Previous /sdg/ thread : >>100368968>Beginner UI local installFooocus: https://github.com/lllyasviel/fooocusEasyDiffusion: https://easydiffusion.github.io>Local installAutomatic1111: https://github.com/automatic1111/stable-diffusion-webuiComfyUI (Node-based): https://rentry.org/comfyuiAMD GPU: https://rentry.org/sdg-link#amd-gpuIntel GPU: https://rentry.org/sdg-link#intel-gpu>Use a VAE if your images look washed outhttps://rentry.org/sdvae>Auto1111 forksForge: https://github.com/lllyasviel/stable-diffusion-webui-forgeAnapnoe UX: https://github.com/anapnoe/stable-diffusion-webui-uxVladmandic: https://github.com/vladmandic/automatic>Run cloud hosted instancehttps://rentry.org/sdg-link#run-cloud-hosted-instance>Try online without registrationtxt2img: https://www.mage.spaceimg2img: https://huggingface.co/spaces/huggingface/diffuse-the-restInpainting: https://huggingface.co/spaces/fffiloni/stable-diffusion-inpaintingpixart: https://huggingface.co/spaces/PixArt-alpha/PixArt-Sigma>Models, LoRAs & embeddingshttps://civitai.comhttps://huggingface.cohttps://rentry.org/embeddings>Animationhttps://rentry.org/AnimAnonhttps://rentry.org/AnimAnon-AnimDiffhttps://rentry.org/AnimAnon-Deforum >SDXL info & downloadhttps://rentry.org/sdg-link#sdxl>Index of guides and other toolshttps://codeberg.org/tekakutli/neuralnomiconhttps://rentry.org/sdg-linkhttps://rentry.org/rentrysd>View and submit GPU performance datahttps://docs.getgrist.com/3mjouqRSdkBY/sdperformancehttps://vladmandic.github.io/sd-extension-system-info/pages/benchmark.html>Share image prompt info4chan removes prompt info from images, share them with the following guide/site...https://rentry.org/hdgcbhttps://catbox.moe>Related boards>>>/h/hdg>>>/e/edg>>>/d/ddg>>>/b/degen>>>/vt/vtai>>>/aco/sdg>>>/trash/sdgOfficial: discord.gg/stablediffusion
>mfw Resource news05/07/2024>CCDM: Continuous Conditional Diffusion Models for Image Generationhttps://github.com/UBCDingXin/CCDM>MediaPipe Hand Crop Fixhttps://github.com/sign-language-processing/mediapipe-hand-crop-fix>LGTM: Local-to-Global Text-Driven Human Motion Diffusion Modelhttps://github.com/L-Sun/LGTM>AniTalker: Animate Vivid and Diverse Talking Faces through Identity-Decoupled Facial Motion Encodinghttps://github.com/X-LANCE/AniTalker>DVMSR: Distillated Vision Mamba for Efficient Super-Resolution https://github.com/nathan66666/DVMSR>ImageInWords: Unlocking Hyper-Detailed Image Descriptionshttps://google.github.io/imageinwords/>MotionLCM: Real-time Controllable Motion Generation via Latent Consistency Modelhttps://dai-wenxun.github.io/MotionLCM-page/>comfy-cli: Command Line Interface for Managing ComfyUI https://github.com/yoland68/comfy-cli>Performance Profiling Report (Forge/A1111/ComfyUI)https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/716>ComfyUI-Video-Editing-X-Attentionhttps://github.com/chaojie/ComfyUI-Video-Editing-X-Attention>AM-RADIO: Reduce All Domains Into Onehttps://github.com/NVlabs/RADIO05/06/2024>Detector-Free Structure from Motionhttps://zju3dv.github.io/DetectorFreeSfM/05/05/2024>ComfyUI Prompt Quillhttps://github.com/osi1880vr/prompt_quill_comfyui>Efficient Implementation of Kolmogorov-Arnold Network [KAN]https://github.com/Blealtan/efficient-kan>controlnetXL_line2colorhttps://huggingface.co/kataragi/controlnetXL_line2color05/04/2024>PuLID now supported in sd-webui-controlnet!https://github.com/Mikubill/sd-webui-controlnet/discussions/2841>ThemeStation: Generating Theme-Aware 3D Assets from Few Exemplarshttps://github.com/3DTopia/ThemeStation05/03/2024>Virtuoso Nodes: Set of nodes to give Photoshop-like functionality within ComfyUI.https://github.com/chrisfreilich/virtuoso-nodes
