Previous /sdg/ thread : >>101209305>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
>>101215144what is it?
Repost cause new bread:What do you guys recommend for someone trying to generate character reference for TTRPGs? I've been trying XL lately and it feels very hit-or-miss, might switch back to 1.5 cause with pony XL I'm just not getting very consistent results, not sure if it's my prompts or what. Just looking for general recommendations regarding what might work.
>>101215180Archway Anon, redbrick with moss.
>>101215197can you post an example of what you are trying to make?
>>101215197Pony should work. I'd just read through https://civitai.com/articles/983/insights-for-intermediates-how-to-craft-the-images-you-want-with-a1111and then spend some time browsing through civitai images and see how people are writing their prompts
Cursed thread.
This is now a blessed thread
>>101215375what's wrong with it?
>>101215468Cat's tongue don't look like that
excuse me
koff is based
>mfw Resource news06/29/2024>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 - Brand 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/spright06/28/2024>AuraSR-ComfyUIhttps://github.com/GreenLandisaLie/AuraSR-ComfyUI>ComfyUI Yolo Nas Pose TensorRT: Ultra Fast Pose Estimationhttps://github.com/yuvraj108c/ComfyUI-YoloNasPose-Tensorrt>OmniParse: Ingest/parse unstructured data into structured, actionable for GenAIhttps://github.com/adithya-s-k/omniparse>Fibottention: Inceptive Visual Representation Learning with Diverse Attention Across Headshttps://arxiv.org/abs/2406.19391>ControlNet Depth SDXL, support zoe, midiashttps://huggingface.co/xinsir/controlnet-depth-sdxl-1.0>xinsir / controlnet-tile-sdxl-1.0 https://huggingface.co/xinsir/controlnet-tile-sdxl-1.0>Compositional Image Decomposition with Diffusion Modelshttps://energy-based-model.github.io/decomp-diffusion/>TexPainter: Generative Mesh Texturing with Multi-view Consistencyhttps://quantuman134.github.io/>Fully Exploiting Every Real Sample: SuperPixel Sample Gradient Model Stealinghttps://arxiv.org/abs/2406.1854006/27/2024>Introducing AuraSR - An open reproduction of the GigaGAN Upscalerhttps://blog.fal.ai/introducing-aurasr-an-open-reproduction-of-the-gigagan-upscaler-2/
>mfw Research news06/29/2024>Hire: Hybrid-modal Interaction with Multiple Relational Enhancements for Image-Text Matchinghttps://arxiv.org/abs/2406.18579>Interdisciplinary Expertise to Advance Equitable Explainable AIhttps://arxiv.org/abs/2406.18563>Arboretum: A Large Multimodal Dataset Enabling AI for Biodiversityhttps://arxiv.org/abs/2406.17720>Directed Domain Fine-Tuning: Tailoring Separate Modalities for Specific Training Taskshttps://arxiv.org/abs/2406.16346>Exploring Cross-Domain Few-Shot Classification via Frequency-Aware Promptinghttps://arxiv.org/abs/2406.16422>Portrait3D: 3D Head Generation from Single In-the-wild Portrait Imagehttps://arxiv.org/abs/2406.16710>ClotheDreamer: Text-Guided Garment Generation with 3D Gaussianshttps://arxiv.org/abs/2406.16815>MantisScore: Building Automatic Metrics to Simulate Fine-grained Human Feedback for Video Generationhttps://arxiv.org/abs/2406.15252>Advanced Multimodal Deep Learning Architecture for Image-Text Matchinghttps://arxiv.org/abs/2406.15306>Landscape More Secure Than Portrait? Zooming Into the Directionality of Digital Images With Security Implicationshttps://arxiv.org/abs/2406.15206>CLIP-Decoder : ZeroShot Multilabel Classification using Multimodal CLIP Aligned Representationhttps://arxiv.org/abs/2406.14830>LM-IGTD: a 2D image generator for low-dimensional and mixed-type tabular data to leverage the potential of convolutional neural networkshttps://arxiv.org/abs/2406.1456606/28/2024>Dataset Size Recovery from LoRA Weightshttps://arxiv.org/abs/2406.19395>Enhancing Video-Language Representations with Structural Spatio-Temporal Alignmenthttps://arxiv.org/abs/2406.19255>Using diffusion model as constraint: Empower Image Restoration Network Training with Diffusion Modelhttps://arxiv.org/abs/2406.19030>AnyControl: Create Your Artwork with Versatile Control on Text-to-Image Generationhttps://any-control.github.io/
>>101215591agree. what can be more based than a pedohile?
