Previous /sdg/ thread : >>101223687>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>Related boards>>>/h/hdg>>>/e/edg>>>/d/ddg>>>/b/degen>>>/vt/vtai>>>/aco/sdg>>>/trash/sdg
Cursed thread.
>mfw Resource news07/01/24>Segment Anything without Supervisionhttps://github.com/frank-xwang/UnSAM>MimicMotion: High-Quality Human Motion Video Generation with Confidence-aware Pose Guidancehttps://tencent.github.io/MimicMotion/>PopAlign: Population-Level Alignment for Fair Text-to-Image Generationhttps://github.com/jacklishufan/PopAlignSDXL>B-LoRA for Kohya-SShttps://github.com/ThereforeGames/blora_for_kohya>Nearly half of US firms using AI say goal is to cut staffing costshttps://www.baka.com.au/world/north-america/nearly-half-of-us-firms-using-ai-say-goal-is-to-cut-staffing-costs-20240629-p5jpsl.html06/30/2024>HunyuanDiT-v1.2https://huggingface.co/Tencent-Hunyuan/HunyuanDiT-v1.2>Hunyuan-Captioner https://huggingface.co/Tencent-Hunyuan/HunyuanCaptioner>MotionClone: Training-Free Motion Cloning for Controllable Video Generationhttps://github.com/Bujiazi/MotionClone>SDFX, no-code platform to build and share AI apps, goes open sourcehttps://github.com/sdfxai/sdfx>Ari Emanuel calls Sam Altman a "con man" who can't be trustedhttps://www.businessinsider.com/ari-emanuel-endeavor-ceo-calls-openai-sam-altman-con-man-2024-606/29/2024>MotionBooth: Motion-Aware Customized Text-to-Video Generationhttps://github.com/jianzongwu/MotionBooth>As AI gains a workplace foothold, states are trying to make sure workers don’t get left behindhttps://apnews.com/article/artificial-intelligence-worker-education-3707cc1d8b3a3aeb1bac5bc357a21c73>LyCORIS 3.0.0 - New Functional API, Parametrize API and Module APIhttps://github.com/KohakuBlueleaf/LyCORIS#20240629-update-to-300---brand-new-functional-api-parametrize-api-and-module-api>FastSD CPU v1.0.0-beta.33 release with Aura SR 4x upscalerhttps://github.com/rupeshs/fastsdcpu/releases/tag/v1.0.0-beta.33>SPRIGHT (SPatially RIGHT) Correct Caption Dataset with 2.3 Million Imageshttps://huggingface.co/datasets/SPRIGHT-T2I/spright
>mfw Research news07/01/24>Auto Cherry-Picker: Learning from High-quality Generative Data Driven by Languagehttps://arxiv.org/abs/2406.20085>Wavelets Are All You Need for Autoregressive Image Generationhttps://arxiv.org/abs/2406.19997>Parallax-tolerant Image Stitching via Segmentation-guided Multi-homography Warpinghttps://arxiv.org/abs/2406.19922>Solving Token Gradient Conflict in Mixture-of-Experts for Large Vision-Language Modelhttps://arxiv.org/abs/2406.19905>PM-VIS+: High-Performance Video Instance Segmentation without Video Annotationhttps://arxiv.org/abs/2406.19665>SK-VQA: Synthetic Knowledge Generation at Scale for Training Context-Augmented Multimodal LLMshttps://arxiv.org/abs/2406.19593>What Matters in Detecting AI-Generated Videos like Sora?https://arxiv.org/abs/2406.19568>Comparative Analysis Of Color Models For Human Perception And Visual Color Differencehttps://arxiv.org/abs/2406.19520>A Sanity Check for AI-generated Image Detectionhttps://arxiv.org/abs/2406.1943506/30/2024>ReflectanceFusion: Diffusion-based text to SVBRDF Generationhttps://arxiv.org/abs/2406.14565>Whiteboard-of-Thought: Thinking Step-by-Step Across Modalitieshttps://arxiv.org/abs/2406.14562>SiT: Symmetry-Invariant Transformers for Generalisation in Reinforcement Learninghttps://arxiv.org/abs/2406.15025>Masked Extended Attention for Zero-Shot Virtual Try-On In The Wildhttps://arxiv.org/abs/2406.15331>Enabling more efficient and cost-effective AI/ML systems with Collective Mind, virtualized MLOps, MLPerf, Collective Knowledge Playground and reproducible optimization tournamentshttps://arxiv.org/abs/2406.16791>Geometry-Aware Score Distillation via 3D Consistent Noising and Gradient Consistency Modelinghttps://arxiv.org/abs/2406.16695>Evaluating Visual and Cultural Interpretation: The K-Viscuit Benchmark with Human-VLM Collaborationhttps://arxiv.org/abs/2406.16469
>>101234197This thread is the only thing he has
is the "ani question" solved?