>mfw Research news05/07/2024>Generated Contents Enrichmenthttps://arxiv.org/abs/2405.03650>Is Sora a World Simulator? A Comprehensive Survey on General World Models and Beyondhttps://arxiv.org/abs/2405.03520>Gaussian Splatting: 3D Reconstruction and Novel View Synthesis, a Reviewhttps://arxiv.org/abs/2405.03417>Animate Your Thoughts: Decoupled Reconstruction of Dynamic Natural Vision from Slow Brain Activityhttps://arxiv.org/abs/2405.03280>Mind the Gap Between Synthetic and Real: Utilizing Transfer Learning to Probe the Boundaries of Stable Diffusion Generated Datahttps://arxiv.org/abs/2405.03243>Adapting Dual-encoder Vision-language Models for Paraphrased Retrievalhttps://arxiv.org/abs/2405.03190>Video Diffusion Models: A Surveyhttps://arxiv.org/abs/2405.03150>SketchGPT: Autoregressive Modeling for Sketch Generation and Recognitionhttps://arxiv.org/abs/2405.03099>Matten: Video Generation with Mamba-Attentionhttps://arxiv.org/abs/2405.03025>Paintings and Drawings Aesthetics Assessment with Rich Attributes for Various Artistic Categorieshttps://arxiv.org/abs/2405.02982>VectorPainter: A Novel Approach to Stylized Vector Graphics Synthesis with Vectorized Strokeshttps://arxiv.org/abs/2405.02962>iSEARLE: Improving Textual Inversion for Zero-Shot Composed Image Retrievalhttps://arxiv.org/abs/2405.02951>MVIP-NeRF: Multi-view 3D Inpainting on NeRF Scenes via Diffusion Priorhttps://arxiv.org/abs/2405.02859>Stable Diffusion Dataset Generation for Downstream Classification Taskshttps://arxiv.org/abs/2405.02698>Enhancing Social Media Post Popularity Prediction with Visual Contenthttps://arxiv.org/abs/2405.02367>Efficient Text-driven Motion Generation via Latent Consistency Traininghttps://arxiv.org/abs/2405.02791>Adapting to Distribution Shift by Visual Domain Prompt Generationhttps://arxiv.org/abs/2405.02797>U-DiTs: Downsample Tokens in U-Shaped Diffusion Transformershttps://arxiv.org/abs/2405.02730
>Cold? Why would I be cold?
just crusin
seeking the truth
>>100371151whatever will the eurofags think, when they wake up to dumb big titty whores? they'll probably ban the lot of us, on grounds of heterosexuality. lel (if its happened to me, it can happen to you!)
>bouncehttps://www.youtube.com/watch?v=gdHao9henOs
gaze upon and seek >>100371229goes hard
>>100371191The line seems to be full coverage of the areola, so we'll probably be fine
>>100371246hey anon, it's me. demon jesus.what's up?
>AIYY YAI YAI YAIIIIIIIIII>>100371279My cholesterol
>>100371279what's the difference, you ask?well let me tell you!regular jesus loves you and stuff,demon jesus is all mad about ... uh stuff!
Out of the loop for a year, whats the QRD
>>100371296A dwarf yakiniku restaurant sounds comfy
>>100371360shit's all fucked, yo!
>>1003713771girl posters are still a waste of space i see
>>100371381it is what it is.we long to see your superior proompting skills! put us to shame, anon-sama!
>>100371395okay this confirmed my suspicion - complete and utter stagnation. I'll come back in another year
>>100371409дo cкopoй вcтpeчи!
Last one from me, good night anons
>>100371409dear diary,some fag barged in and declaimed his retardness. it was pretty weird.anyway, the ""CRITICAL SHIPMENT""" is coming soon so, look out.Cheers,Wyatt MannPSPS: TAKE A GOOD LOOK AT -53.2332W,23.NISHJW;WOWohmyOOOWOWthat's doog.good. that's good.