>>101215375
pedo den
>>101215677Just say Trani den
>>101215692It wouldn't be accurate as there are more than one pedo sickos which frequent here.
>>101215704wait until you see whats going on in the discord ...
Trani is their king
>ranfag seething again
>>101215739queen*
>>101215750Yes, she is.
How do I make gandalf?
>>101215523nice catgal
>>101215780White of course. Epic.
Annie, are you okay?So, Annie, are you okay?Are you okay, Annie?Annie, are you okay?So, Annie, are you okay?Are you okay, Annie?Trani, are you okay?So, Trani, are you okay?Are you okay, Trani?Trani, are you okay?So, Trani, are you okay?Are you okay, Julien?
why is schizo so mean to ani?
Now that SD3 is doa, and stability is dead, buried and jeeted, what's next for localgen?Any news from the chinkslop front?
>>101215992we dont discuss chinese models here
>>101215992Just use whats avail, don't fall for the drama, who cares. kek IT'S OUT OF YOUR CONTROL
>>101216067Trani moment
>>101215992SD3 is """"dead"""" for coomers>it isn't really though
>>101216182>every poster is part of my persecution complex schizofantasmagoriaget well soon
>>101216223is anyone even hosting let alone making finetunes?
i wish it was winter right now
Morning /sdg/! Hope your GENS have been going well!
>>101216333I shouldn't have to explain how SD3 can be wielded without finetunes. If you can't conceive how it can be used you're unironically not going to make it.
>>101216374good morning
>>101216389show some of your SD3 gens then, big guy
>>101216370But is summer time! Though I should do some Winter gens, hmmm.>>101216392Morning morning!
>>101216522>But is summer time!based northern hemisphere enjoyer
I think trani is shit
>>101216541Truly is a blessing
>>101215144>>101215437>>101215493>>101215838>>101216239these are great
>>101216619>Truly is a blessingindeed
>>101216612Me too
>>101216632thank you
>>101216619prompt?
>>101216765Enjoy!https://files.catbox.moe/gqcnv4nkqe.png
>>101216652>Default womanif only
>>101216959:\
>>101216959not sure you copied the url correctly as that's a bit long for a catbox filename
I keep getting the message:>NansException: A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.Anybody knows what could be causing this and how to fix? Its odd, it will work ust fine for a time, then this shows up when I create.
>>101217172>Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.
>>101217009>>101217046Huh, I clicked the little thing and just copied it over. How odd, yeah, here it is again, I see that it added stuff.https://files.catbox.moe/gqcnqe.pngSorry about that!
>he didn't defrag his tensor cores
i miss debo
>>101217233thnx
any1 who uses the exponential or polyexponential schedulers: why
>>101216981weird skin textures
>>101217266what do you think he's up to rn?
>>101217363indeed. Bad upscale
>>101217295Nice. What is that?
>>101216223>SD3 is """"dead"""" for coomersdead for coomers means dead period, remember SD2.0/2.1, SD3 did a repeat of that mistake
>>101217701a doggie
>>101217295>>101217701>Nice. What is that?looks like Scorpiushttps://www.youtube.com/watch?v=pHW6ntf1gME
lol the background
Morning anons
>>101217380bet he's busy thinking about schizoanon and his truth bombs
my eyes are up here
>>101217358A1111 uses them by default with SDE samplers, if you leave the scheduler on Automatic>>101217716Have you used the previous version of that checkpoint (1.1)? I can't tell for sure if 2.0 is really an upgrade yet
>>101218263https://www.youtube.com/watch?v=Vdd4rBlsj2o
>>101218279>HOW ABOUT SOME SCAT YOU LITTLE TWATI loved that game, still surprised Nintendo allowed it in the first place kek
hate to fucking ask this because it means I've become what I hate, but just suppose I wanted a caucasian female in an SD3 gen, how do I maximise my chances of that happening?
>>101218457Put african, hispanic, asian in the negatives
>>101218457remember to put trany in negative
>>101218505yep, I'm missing every other nationality but I think that's pissing in the windthe only reliable way I've found so far is make her ginger, but that's not what I want
>>101218527>black, Black, BLACK
>>101218527ginger with a different coloured wig
>>101218457have you tried not being racist?
>>101217233Do these wall-of-text schizo negatives actually have a measurable impact on anything?
>>101218562he wants to make a caricature that makes fun of white women for being racistsso preventing him from making that gen is raciststop being racist please and gen white women
>>101218527more specific nationality in the positive helps, like irish or finnish or something
>>101218671Yes, they can and will negatively (pun intended) affect the gen. The less you have any negs the better it is.Only techlets think otherwise.