>>101233183>>101233337>>101234200>>101234410slider lora is best lora. Also mind if I ask what the checkpoint is?
headin into work
>>101234505I don't know - https://files.catbox.moe/zmfkzj.png
koff is based
>>101234505I think it's Zonkey anon
so /sdg/ will remain split forever?/ldg/ for good gens and technical discussions/sdg/ for gossip and slopdiscord for all the normie bastards?
>>101234327can i use my own checkpoints with fooocus?i want to go for a very specific concept art/black and white style and I don't think just loras will cut it.
>>101234927nope
Call me by my real name
>>101234970nigbo
>>101234970spaghetticat
>Trani destroyed the general
lets discuss technology
>>101235082fuck koff
>>101235082i want to hug youand trani is shit
>Trani is crying
>>101235153trani and debo**
am i still supposed to believe that debo has nothing to do with the ani hate?
>>101235163Agreed
>>101235163Debo hates 5 anons
>>101235223who are the anons? the 5 he hates and the other ones who post here and there
Don't ever post here again trani
>>101235236In order>Ran >Pastebin anon>Discord mod>Troony> inferior anon
>>101235281>tfw not on the list
>>101235281>I'm one of these
>>101235082unironically the last sane anon in this shithole
>>101235313idk about being sane, i just dont care about the ongoing drama/endless trolling, im here to discuss technology(post my pics)
>>101234681/sdg/ should merge with /ldg/ and the name should be /limg/, local image model general. For local video, it could be /lvmg/. Music&Audio is /lamg/.
>>101235348nta but that makes you quite sane compared to the schizos and debo
>>101235414Debo is the schizo
>>101235168its pretty well known that I'm a big fan of ani
>>101235476Pedo lover
>>101234560>https://files.catbox.moe/zmfkzj.png>>101234636Thank you, friends!>>101235382Loooove it>>101235348I feel the same way, I just want to post pics but every time I'm here it is just the same circle of people yelling at the same names. Just...Forget them?
>>101235564nice booba
>>101235348>>101235564makes you both quite sane desu desu
>>101235564>Forget them?Yeah, let them find a new identity and we'll start the cycle over again.
>>101234505What's this checkpoint?
first for "eye still suck"or maybe i do
this place just isn't the same without the avatars :/
>>101235382what is your process for generating higher resolutions like that?
/sdg/ is healing
Pony and SDXL are two different models. Stop mixing and merging them and posting them on civitai as SDXL. Pony isn't compatible with SDXL.
>>101235869Retard
debo come back to the discord and give me your personally identifiable information
>>101235869unironically stupid person
>>101234327Stop with the infighting between /sdg/ and /lmg/. Fucking grow up and make one proper general for people who aren't subhuman brown spergs.
>crying
Stop splitting up the general.
where is there a comparison of the different samplers and schedulers? I can't find it on any of the rentry's.
>>101235570The perfect booba>>101235620As sane as one can be!>>101235630Well glad I have no idea who they are nor even care/know of this cycle. Feels good just to post GENS. >>101235710https://civitai.com/models/520661/2dn-ponyEnjoy!
Are any of the online AI generators able to take this and turn it into a full colored image like the Frieren anime? If so, which would be the best to accomplish the task?
>>101235783Original size is 512 x 768. Usually hiresfix 2X using remancri @ 0.3 then another 2X using waifugan (Ultimate upscale). Sometime it works good.
>>101235917They're compatible in terms of merging and working without loras. But, their training data is is incompatible you retarded brown weeb subhuman. Also, Pony was trained by a retard.
>>101235564>Loooove itty. love those jugs
>>101235922/lmg/ is the large language model thread and I have no problem with them, they were helpful
Do you really need a refiner for anime/pony models?
>>101236083You don't need the refiner for anything, I would just not use it ever.
>>101235714seriously, eyes are the main thing with anime yet ai cant seem to do more than blobs
>>101236015yesread the op
>>101236054Retard
>>101236080>/ldg/You know what was mean't. These retards need to fuck off and let normal people make threads.
>>101236056Have some old jugs as I figure out what to GEN for later. Gotta cook up ideas.
