Previous /sdg/ thread : >>101368376>Beginner UI local installFooocus: https://github.com/lllyasviel/fooocusEasyDiffusion: https://easydiffusion.github.ioMetastable: https://metastable.studio>Local installAutomatic1111: https://github.com/automatic1111/stable-diffusion-webuiComfyUI: https://github.com/comfyanonymous/ComfyUIAMD GPU: https://rentry.org/sdg-link#amd-gpuIntel GPU: https://rentry.org/sdg-link#intel-gpu>Use a VAE if your images look washed outhttps://rentry.org/sdvae>Auto1111 forksSD.Next: https://github.com/vladmandic/automaticAnapnoe UX: https://github.com/anapnoe/stable-diffusion-webui-ux>Run cloud hosted instancehttps://rentry.org/sdg-link#run-cloud-hosted-instance>SD3 info & downloadhttps://rentry.org/sdg-link#sd3https://education.civitai.com/quickstart-guide-to-stable-diffusion-3https://aitracker.art/viewtopic.php?t=57>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://openmodeldb.info>Animationhttps://rentry.org/AnimAnonhttps://rentry.org/AnimAnon-AnimDiffhttps://rentry.org/AnimAnon-Deforum >Index of guides and other toolshttps://rentry.org/sdg-linkhttps://rentry.org/rentrysd>View and submit GPU performance datahttps://vladmandic.github.io/sd-extension-system-info/pages/benchmark.htmlhttps://docs.getgrist.com/3mjouqRSdkBY/sdperformance>Share image prompt info4chan removes prompt info from images, share them with the following guide/site...https://rentry.org/hdgcbhttps://catbox.moe>Discord6wUwtcJsr2>Related boards>>>/h/hdg>>>/e/edg>>>/d/ddg>>>/b/degen>>>/vt/vtai>>>/aco/sdg>>>/trash/sdg
>mfw Resource news07/11/2024>Introducing AuraFlow v0.1, an Open Exploration of Large Rectified Flow Modelshttps://blog.fal.ai/auraflow/>FlashAttention-3: Fast and Accurate Attention with Asynchrony and Low-precisionhttps://pytorch.org/blog/flashattention-3/>LLaVA-NeXT-Interleave: Tackling Multi-image, Video, and 3D in Large Multimodal Models https://github.com/LLaVA-VL/LLaVA-NeXT>SD Webui Resource Monitorhttps://github.com/Haoming02/sd-webui-resource-monitor>AI's Energy Demands Are Out of Controlhttps://www.wired.com/story/ai-energy-demands-water-impact-internet-hyper-consumption-era/>German court rules AI output can be protectablehttps://devclass.com/2024/07/10/german-court-rules-ai-output-can-be-protectable-ups-stakes-for-machine-generated-code/?td=rt-3a>EchoMimic: Lifelike Audio-Driven Portrait Animations through Editable Landmark Conditioninghttps://badtobest.github.io/echomimic.html>LaRa: Efficient Large-Baseline Radiance Fieldshttps://github.com/autonomousvision/LaRa>Grounding Image Matching in 3D with MASt3Rhttps://github.com/naver/mast3r07/10/2024>ROCm 6.1.2 HIP SDK 24.Q3https://rocm.docs.amd.com/projects/install-on-windows/en/latest/>ConceptExpress: Harnessing Diffusion Models for Single-image Unsupervised Concept Extraction https://github.com/haoosz/ConceptExpress>EasyAnimate-V3-XLhttps://huggingface.co/alibaba-pai/EasyAnimateV3-XL-2-InP-960x960>AMD to Acquire Silo AIhttps://www.globenewswire.com/news-release/2024/07/10/2911168/0/en/AMD-to-Acquire-Silo-AI-to-Expand-Enterprise-AI-Solutions-Globally.html07/09/2024>Tailor3D: Customized 3D Assets Editing and Generation with Dual-Side Imageshttps://tailor3d-2024.github.io>PaintsUndo: A Base Model of Drawing Behaviors in Digital Paintingshttps://lllyasviel.github.io/pages/paints_undo>Video-STaR: Self-Training Enables Video Instruction Tuning with Any Supervisionhttps://orrzohar.github.io/projects/video-star
>mfw Research news07/11/2024>4DiM: Controlling Space and Time with Diffusion Modelshttps://4d-diffusion.github.io>Multi-task Prompt Words Learning for Social Media Content Generationhttps://arxiv.org/abs/2407.07771>VEnhancer: Generative Space-Time Enhancement for Video Generationhttps://arxiv.org/abs/2407.07667>Tuning Vision-Language Models with Candidate Labels by Prompt Alignmenthttps://arxiv.org/abs/2407.07638>MARS: Mixture of Auto-Regressive Models for Fine-grained Text-to-image Synthesishttps://arxiv.org/abs/2407.07614>InstructLayout: Instruction-Driven 2D and 3D Layout Synthesis with Semantic Graph Priorhttps://arxiv.org/abs/2407.07580>Zero-Shot Class Unlearning in