>>101218671It's easy to test mate...In general not all negatives will affect all gensPrompting against some stuff can have nasty side effectsAlways remember that SD works by associationSo fine tune your negatives like any other part of the prompt
>>101218671My approach is to start with a much shorter list of general negatives (stuff to address extra/bad hands, watermarks, low quality images). Then I add negatives based on problems that seem to recur as I keep generating images. Like, I might not put "patreon logo" in the negatives at first, but if I kept seeing the patron logo, then I would add it.
>>101218687thinly veiled racism
>>101218969reporting to reeducation center immediately
imagine still using negatives in the year 2024
>>101219064aaaand I only just realized the hands are fucked up. But seriously, getting that swimsuit was the bigger challenge
Does anyone have examples of images made by a lora of a single artist's drawings?
>>101219071>the hands are fucked upin an AI generated image?no way, I refuse to believe it
>>101219071>getting it to hold in that positioni can imagine
>>101218892this++
i swapped prompt with negative
>>101219151
>>101219036this attitude is why XL gens are shit
>>101219151>>101219156it's not really a negative prompt, its two generations mixed together, you could have as many prompts as you want and just make up another classifier free guidance type calculation to mix them together
>>101219194-_-
>>101219180anon is simply a promptlet
>>101218671Yeah, just an old holdover I kept from looking at other GENS and making one of those quick pos/neg load thingys. Really the only stuff I added for the picture was the early stuff like blood and clothing parts. I really haven't gone back and cleaned some of the stuff up which I should!
>>101219298realistic XL gens look blurry, everybody knows this and people stick to 1.5sure it doesn't need negatives since they tried to bake "beauty" into the modelin the process they made something that's only capable of producing smooth airbrushed fashion photo bullshitbut heeey, it makes le pretty pictures by default, no negatives! hur durrrr
>>101219359oooohhh i see. you were never able to git good with xl KEK. whats it like being stuck in the past?
"Fine, you can walk home with me! Just don't ride me!"
>>101219413pretty indecent of her to have her ass hanging out like that
>>101219546Eh, centaur modesty standards. You wouldn't understand. Besides, her tail does enough work as it is(I did have a gen with a "full skirt" ,but it still looked odd. In fact, most of the gens with a skirt basically looked like a biped's skirt slapped on)
>>101218197>can't tell for sure if 2.0 is really an upgrade yetInteresting. I might have. I do like this version.
>>101219413cute!
>>101220087Beautiful.
>>101220167
trying SD3 on A1111 sd3 branch.
>>101220328how is itheard it's barely functional, plus controlnet doesn't work so it has no advantage over comfy for now
>>101220087Nice devon island picture kek>>101220309>>101220328Incredible
you are all lazy thieves and should be ashamed
anyone else think it's kinda fucked up how much the resolution aspect ratio affects the output?for example i was making hentai and couldn't get the bitch to deepthroat the dick until I made it widescreen.
>>101220387
>>101220355meh, not sure on what makes it gen a good image. For this image (which is nice) I had to copy the prompt from another image on civit.
>>101219064cool swimsuit indeed
drank a bunch of ckoffee and napped a bit
It did add light rays as instructed but removed the waterfall in the gen. I had to put (full body shot:1.8).I must say that it's pretty good as a base model. I couldn't do anything like this with SDXL base.
>>101219345>>101220309beautiful
>>101220481from the little I've used it, parsing text is its biggest strength. No bullshit like in 1.5/xl saying "red shoes" and everything is red, even the grassit encourages verbosity for sure
>>101220387I may be lazy and a thief but I'm not whatever the third thing you said
>>101215523nice
>>101220365Thank ye thank ye. Just messing about with new things and really loved how it turned out.
what the sigma?
sigma all nite long
>>101220387I know
>>101220715she thin
without loras, especially no age or weight sliders, sadness
>Trani is leaking fecal juice
>>101220418Uncannily Skyrim-like metal/leather textures.
>>101220895oh no, centaur anon got jannied
claws are hard
>>101221187thats weird. does that count as mlp or something?
No, I deleted it myself. I realized that that might be a bit too much.
>>101221356>>101221187
>>101221258>>101221297>>101221221I hate trying to GEN weapons but cool stuff either way!
>>101221368one of her Loras tries to pop claws here and there, if you prompt claws it's usually pretty jank
2B's butt overpowers the Samurai armor lora
>>101221408>2Bwas it being outputted by SD3 2B though?>badabum tss...