>>101236294add "pregnant" to your prompts for bigger yet - kek
>>101236294full body shot, sophisticated woman high class party, giant lifted breasts, sly smile, 1960s style, maximally open red evening dressI want to see what you can do with that.
full body shot, sophisticated woman high class party, giant lifted breasts, sly smile, 1960s style, maximally open red evening dress
Having fun with ip-adapter controlnet
>>101235732I'll never leave you
>>101235900thats a pretty heckin cute debo ngl senpai
>>101236411if i cared enough i would link back to everytime you said you werent an avatar
>>101236410I could never figure out how to make it work
>>101236294Nice
>>101236432you should. that would own me so hard
>>101236369Here's one attempt>>101236434yeah, It's tricky.
>>101235732I vacuumed them all up into my doxxing discord where I threaten their employers and force them to goon to lewd imagery
>>101236479ip-adapter is the one where you can get consistent faces right?
lets ripping boysi need more spotify premium accounts desu
>>101236518You can do that with any model.instead of 1girl, woman, femaletry Lexi Belle, or just any girl name. You'll get constant face.
>>101236518yeah, it gets the clip embedding of an input face and lets you generate images of a person with that face but generally following the prompt unlike i2i or other controlnets like canny/depthgreat for what normies call "deepfake porn"
>>101236533>You'll get constant facemaybe. if if knows about lexi belle>>101236518For me it"s more about using a face in particular.
>>101236564Just having a name will gen similar face on all seeds.
>>101236460>>101236411
>>101236369>>101236479>>101236564Nice
>>101236604>pictured: my avataroh shit, you got me good
>>101236590lol 99% will result dreamshaper default face, lora is the only way
>>101236590If it doesn't recognize the name, you get the default face. At least with SD1.5.
you need to realize what's at stake if you join the discord....FUN!!!
>>101236460kumail nanjiani
You like sexy man?
You like fennec fox?
>>101236729>LonemIpsumI think this might not be a valid liceice
https://files.catbox.moe/nudlxk.flacStable audio is great.
Seeking fine tuning advice. I have made a Pony concept lora that is very good. (this one, if you must know (NSFW!!!) https://civitai.com/models/333006/ponyxl-golden-nectar)I am now trying to make a realistic version of it. Applying an earlier checkpoint of that lora to PonyRealism works surprising well at capturing the bulk of the concept in 3d. But it's still "off". So my idea is to then fine-tune that starting lora, on top of PonyRealism, on real images of the concept. So far it's showing promise, especially since I'm still gather the dataset and what I do have is incompletely tagged.One thing I've noticed though, is that early on in training it always gets deep fried. Picrel, this is a basic 1girl prompt, using PonyRealism, with no quality tags whatsoever, and no negative prompt. Top: starting anime lora, then middle and bottom are epoch 10 and 60 of the real image continued finetune. As you can see, training long enough partially recovers from the deep frying. UNET only training still deep fries. Extremely low learning rate still deep fries, just more slowly. And weirdly, having any negative prompt at all (even just a single period) makes it NOT deep fry the middle row.I spent an entire day trying to stop this from happening, but is it actually just working as expected? I've never really looked at gens early on in training before. My current working theory is that the model doesn't learn all the aspects of the data at the same rate. So if it's incompletely trained, all bets are off, and it could produce unpredictable results. In other words the training process doesn't smoothly interpolate the starting and ending point. But then, still a remaining mystery is why any negative prompt at all stops the early training deep fry effect.
>>101236670nice gen
>>101236326Lol maybe one day!>>101236369Leeeet's give it a go!>>101236445Thank ya thank ya! Baphomet and goat girls go hand in hand!>>101236479She cute!
mfw without any change in settings starting to get random outofmemory errors. might have to restart pc. techology
>>101236924nice booba
>>101236810that's what the glowies want you to think
>>101237054licee are a scam anyway. whats next? a licee to toast in my own damn toaster??
>>101237070oi m8 u need a loicee for that Img Model
>>101237070I get this reference
>>101237070Look m8, it's simple you can't get stabbed if you don't have 'getting stabbed loicee'
I dont even know what I'm cooking anymore
>>101237391like this
>>101237448new girl or old one I don't remember? could use some nose piercings imo, though I know those can be tough to get reliably
>>101237448Sasha?
>>101237464Just some goth chick
>>101237448>>101237528I need to make more goth chicks. Fishnet always fucks with me on bodies though. >>101237464Really cool gen!
>>101237547thanks! these gens are all over the place
Anyone got any idea what kind of lora this guy is using? https://x.com/aosiai123I haven't seen anime models that make such interesting backgrounds before.
>>101237547Do you do catbox requests?