CLIP with Synthetic Sampleshttps://arxiv.org/abs/2407.07485>Video-to-Audio Generation with Hidden Alignment https://sites.google.com/view/vta-ldm>A Survey of Attacks on Large Vision-Language Models: Resources, Advances, and Future Trendshttps://arxiv.org/abs/2407.07403>HiLight: Technical Report on the Motern AI Video Language Modelhttps://arxiv.org/abs/2407.07325>Few-Shot Image Generation by Conditional Relaxing Diffusion Inversionhttps://arxiv.org/abs/2407.07249>Reference-based Controllable Scene Stylization with Gaussian Splattinghttps://arxiv.org/abs/2407.07220>ColorPeel: Color Prompt Learning with Diffusion Models via Color and Shape Disentanglementhttps://moatifbutt.github.io/colorpeel>CamFreeDiff: Camera-free Image to Panorama Generation with Diffusion Modelhttps://arxiv.org/abs/2407.07174>Diffusion Model-Based Video Editing: A Surveyhttps://arxiv.org/abs/2407.0711107/10/2024>V-VIPE: Variational View Invariant Pose Embeddinghttps://v-vipe.github.io>Latent Space Imaginghttps://arxiv.org/abs/2407.07052>HumanRefiner: Benchmarking Abnormal Human Generation and Refining with Coarse-to-fine Pose-Reversible Guidancehttps://arxiv.org/abs/2407.06937>Rethinking I2V Adaptation: An Object-centric Perspectivehttps://arxiv.org/abs/2407.06871
self actualization is masturbation
>>101375188For sdxl, specifically from models from civitai, are there any settings inside SD i should be aware of? Like, for example i just follow my old workflow of using stuff like hires fix is it just dumb with sdxl since it can do higher res shit to start?
>>101375247yea, read the descriptions, some come with special negatives and positives to use for quality (mostly the pony based pron models) other than that you can forget all the schizo quality prompts from sd15, they dont work and will make everything worse, especially the negative ones
When merging with SDXL to try find a secret top model recipe, can I ONLY merge models that are already SDXL tagged, is it retarded to do otherwise?
>>101375268ok so its mostly boomer prompting with sdxl?
>>101375281depends.. just look at the example prompts that come with the models and you get an idea, the pony models pretty much work with booru prompts tho>>101375271only SDXL models can be merged with SDXL models, but be careful with merging, can get ugly very fast, fine tuning is more effective on SDXL than merging.. also loras loras loras
i almost forgot i was supposed to try out some new artist spam. can't quite remember who tho. sad!
I see a whole lot of new samplers, both default and hires onesSuggestions?
ah yes, Hilma af Klint, et al.>>101375380i like u>>101375408xyz that shit, assuming you're using a sane UI (aka not comfy)
>>101375415fuck, got excited and posted w/o the image. i am stupid.
>>101375248>armor cameltoeHow is this possible?
>>101375427>he asks how cameltoe armor, and not why anon is a sick and twisted predditor.ngmi
Talk me out of getting a lovelace a6000, I already have a 4090 & 3090 chipping away at dreambooth models. I really feel I could contribute more to the climate emergency with another card
>>101375415i forgot how to do xyz layouts
>>101375427Its no more unhinged than nipples on a breastplate
>>101375460
>>101375427like Roman armor of elder years, primary and secondary sexual chrematistics hinted on the armor.. nah ofc its the model rofl>>101375442ya ofc, its so tiresome>>101375408>I see a whole lot of new samplers, both default and hires ones>Suggestions?dpm++ 2s a or euler a .. if you dont like random, dpm++ 2m or plain euler
>>101375468ty
Sorry one more thing (it wont be the last). Hires fix with SDXL, do you guys still use it? Shit just took forever for 1 image.
ask not what this weird anatomically incorrect whore can do for you, but what you can do for this anatomically incorrect whore.
>>101375474(≖_≖ )
look what,you've done
>>101375530>>101375547this post delta.......! clearly a botwhat a faggatron lmao
damn why didn't anyone tell me we solved catboys
>>101375547>>101375530i tried to copy ur style
plain jane. hold on, reporting myself for spam. it's only fair!eat a barbies ass or something.
