All the weird stuff will go away!
>>100128267>https://huggingface.co/stabilityai/sdxl-vae/tree/mainI put this in the VAE folder, and applied the settings. Is there anything else I need to do? The output still looks weird.
New Thread>>100127849>>100127849>>100127849
>>100128410Hmm that should do it, as long as you use the vae with the modelThe artifacts stopped happening tho, so it's a step forward hahaYour resolution is pretty low, XL generally uses 1024x1024
>>100128444Increased it to 1024x1024 now. Does the CFG Scale and Sampling method matter much?
>>100128523I usually only have about 8~12 CFGThe sampling method is really up to you! Euler A is a good starting point :]
>>100128394>>100128362
>>100129056Oops forgot message, looks nice!
>>100128444Good info, thanks!
>>100129084thx anon
>>100122347Hello quokkAnon, it's be quite a while since I last sent you a paste (i think it was that time just before the new year!) so I've added it to the bottom of the rentry. That way you can always find the latest one by hitting sdg-link#bakery . The difference between what you've got here and in the new paste aren't that major, just one or two links. It's going to be relatively the same until SD3 finally arrives so you don't need to check the link too often.Anyway, thanks as always for playing! @( * v * )@