>>101236054He somehow trained the 1.5 version of Pony V6 so wrong that the VAE doesn't have the same encoding as any other SD 1.5 VAE, you have to use the Reencode Latent node in Comfy to use a different one than the baked one lol
>>101234576very cool. I'd love to see more like this. what's your prompt if you don't mind me asking?
>>101237644I never have before but I could try! No guarantees though!>>101237568Haha, they are but it gives them a cool vibe.>>101237631More of this please and thank you.
>>101238164can you gen that style of horn reliably? I've always struggled with it
my brain is melting reading the op. I got a dalle reverse proxy what front end do I shove it into?
>>101238180The ram horns? Sometimes it gives me a lot of them but nah, it can be hit or miss. I'd love if it REALLY did a good curl but that is very far and few in between when it does.
>>101238386>dalle reverse proxynever heard of itand I've heard of everything>>101238388a shame>>101238402japanese quokka looks like a tempura bear
Good days
>>101238419It is, buuut I take what I can get. And I usually gun for ram horns with the red fit demon girl I make. At least she stays good, lol
>>101238499nice booba
weird
>>101238540Blessed with much booba tonight but sadly I am booba out and must go sleep.
https://suno.com/song/47e4d17b-1e38-4bf5-9b6f-ac2cd26cb6a8
hello puter friends
>>101238895hello
Where da weebs athttps://suno.com/song/2cb5c6da-781e-4c9d-bc5d-830227001c38
I wanted to post da nooz but we're stilla billion posts away
>>101238932That drum track is pretty good
sd3 banned belly buttons
https://www.youtube.com/watch?v=-2yxPal4wQI&t=1029sINTEL WONTHANK YOU INTEL2X FASTER PEFORMANCE
>>101239032with nooz ya looz
>>101235707>>101235608>>101235591wow catbox?
Last one from me, good night anons
>civit ai started to crumple on loli contentit was good while it lasts but it still hurts, this site is pretty much the perfect gooning machine
>>101239941>trani in shambles
who are the regulars here?
>>101240059meeveryone else is a tourist
>decides to give auto1111 a try because it contains some extensions that arent available in comfy or forge>gets oom on img2img at 1500x1500 with 24gb vram>cry in a cornerIm legitimately sad
>>101240195tiled vae bro
>>101240059idk I'm just here on vacation
>>101234349>Nearly half of US firms using AI say goal is to cut staffing costspretty based. i hope i live to see the end of useless jobs
>>101240494me too tbqhwyf, not only their jobs are useless but they get them by either being ass kissers or because they know someonethey don't earn their money
>>101239958Loli isn't exclusive to troons retard, fuck with that forced narration
>>101240514>they don't earn their moneyfucking exactly. if they did they wouldnt be getting replaced. motherfuckers wanna get paid minimum wage but nobody wants to earn it.
>>101240288It doesnt OOM when applying the VAE, it OOM before even starting to generate
>>101240589To add to that, yes multidiffusion makes it work, but then I can't use some other extensions that improve image quality because they are incompatible, its ogre
White Fox anime studio
>>101240835fuck nugget saviorfag dream
how and why is this lora 2gb?https://civitai.com/models/251417/midjourney-mimic
ded general
>>101240970high rank/dim, I think you can make even bigger ones, 90% of that is random numbers that don't contribute in anything to the output though :D
>>101239046cute heart choker desufixed her eyes for you tho
Anyone know how to improve clothing in comfyui? Something like creating a mask with clothing in it then inpainting it until it comes out great?
i miss schizo anon
Why do AI pics always have the same smiling face? Can AI not yet handle other expressions?There are mountains of AI big booba images and always they have the same smile as if to say 'cheeeeese' for a Facebook photo.Why?
>>101239609Sorry I dont share
>>101240195We heard you the first time let it go
>>101241557> smileAlthough you can get a smile it seems most realistic style images feature that vacant dead-inside stare of a woman getting paid to undress for a camera. Which makes sense given the image data fine tunes are usually trained on.
>>101241578How do you get an AI to make a gasping or surprised expression? There is no data for this
>>101241597convince the open source community to take pictures of themselves with various facial expressions desualso get non-autistics to label them because autists don't understand facial expressions
>>101241679why is she crying?
>>101241700/sdg/ is dying ...
i want to post nipples
>>101241679just some slime
>>101241806I do it all the time, what's the matter?