>>100371451good night.
you did not create youyou were made
>>100371451sleep tight
.i would be honored to have my work enshrined into latent space. "don't use my drawings in your dataset" boo hoo.
what model do people use for CLIP interrogator?
>>100371579I don't think many people use it, it's 100% inaccurate.
>dude's been genning giant woman with massive tits and wide hips for hours nowHaving a nice goon sesh there?
>>100371773Actually, I've been watching old nostalgia critic videos while waiting on gens
Anyone know if there is a node that lets me see what the T5 encoder is doing with my prompt?
Good night everybody
burn SAI burn
>>100372050*Released unusable 3B parameter llm*
i dont understand how to use this. so i downloaded the model and installed everything but it just gives me cave painting like things. what settings do i use for things that are high res and stuff?
>>100371976night
>>100372081You downloaded what model on what UI following what guide?Your post tells me nothing except you're some kind of cave man.
>>100372186i was using that retard guidehttps://rentry.org/voldy and downloaded the 1.5 stable diffusion model
>>100372232Go to civitai and download a couple of newer models. Or use sentences to describe what you want with the one you have.For example:A painting by matisse of a woman. She is saying "paint me like one of your french girls".
>>100372232This is why I never read the OPs in generals.https://github.com/lllyasviel/stable-diffusion-webui-forgeJust download the one click package from here and you're off. Talk about getting new and better models once you confirm it works.
Anyone have any cool styles they want to recommend?
>>100372288my latest ones. Christian Wilhelm Allers painting -- try drawingChristoph Niemann minimal illustration -- might not workChristophe Vacher concept art -- seems to favour fantasy scenesClarence Gagnon paintingClarence Holbrook Carter paintingClaude Monet paintingClayton Crain digital comicClive Barker illustrationClyde Caldwell illustration -- fantasyClyfford Still painting -- limited palette?
>>100372255>>100372259oh i think i got it. i just put the sampling steps all the way up to 150
>>100372302Bingo
>>100372296Hit me like a fucking truck that I asked a question and then just got a fucking answer.Thank you. That never happens anymore.
damn extra leg
>>100372259Sampling steps or cfg?
>>100372069what i dont understand is, if they wanted to make more different stuff, then why not make things that would help themselves make their main thing better, this way they just giga split their resources and are below mediocre at everythinglike making their own vlm (which ironically im pretty sure is the ONLY thing they have not done) and releasing it to help people align with base checkpoint captioning when finetuning would probably go a long way
>>100372389I sound like a broken record at this point, but there is no way a huge amount of SAIs financials weren't improperly used or straight up embezzled.That had basically an unlimited money spigot to train endless top of the line models and they somehow screwed it up.
>>100372450cool
>>100372450milton bradley dark souls
A1111 > comfyuiSD 1.5 > SDXLif you believe otherwise you're a midwit
>>100371451good night>>100368842>this is the quokka we've come to knowdem macropods
>>100371451night
sigma > balls if you believe otherwise you're a midwit
>>100372507I decoded your autism. You got better results and were more pleased with the supposedly objectively better options.Rather be a midwit than a dipshit.
>>100372595lol
>>100371875hmm wouldn't this have to be added as output to the t5 text encode node or something like that?I actually don't remember what format it encoded to tho.
One thing I find pixart is really good at following is where things go on the screen, if I say a demon is on the right, it puts the demon on the right.
>>100372631Yeah I actually don't know what it's spitting out after the text goes in, but I'd kinda like to see it.
>>100372634Cool. Is there a high success rate?
>>100372673Actually details concerning the people themselves can be a bit of a crapshoot sometimes, but it listens very clearly to what side of the images something is supposed to go on.
>>100372689That's what I meant. Thanks
Okay, I'll stop posting faux fantasy art. I just really liked this last one.
why is cumfart so silent since weeks?
>>100372717
>>100372649might have to go and add it to the node yourself
>>100372751Huh.That gives me an idea for my comics. what node are you using to add text?
>>100372717maybe d*b* could ask him in their discord?