Are hands fixed yet, is there a magic lora out there that works
>>101375597have a taste; learn from the master. https://files.catbox.moe/9f4xze.png
>>101375616no, and no.
>>101375597thats a $20 tshirt
>>101375616it got better ain't perfect tho, if you start lora mixing or asking for holding things it still fucked tho
back to back to back to https://www.youtube.com/watch?v=hr-bX9Wlo_I
>>101375619i need to tweak i only got b/w
>>101375678yeah well, you're missing a lot of stuff there. dynamic prompts are a hell of a drugshould i divulge the secret sauce? nah, i think not. i'll say this much{1$$best quality,lineart,(sketch:1.1),highly detailed|best quality,lineart,highly detailed|best quality,highly detailed} you rolled (sketch:1.1), i'd guess. it works pretty good for that tho
Last one from me, good night anons
>>101375487i hiresfix with a lightning sdxl so i just throw like 6 steps on there and it generally works pretty great
>>101375707night night nigga rest in peace. i think that's what they say. if not, they should
>>101375707Night
>>101375704i dont care that much desu, i like trying to recreate just by looking at the img anyway
>>101375797word. recreate away
holy shit /sdg/ dies so early now
>>101376181Your AI slop is getting old
>>101376181it's Thurs and hot as balls outside
>>101376194well yes this is an old prompt but i havent posted in months
>>101376181Nice grid, a shame it can't fit in schedule types.Mass reporting banned the few posters left
had to dust off and nuke this paki from orbit. this piece of shit has been getting his passwords compromised (i assume he logs on to public wifi in Karachi somewhere) and has gotten site after site compromised. fucking goddamned piece of paki scum
>>101376246the fact that pakis pretend they're not pajeets is fucking amazing to me. you're the same picture
hindu? islamic? whatever, you all shit in designated streets. none of you belong in the west
Assist for this?Trying old prompts with SDXL, im just curious whats making this error. Obv to do with hires fix, but not understanding what its meaning by the all_hr_negative_prompts, i dont have any schizo neg embeddings in the negatives so idk whats up, same with wildcards, dont have any active
>>101376311if you scroll down abit you will find that an option from your old prompt has been added in an additional field, remove that and it will work, cause the old option has been removed "dp_wildcard_manager_no_dedupe" some old extension you maybe used or some deprecated option
So what's minimum req for flow? Can it run on 6GB vram?
>>101376379nice gen>>101376460cool
>score 1, score 2, score 3ponyfags explain this
>>101376504magic
>>101376504bad dataset captions
>>101376310nice aeon flux
>>101376543Thanks
i am not having fun
I completely accidently managed to gen a girl looking exactly like a chick I once picked up from a bar as a young man. And now I can't stop fapping because that bitch was nasty as fuck in a good way.
>>101376588i keep getting old hags
i miss schizo anon
>>101376657why you do this?
WWE used controlnet and stable diffusion to make one of their trailershttps://www.youtube.com/watch?v=t8961pFrH-I&Based
>>101376759https://desuarchive.org/g/search/text/%22i%20miss%20schizo%20anon%22/page/1/>31 results found.>earliest post is from January You are the schizo
heh
>>101376787not true and kinda mean
progress
>>101376829>>101376845>>101376863>>101376885Loving these towers.
just woke up from a 2 week coma, is SD3 still dead?
>>101377136always will be, formidable alternatives are now out like kolours with more to come (pixart bigma, that new lumina model and whatever that auraflow guy will cook up next)
>me after reading this retarded thread
>>101376999thanks
>>101377457these are cool
>>101377517Nice, how does that work, is it a Lora giivng directly this result, or some kind of filter applied to a "normal" generated result?
>>101377243>checking /sdg/ during India evening hoursdo not
Why do I get these results sometimes, when using prompt weights like (prompt:0.8) or (prompt:1.1)?
>>101377615it's an extension for A1111 that uses a neural network to generate the pixelartI use this fork because it can do color paletteshttps://github.com/luhovei/stable-diffusion-webui-pixelization
mfw wakeup to see my robot cat gets used as op picim so proud
im not getting the anime style looks i want from these sdxl models im testing
>>101377944try anime screencap of xtry 3d , render , sfm in negatives or use aniimagine or something
>>101377944SDXL_Niji_v6?none of the example images seem to be particularly anime styled
I have to reinstall Automatic1111's webui on a new PC after a year or so, but what are the pros and cons of the automatic installation vs the Python + Git one? Back in 2023 I did the Python method, but is the automatic install just as good?