>>101241886tranny jannies ban me when there are visible female nipples on my pics
morning in america
>>101241905I know, just teasing
i bet i'm being retarded but can any anon point me in the right direction for cleaning up inpainting or masked areas? i swear i've seen it before, but basically i want to inpaint an image with several alterations and then keep those changes in the subject of the prompt while removing artifacts and creating a cohesive image. i've searched but can't find anything other than generic inpainting tutorials. what the fuck is this technique called?
me on the right
>>101242385>generic inpainting Isn't that what you're describing, how is what you want different? For the cleaning itself try using lama https://github.com/advimman/lama or migan https://github.com/Picsart-AI-Research/MI-GAN
>>101242675yeah it is generic inpainting, i guess my question is how to avoid artifacts when inpainting and make the image more seamless? or if there are artifacts present, is there basically an img2img that refines an image only very slightly. i swear comfyUI had an attribute for this, something that controlled "the strength of a prompt". i know its a shit explaination but thats the best i can think of. i'm playing around with the krita SD plugin so i don't have as much control as with comfyUI atm
>>101242941Try posting an image instead of these vague ambiguous descriptions lol
>threads are kept alive by 0anon and mint pussybless those two
>>101243262sorry anon im so very braindead todaylets say i have this image of fat man with a slug head, that's a hard prompt to get without inpainting. now i've inpainted it but you can see his clavicle/neck area looks all fucked where it joins the inpainting. basically i want to take this image and have the model try and work out a more cohesive way of joining that inpainting. i remember seeing some video in a comfyUI workflow where the last image in the workflow went through this step, it took the prompt but the prompt had very little influence on the final image, basically just cleaning/making the image neater and i have no idea what that adjustment was
has anyone gotten this 16channel vae to run in comfy with pixart?https://huggingface.co/AuraDiffusion/16ch-vae
>>101243535>chink shill
>>101243535this general is dedicated to Stable Diffusion onlyplease go away
>>101243720nope. NAI, MJ and outputs from other models are posted here all the time. also its the comfy general too
>>101243806thats considered spamits called stable diffusion general for a reason, shill
>>101243535it says its for sd3? I saw some people on twitter saying it could work for sdxl but idk how applicable the vae is to other model architectures. waiting for some smarter people to see it and chime in
>>101243523I would just create a lora for this.
>>101244258thats a very cool aesthetic. what tokens are most driving this style?
>>101243523You're thinking of img2img with low denoise. I'd try drawing on it first. You could also try lama/mi-gan on the neck area
>>101244304>abandoned University campus, cyberpunk, sci-fi, futuristic, graffiti about "NIGGERS", dust, smoke, ghosts, racism, hatred, sketch, misery, highly detailedthat's the prompt
>>101244336lol, uhhh... t-thanks
got any spicy catboxes?
>>101244366I'm not racist, i just do it for the art
Ah re can I find more stuff like this?https://x.com/solisolsoliHe posts very good AI stuff from time to time>>101244369What's that anon
Does SD (any iteration) benefit from >24GB of VRAM? I haven't done much proompting or training since pre-SDXL days since I went down the LLM rabbithole for a year instead. I am now in possession of 2x Ada A6000s, but my understanding is that having lots of VRAM isn't really that useful for SD, or at least not SD 1.5 due to the model architecture?
>>101244854just larger batch sizes, bigger upscales. also since you've got that much vram you can run sd on one card and llamas on the other.
>>101244881that's what I figured, yeah. higher batch size is nice though for finding a nice-looking seed more quickly though, I'd forgotten about that benefit.thanks
SD3 doesn't seem to know what a lithograph is. Prompting lithograph just makes the image greyscale.
>https://github.com/v0xie/sd-webui-incantations/blob/master/scripts/scfg.pyHow hard would it be to implement this on comfy/forge assuming you have moderate programming understanding and low/moderate SD understanding?
>>101244913its also useful for speeding up workflows that use more models in parallel. like if you're using controlnets or local LLMs or whatever, more vram means you can keep multiple models in memory vs wasting time swapping them around. gives you more space to be creative/liberal in how you stack resources into workflows
>SD3 doesn't seem to know what a wifebeater is. Prompting wifebeater just makes images of domestic abuse
>>101244914what about other art mediums? https://rikkar69.github.io/SDXL-artist-study/art-mediums/
>>101245001can you shut up?
>>101245001keep talking
>>101245001uneducated people trying to use slang termsmany such cases
MMorning anons
>>101245083morning friend
>>101245506>>101245506>>101245506
>>101245773https://files.catbox.moe/thnriu.pngthe prompt is a bit all over the place but hopefully you find something interesting in there
>>101245001so that means that they trained this model with domestic abuse pictures, but it's safe all right? :^)
>>101244464What's this New rule34 pedo crap?
>>101245001Still pretty closeWhat were you expecting to get