>>100372738>correct fennec ears SAI is so done kek
Good morning anons
Hello fellow sigma male
>>100372507Straight diffusers scripts > any gay uiSDXL > 1.5You suffer from dunning kruger.
So sigma we don't even >> each other?
>>100371114>the gothic vintage hair dryer>>100371144Very hot>>100371191As a eurofag, I like big tits.
>>100373184gib
>>100373216Hungry anon?
>>100373216I have extra cupcakes after you finish breakfast!
>>100373235very, your gens make me more hungry, stupid office daysgreat gens, they look really good anon
>>100373272Thank you anon. I made one without sprinkles. Just how you like it
what's the best furry model outside of pony diffusion?
>>100373300Sameface problems?
>>100373292can you give me a quick rundown on how to get sigma setup? would love to try it this eveningiirc you only need the extra models node for comfyui or?
Prompt: wawaweewaawoopeepeepeooopooodawd ad aw dwad9
>>100373353needs 20gb ram for the text encoder, do you have that.
>>100373373yeah got 64gb ram
Prompt: wawaweewaawoopeepeepeooopooodawd ad aw dwad9 faF AFAF awfawf3 3t3 4ef hdafsrgv eat3 R
>>100373382looks like a well trained model
name my company
>>100373353https://github.com/city96/ComfyUI_ExtraModelshttps://huggingface.co/PixArt-alpha/pixart_sigma_sdxlvae_T5_diffusers (cp text_encoder files into to models/t5, vae to models/vae)Get your favorite finetune (bunline or vintage knockers), or the default model at https://huggingface.co/PixArt-alpha/PixArt-Sigma/resolve/main/PixArt-Sigma-XL-2-1024-MS.pth?download=trueAnd here's a catbox https://files.catbox.moe/uv0h8c.png
anyone has a good anim diff workflow ?
>>100373400Probably just the text encoder trying to make sense of it.
>>100373401microsoft
>>100373414thank you very much anonsounds like my smooth brain should be able to handle that
>>100373418only guy doing animations here doesnt explain shit
>>100372312Easily fixed. It is the castle on the background that seems covered in cobwebs that I don't like much. Great gen though.
>>100373546In the meantime, you can make some okayish gens online for free https://huggingface.co/spaces/PixArt-alpha/PixArt-Sigma
PLANT
time for another day of high art (especially coomer elves)
controlnet lineart inpainting is fun, just mask the face or body and poof, edits that follow the original
>>100373720
>>100373741
>>100372762>That gives me an idea for my comics. Nice. But it's actually not a text node but a whole cool new thing that keeps a character very consistent by comparison to what SD[XL] otherwise does: https://github.com/HVision-NKU/StoryDiffusionThat said I also remember an neat example of just doing stuff with text and bg images:https://civitai.com/models/377483/artismysteriums-basic-playing-card-workflow
made you look
If the T5 encoder is doing a lot of the lifting for prompt adherence, I understand the model tags play a part as well, then isn't worth using everywhere as a default?The two nodes i use in comfy are the default and advanced CLIP encode nodes, I presume they are not "T5" encoders??
>>100373892Think of the T5 encoder as a pre-sorting mechanism for meaning. The output is encoded as roughly 300*2 bytes (16bit) of input to the transformer (if you max out prompt). It's just a cleaner source of info. You don't get access to T5's weights.This is literally as bad as it's ever going to be. T5 isn't the only encoder model and it isn't even close to SOTA. Wait until we have smaller and better encoder models on transformers.
>>100373822sloppy sloppy slop.>Le AI face>Chair not even facing in the correct direction>1 girl
>>100373892Everyone is trying stuff. SD3 too also used T5:https://stability.ai/news/stable-diffusion-3-research-paper
>>100373929>>100373865>>100373715Your posts are deliciously artistic
>>100372407you underestimate bad management, Emad wanted to make ai for african kids or somethinghttps://www.forbes.com/sites/kenrickcai/2024/03/29/how-stability-ais-founder-tanked-his-billion-dollar-startup
>>100373971I think I misspoke in my post. I wanted to say that his financials were misused. (if not straight up embezzled).All he had to do was make SOTA AI image models. He had the funding, the knowhow and still somehow screwed it up.