>>101221436pony
>>101221436i got it
>>101221468nobody likes a brown noser
>>101221329does what, his image? no, it just had too much ass in it
Not sure why the legs look like melting wax.
>>101221557Why knot?
>>101221679real women have cellulite
>>101221368lora for magazine cover?
When the uniform designers aren't fully accustomed to the centaur exchange program
>>101221839
>>101221901nice
>>101221941thanks
wow such a.i.
>>101221901>>101221937>>101221978>>101221839What are you using to copy the pose?
>>101222054it's the same image, so probably img2img>>101221978amirite?
>>101222054img2img>>101222065yeah
>>101222095without age slider version
>>101222065I've tried ControlNet for that but it has several options: lineart, Depth, canny, openPose etc.>>101222065oh ok thanks
>>101222176wrong image.this one is SDXL, other is sigma.
>>101221751No lora, just prompts! "magazine cover, posing for fashion magazine cover" Can change out fashion for what you want or just posing for magazine cover.
>>101222250how long until it does text like that in the image perfectly
Something like this
>>101222368or this
>>101222291you're kinda stealin my vibe bro
>>101222454kek yours are more coherent, back to arches.
>>101222301You mean how many GENS till it does perfect text? That I wouldn't know since I'm really not trying to hit that perfect lettering. Just want the feel of the cover. Sorry, lol!>>101222505Love the demon, makes me want to try some of my old demon lady GENS again.
>>101222590nice ass!
>>101222590>You mean how many GENS till it does perfect text?no i meant when will the models be that good
>>101222590And thanks!
>>101222250awesome, thanks
sunday unfunday
Koff > jules
what's the best/proper way to prompt the comic panel style thing
>>101222613really makes you think
>>101222682Planning a heist for the squirrels seed silo?
>>101222658I've seen some loras but the best method i've seen was to gen each image then get comic panel layout software and design your page layout and cut and insert each image
>>101222590nice butt
>>101222601>>101222610Haha, you're welcome! I do miss doing 1.5D stuff like your GENS so at least I can get KINDA close to it with this Pony model.>>101222605Oh, man...I have no idea. That sort of stuff I don't even look at. I just let it happen and then be like 'cool', I can do that.>>101222622Welcome! Hope you do some great stuff!>>101222766Thanks! Love your lighting! And the abs/face are fucking great too.dgs8d
I like it when centaurs have "waist fur" that's visible above whatever they're wearing. Too bad getting it to actually show is a hit or miss sometimes
>>101223427could train a lora off the gens where it pops up
>>101218259Eating disorders are no joke =)
>>1012234411) it comes up way too little. Getting enough gens would take way too long, putting aside I have no idea how many gens I would need2) it mostly is only an issue when the centaur isn't "bottomless"3) I feel like I would need a better way to describe this...thingAnd yes, picrel is what I'm trying to avoid
>>101219088>>101221329>>101223081Such a cool style
>still no news on 5000 series release, specs, and pricingBros I can't live with 30 second generation time for medium resolution images any more. Should I just buy a 4090 and KMS when the 5080 beats it for half the price?
>>101223511thanks! I have a sprinkling of some MtG artists along with (graphic novel), which turned out as a decent combo. its ironic though that sd3 doesn't really produce the artists styles but the artists influence the aesthetic. I just wish I didn't get sd3 hands every gen....>>101223536if you've got that kind of loose change, why not just pay for a cloud service while you wait for the 50s?
>>101223509Gate expectations.
>>101223563Because I don't want to generate spicy material on someone else's computer, who also has my contact details.
>>101223636oh, ok. thats fair lol. another option is to buy a 4090 now and then resell it when the 50s drop. it'll probably hold its value pretty well
>>101223687>>101223687>>101223687
>>101223679>sky isn't filled with dozens of planetsas a fellow space gen enjoyer, I'm quite impressed
>>101223738Air brush extra stuff out : the little man and extra planets. You can smudge the planets into perfect circles with select circle in an image manipulation program.
>>101223622But what is gate reality?
Why is it when i user some Loras (on a Pony merge) and when i have its weight at lets say <lora:xyz:0.5> it works, <lora:xyz:0.95> also worksbut inbetween somewhere - <lora:xyz:0.75> - i just get this weird colored noise.Anyone know whats up with that?
4
3
I'm helping
1
Does anyone know a keyword or maybe Lora for Pony that makes a woman really tall and bulky? Basically think space marine stature thats 10+ heads tall.Also, something that makes a woman buff and stocky but not ripped, more of a strongwoman build?
>>101223536Why not a second hand 3090?