>>101375191>>101375367>>101375671>>101376368>>101376752very nice
>>101376591face wise she looks exactly like my ex lol
My faces are all fucked up, where did the "Fix faces" button go on the latest WebUI?
https://youtu.be/RoSfTe9cI-M?si=9Gb5OhmCD27bVfqx
>>101379004Try using adetailer instead. But if you really want the old face restore you can add it back by going to settings > settings in ui > add face_restoration in the txt2img and img2img (pic related).
>>101379004>My faces are all fucked up try to put (detailed face) in your prompt, might not fix it but mitigates this issue
>>101379311>>101379423Cool thanks, didn't know about these
Goo'd morning anons!
>>101379682are you og goo anon
>>101379693you my fan club? some ppl called me goo anon ya and I made some kinda wonkly SD15 models like anime10 and animeElves
What's your favorite cereal?
>>101379745>I made some kinda wonkly SD15 models like anime10 and animeElveswell holy shit, welcome back... again!
>>101379768thankgyoo! <3>>101379761Bicher Muesli? Thats pretty rad .. for the American sugar cereal type.. hm probably anything with cinnamon?(yea I know her hat looks like a condom)
>Breaking News: middle aged woman seen molesting underage slimes, know anything? Contact the FBB.
>>101379820reminded me of this kekhttps://youtu.be/pUscQZrdzAs?t=110
>>101379886oh yea that scene in DQ8 is legendary.. Japanese game devs can be so cool it kinda is a brothel .. but slimes... damn I miss Akira Toriyama
>>101379987Same anon, same. God bless japan.
everyday is a goo day for goo
>>101380010such a clean gen, impressive! very good..also have some goo everyday!
>>101380066sank yew!
for the pixart lovers900M PixArt Sigmahttps://civitai.com/models/573014?modelVersionId=638760
>>101380795did you ever muster up enough brain cells to get it working
>>101380795never have i been more sold
>>101375191working on lora generation, if the denim shorts keep blending into the black tank top making a wierd denim tank despite both of them being tagged properly (tank top, denim shorts) then what's going wrong, its also giving the shorts pockets despite being trained only on shorts which don't have pockets, do i need more epochs or greater training speed? its on 10 epochs at the moment but the training data never dramatically increases so its probably not overfitting
>>101380741have you considered doing a ponyxl fine-tune?>>101380819no. Im saving those brain-cells for killing with alcohol
what a retard kek not surprising desu
>>101381004who?
>>101381041forgot your image
>>101380982>have you considered doing a ponyxl fine-tune?the thought crossed my mind, but honestly there are already so great ones around and the loras for XL are excellent .. plus the amount of hardware you need to make a good one is staggering .. so I kinda am just floatin with the vibe for now, maybe if the 5090 has like 40gb vram and I get one I think about it again
>>101381065>maybe if the 5090 has like 40gb vram and I get one I think about it againnot ready https://www.techradar.com/computing/gpu/nvidias-rtx-5090-now-rumored-to-have-superfast-clock-speeds-as-well-as-being-super-slim-could-this-gpu-be-too-good-to-be-truehttps://videocardz.com/newz/rumor-geforce-rtx-5090-base-clock-nears-2-9-ghz>Panzerlied also recently informed us about the RTX 5090 supposedly using a 448-bit memory bus (as opposed to the previously rumored 512-bit bus), which would mean a VRAM loadout of 28GB (and not 32GB)
>>101381055do I need one
>>101381108>pretending to be this new
>pelvic curtain>erect nipples>sagging breastsAre there any patrician tags I'm missing?
>>101381124ive been here longer than u
>>101381150sure "anon"
>>101381098>28GBsad .. so the only enthusiast option is still the overpriced workstation line, and I ain't selling a kidney nor any of my synthesizers for thatthat said I still kinda want a 5090 when it comes out for the energy savings alone, my 4090 probably cost me more in energy cost than purchase price
>>101381181>sad .. so the only enthusiast option is still the overpriced workstation linethat's what Nvdia wants you to do yeah, 90% of their revenue are from AI data centers, so of course they want to prioritize overprised workstation gpu's instead of democratizing it
>>101381201>90% of their revenue are from AI data centerssource?