>>100373933>>100373956Thanks anons, I was worrying about my 32GB ram being enough for the near future.
>>100373990Still, don't attribute to malice what can be explained by stupidity
>>100373971The jews will never get me to sub to forbes, kys
>>100373969Thank you :3Your brownies are making me hungry. Promptmaxxing in sigma is so much fun
>>100373971but according to d*b* emad is a good IT worker by default because emad is brown?
>>100374013Here, dear dumb ni****, a bypass for your pea sized brainhttps://12ft.io/
>spend 80 seconds rendering this beautiful 4k prompt>wizard is facing the wrong wayd'oh
>>100374024no thanks not clicking that dolphin porn
>>100373994So far there is little evidence that the FP16 version of T5 won't work just fine on Pixart Sigma and even that may be bloat. On the other hand of course having some text model with the level of understanding that llama3 has connected to an image AI would be sort-of nice, but I'm not sure this will happen that soon and even if for imagegen running THAT model on CPU may be enough? We'll find out.They'll still find you a reason to get moar VRAM than 24GB and more system RAM than 128GB sooner rather than later
>>100374045FP16 straight up works. Why are you even mentioning there being little evidence of it not working? Just try it instead of going full no-gen.
>>100374072>there is little evidence that the FP16 version of T5 won't work just fine on Pixart Sigma and even that may be bloat
>>100374045>llama3 ... connected to an image AIThis is where i was sort of thinking things were going, perhaps faster than i expected, not knowing much about what's in the pipeline however.
forbes shills and dolphin porn general
Why can't I use more vram to make 4k pixart gen faster?
>>100374084>>1003740454bit and 8bit T5 work w/ bitsandbytes too. You're literally just noise. Try things out instead of pontificating about things you don't know.
>>100374109for the same reason you cant use more RAM to make the CPU faster
>>100374112I literally said there is little evidence that it won't work fine and that it probably still is bloat (=>less precise variants probably also still work)stop making up shit
>>100374021i think indians self identify as white therefore emad is shit according to debo
>>100374132My point: there's only evidence that it works on F16 (and below), so your words are wasted on everyone. The trainer uses FP16 by default if you look at the repo. Llama3 isn't an encoder model. And there's no "they" trying to get you to use more than 24GB of RAM. Read up bobble head.picrel of your no-gen additions
>>100374155There was a time where coomer art interested me. But now fully comprehend that stable diffusion can do that because that's all it has been fed on. For the better part of a year every model has been dripfed porn, merged and fed more porn then merged again. Of course they're good at it.And it no longer impresses me.
>>100374198Post-coomer age of enlightenment?
cammy, except as a fire emblem character (was tharja)
>>100374198thanks for the essay
>>100374198>>100374228niiiice
>>100374021Maybe he is good at IT, but he is a demented CEO/manager
>>100374194> refuting what I didn't say in the first place, then saying the same thing as I didvery helpful sarAs for the other topic: More than 24GB VRAM / more than 32GB system RAM total is a fact for all sorts of published AI models, they'll use and publish more of them. Perhaps Pixart or Stability.ai won't, who knows, but even then yet more useful additional nodes will - sooner or later.
>>100374234>same ai face
>>100374227>Post-coomer age of enlightenment?My enlightenment is that I like gyarus
>>100374099how
bonus points if you can guess the game:
>>100374292Crazy
>>100374310kingdom cammy: dead redemption 2?
>>100374263ohhh i see@debo can you clarify which skin colors CEOs need to have to not be absolute garbage?
>>100374343>trying to talk to debo when he's not herepeak obsessed schizo behavior
>>100374331yes siralso "prompt more important" in controlnet seems to work well
Not quite what I was going for, but cool nonetheless
>>100374343>brown, black and shti
nobody fucks with Miyamoto Musashi
Almost actually got it that time.
>1girl
>>100374497
>>100374483i like it
>>100373971>Emad infuriated our initial investors so much it’s just making it impossible for us to raise more money under acceptable terms - A current Stability executivelol , lamo even
can't wait for T5 FaceDetailer, it still uses CLIP :(
Man I fucking live pixart
>>100374663do you have a reason to believe this is actively being worked on, or is this just hopium
>>100374364he reads all messages in every threads sooner or later anyway, he can answer then
>>100374707pure hopium that some of the most popular SD tools will work on Pixart in the coming months
>>100374727You can use the unsampler node, plug that latent into a ksampler at cfg-1 dmpp_2m karras then plug those samplers into a model of your choice and use your SD tools on it from there.