>>101381148nice one
>>101381222https://www.techpowerup.com/309125/nvidia-announces-financial-results-for-first-quarter-fiscal-2024-gaming-down-38-yoy-stock-still-jumps-25ok not 90%, but data center makes them more and more money yeah
>>101381222not that anon, but its true
>>101381256>>101381257Wow! TIL
>>101381257I've been saying that forever, that nvidia doesn't care about retail consumers and could happily jettison them and still make ungodly money
>>101381306>"I've been saying that forever">has never said it before
>>101381306for once I agree with you on that one debo
>>101381306true they could right now, but the competition is slowly creeping onto them in the data center market, its always good to diversify in you offerings ... alteast if they are smart they dont drop consumer ever, cause it also drives innovation, they would not have this data center product if they hadnt pushed GPUs to what they are now due to gamer demand
I don't understand how it could draw even toes, fingers and labia right but get confused by eyes of all things.
today is my first day on 4 chan
Just need massive amounts of VRAM. If NV wants to be cunts someone will squirt something else out to replace cuda for accel with torch. (Apparently rocm is awful but I can't experiment with it.)That would imply competency on AMD's behalf, for example, or Intelaviv realizing there's room for them to drive a big fat hammer between NV and the whole AI/Nerd/Gamer/Retail market.
>>101381306>I've been saying that foreverLink anon to it you massive faggot
>>101381306>thread schizo opinion
>>101381519She looks so sad and pathetic, very cute, good job anon
>>101381519close your eyes princess, you're scaring me. teeheek!
>>101381579He's right, though.
>>101381601>a broken clock is right twice a day
>>101381552I only have 12gb of vram, bought my card thinking I was future proofing. At least I can run SDXL with Forge
>>101381652Yup. I have two 8gb 1080 but it doesn't work like that apparently. (I.e no pooling.)They have lasted ten years though so I shouldn't really complain.
>>101381670what model is this? sloppy yet sovl
Are there any all in one models yet? I mean, one that combines the ability to understand text and respond, or generate media based upon it as required?
>>101381652yea I never estimated the amount of VRAM needed, about a year before the AI hype in 2022 I bought a 3070 8GB .. barely scraped through SD15 .. but SDXL just killed that card, I am super happy with the 4090 tho, just wish it wouldnt draw 350-400 watt when genning>>1013816581080 series ditched NVlink for consumer cards right? sad .. even if it had it tho, running two 1080s in one case sounds like a fire hazard (not like the 4090 isn't one by itself)
>>101381658That's a good run. Time to upgrade. Shame gpu cards don't have removable ram?
>>101381731These still have it, I can SLI them. (Which is why I had them; Fallout 4 at 5760x1080 on 3 monitors for a long time.) Unfortunately very little I can use/play actually benefits from it. I used to be able to use the huge XL models with a111, but can't even run it now. (cuda memblock splitting nonsense where it basically looses 1gb of vram to the ether and only a111 with the right pytorch conf 'fixed' it.)>>101381735That would be beautiful. Performance wise these are fine for me. (Starfield, Skyrim, etc) But I need more vram for SD/LLM shenanigans. Or, you know, pooling. :^)
>>101381735the whole GPU as workhorse computing center piece in an extension slot kinda is silly design of the ATX mainboards, upside down, locked airflow up, not really combinable with different GPU/VRAM like CPU/system RAM .. but when ATX was specified ages ago the systems looked so different, you had normal sized GPUs etc. the CPU was the thing that needed the coolingthe SXM standard they use in servers is so much cooler, but we probably wont see that in consumer products got my 4090 vertical tho .. helps alot(pic related is 4 SXM GPU modules connected by NVlink)
>>101381731>350-400 watt when genningYeah, that's a lot.>>101381849>SXM standardWho knows. Maybe demand will soon be there.
>>101381844>with the right pytorch conf 'fixed'sad .. I hate pytorch with a passion, that software makes so much trouble, also yea SLI kinda is dead for gaming, just more sensible to put more cores on a die/add more chiplets and sell bigger cards
>>101381885It's a necessary evil apparently. lol
>>101375474catbox?
Another hands-free umbrella
>>101382041ya sure, you are lucky I am awake again .. that's been what 11 hours ago? https://files.catbox.moe/gaweli.png
>>101382111Thanks. Weren't you posting slimes recently too? I liked those.
>>101382188ya thats me <3
Goo?
>>101382233Pony usually does slimes in a kinda boring way so it was neat seeing how those loras affected them.
The gangs all here hahahah :]
>>101382267always>>101382272ya, masha lora is probably the best I seen for cute slimes I have seen yet
Morning anons
>>101382327gm
>>101382327Good morning.
What should we discuss today, fellow avatars?
>>101382358>I wanted more pony merges a couple of weeks ago>find some>my ssd is now full and I have like 10 tabs with more modelswill this ever be solved?
>>101382358choose any news item from the top of the next thread :)
>>101382373pic related only solution .. I got 10TB atm .. but I feel ya, maybe I should archive my 2022 gens?
>>101382433>>101382433>>101382433
>>101382444lets fill this up with some nuns then
>>101382444>>101382508
>>101382508
I'm helping
>>101382526cute novice, I got one to