>>100374723the fact you spend time thinking about him when he's not even here is pretty sad
Good bye for now anons
>pixart shill gonefinally some breathing room
>>100374727what are the ram/vram requirements for sigma-1024? 2k? I'm going oom w/ 32gb RAM & 16gb VRAM
>>100374804they call me schizo for a reason
>>100374837It's like 4, 5gb at most for even the 2k model. 20gb of ram tho.
>>100374815later
>>100374865much of which was for T5 where you can apparently actually use a less precise version, right?
>>100374764Thanks I'll keep it in mind when I'm feeling more technically inclined :3>>100374837Less than that. Load the T5 encoder on CPU, that's probably what's maxing out for you and it's fast enough like that. Otherwise close your porn tabs.
>>100371229Good taste
>>100374902I found this schizo workflow that takes a pixart image and uses unsamplers etc to refine it using SDXL.From the few times I've used it works really well.https://files.catbox.moe/gfpxpc.json
Save pupper>>100375012TYVM I'll try it out later
It worked!
>>100375156The schizo workflow?
>>100375167joan saving pupper, I'll have to schizo it up when my schizo energy is higher
>>100375156Bruh, Your shit's all retarded.
>>100374858based schizo
>>100374902okay, it's working now... for whatever reason after it initially failed, python was stuck in the background with 20gb of committed memory and wouldn't release it, so I had to restart my pc
>>100375191oh look, a child skipping school and getting mad at watercolor style
>>100375233nisu! that's a pretty watercolor
to the baker of the next thread: remember to NOT, i repeat, NOT include the pastebin
>>100375251sank yew, pretty impressed with the initial results.. wondering if you can concat t5 prompts now..probably not, but will try
>>100375263By mentioning the pastbin you are drawing attention to the pastebin to those who otherwise knew nothing about it. What is the pastebin?
>>100375299pure autism and it doesnt have an effect to post it anyway
>>100375299Newfag or just playing dumb
>>100375322>>100375317I've been here almost as long as these generals have been a thing, I just never read the OP because only autistic hall monitor faggy boys do that.
>>100375317>it doesnt have an effect to post it anywaywrong
>>100375337>I've been here almost as long as these generals have been a thingThen it's impossible for you to have missed the pastebin
seems like concat conditioning does work. Wish I had more time to mess around with these workflows this morning
>>100375356that gorilla got a nice pair of tits10/10
>>100375235That's a cope, ngl, looked through your past ones, and some of them aren't too bad. But if you aren't willing to admit you could do better with some of them, you're ngmi.
>>100375235Also, this is a good water color example you namefag.>>100375269>>100375356
>>100375356What are you using concat conditioning for here?
Why does civitai have an SD3 tag in filters? Did it come out or something?
>>100375479some people requested it so they can tag their blurred out api 1girls (god bless SAI for making it safe)
>>100375479>Did it come out or something?Yeah but it's called sigma now. Check it out under that tag.
>>100375479Been there for weeks anon.
>>100375479>Did it come out or something?no, they're just getting ready for when it does. At the moment only select few people have local access to it. No they haven't leaked it.
>>100375479I think the civitai dev simply sometimes adds stuff ahead of time.
>>100375376Sorry you're not entitled to more than the roughly 10 seconds I spend writing and generating each batch of pics I pick from. I spend actually zero seconds inpainting or refining these. Pixart is just that good :3
>>100375470that gen wasn't actually using any concats. I did some others with different hair colors to confirm if it even worked or would error out, but didn't like the images enough to post
>>100375527>I spend actually zero seconds inpainting or refining these.That's why your ngmi.
>>100375527Low effort, high quality
Next>>100375528>>100375528>>100375528Thread
>>100372595>>100372600utter nonsense: checksamefag: checksad!
>>100373076>use thing>hate it>"NUH UH YOU LIKE IT"92 iq at best