Previous /sdg/ thread : >>101360320>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>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>Compositional Video Generation as Flow Equalizationhttps://adamdad.github.io/vico>Big Tech needs to generate $600 billion to justify AI hardware expenditurehttps://www.techspot.com/news/103699-big-tech-needs-generate-600-billion-annual-revenue.html
>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
First for desk toys
I think Gradio SUCKS!
>>101368617about right.. but whats the alternative? electron? or god forbid write a native app?
anon asked for butterflies, you did not deliver.
>>101368743merkava just refilled
>>101368376merkava just refilled, let's go boyhs
gotta take a quick break
>>101368661How does writing HTML/JS melt the brain of Python devs?
>>101368833HTML/JS isn't programming, Python is
Beep boop
>>101368863That's why you need daddy Gradio to hold your hand.
you don't need gradio
>>101368833python devs cant into async or dom interactions
>>101368882you don't need async in python
>>101368882>thinking multithreaded applications need python>not just java
>>101368786>>101368869rodit
>>101368886thats what I'm saying. python dvs can't into async, which is why javascript melts their lil puny brains
>>101368833>>101368873Can't you just use a CLI to SD?
>>101368958Nice rodit
>>101369170prompt: "big booba animu gurl"the style is extremely fucking generic anime shit
>>101369170Get better taste in styles
>>101369220ironic
>>101369253forgot your gen
>>101369210>>101369220thanks guys :3
>>101369170>since I'm here, I was wondering if anyone has any idea how to prompt something like this, or at least the style picrel has?serious answer, not just for your specific question but in general https://huggingface.co/spaces/pharmapsychotic/CLIP-Interrogatorupload your picture and get a prompt
>>101369410clip similarity is bullshit
>>101369272You're welcome
This is jungle fevertesting stealth PAPA NOVEMBER GOLF info transmission on channel Alphaplease acknowledge
>>101369511Weapon, hands, checks-out.nice gen
>adult body>childs face
That's not all she's got that's child sized.
>>101369620Well women refused to put on the anti-aging helmets so
Finally figured out how to make my model do what I want how I want.Except it flat out fucking refuses to properly light a scene. The colors aren't washed out like the vae-issue, but it just won't fucking illuminate anything properly.
>>101369690....nevermind.Adding a lamp works. wtf
>>101369700I really hate it when terms like backlit,sidelit,underlit, etc stop working.
>>101369700like in real life
>>101369170T-ponynai3
>>101369170masterpiece,bestquality, score_9, sweating massive big tity bitch in a white [see-through] tank top looking down at viewer astonished to see you in the bushes, one tiny tooth, grey hair, big hat has cattle, pink bow, sweat droplet on chin, it's nice out wow look at those clouds
>>101369898>>101369855 interesting styles
What's my best option to fix hands? From what I've read it's inpaint with a controlnet? How do I go about that and would it matter what checkpoint I used for my base image when doing this?
>>101370009Adetailer used to do pretty good but I can't use it on my current set up.Oddly enough, 2dnpony v10 does hands and even feet pretty reliably
>>101370009The vast majority of pony derivatives can do hands and feet pretty well and then running it through a detailer helps even more. It’s more worth finding out why you gen very bad hands instead of fixing them.
>>101370041you could use bokeh to avoid twisted horror creatures in background
>>101370151in negs or positives?
>>101370151that’s their style bro
>>101369602confirm trans(mission)
the eff.. I just updated automatic1111 and everything is 1/4 the speed .. heck I should have backuped .. it upgraded from pytorch2.0.1+cu118 to pytorch2.1.2+cu121is there a known bug? cause with 2 loras and an embeding I suddenly have to wait 5 minutes for the picture it ran 1 minute in previously
>>101370165positives, of course, depth of field should work as well but it's longer it also blurs wrecked cars or errors in buildings and gives a more natural photographic feel ,or pick a simple background where nothing can go wrong
>>101370121More quadriceps
>>101370279Okay thanks I'll try that>>101370280
>>101370151dont listen to this anon he's trying to make the twisted horror creature blurry so you don't see it coming
>>101370337cool
man those threads are unironically depressing now
>>101370174did you get the prompt?
>>101370436which threads
>>101370436yeah
>>101370436>>101370455He means her clothing invokes feelings of meloncholy in a literal sense at present
https://xcancel.com/cloneofsimo/status/1811062800695054500#mLooks like this mf is making a really uncensored model based on the SD3 architecturehttps://github.com/cloneofsimo/minRF
>>101370534Aren't base models really expensive though? Is he going to train it or just provide the tech?
>>101370436i miss schizo anon too
>>101370573Looks like he's gonna train the whole stuff and give us the model, let's wait and see I guess
>>101370241meh.. auto1111 is dead it seems, no updates since may! .. just installed SDNext works fine and flawless.. guess I have to wrap my head around new slightly different interfaces
>>101370534>>101370573Nvm dug up this>There were unintended hype here, im sorry for overhyping it on the last post.>* The model is not open-midjourney-class model nor should you expect it to.>* The model is very large (6.8B) and very undertrained. (30% of SD3 compute budget) So it will be more difficult to train with a consumer gpu, but we might continue to train it in the future.>* The model is doing great on some evals, and imo is generally just better than sd3 medium, but not by a large margin.>* but it's not finished yet. I'll upload some post now and then but it's never intended to be like, a statement to claim victory or whatever
>>101370667>The model is very large (6.8B)that's the imagegen model + the text encoder? That's kinda big if it's just the imagegen model
Need some distributed computing BOINC style load sharing.There's millions of us coomers out here that'd pitch in.
>>101370534Here's to hoping.
>>101370667hoping/>
Free from 3 day jail
>>101370797wish it were longer
>>101370534I don't know if I should keep scrolling through his twitter posts lest I acquire more anticipation
>It's the episode where Truman realizes he can generate all the Tinkerbell bukkake porn he wants then spends a week doing literally nothing else while drinking.
>>101370797I bet there's some obvious reason money is not square shaped that we never have to think about.
>>101370827Was an Azula nip slip>>101370868Wasn't there a paper coin bill in Europe at one time?
>>101370855did something similar for a few months
Does Kolors work on windows if we use ComfyUI with it?
>>101368376anyone else here experimenting with lora training? whats your go too settings for simple characters (not photorealistic), an anon mentioned using clip skip 2 instead of one and lowering network alpha to 32-16 but hasnt seemed to have made any significant difference the character has realistic puffy lips despite being an anime girl, also my training data loss never goes higher than 0.1 in 10 epochs, am i underfitting?
downscaling makes all my images look bad, what do? do i just have to resize in an external editor? all resizing i tried in comfy makes it come out blurry when i just wanna shrink an image
>mfw Resource news07/11/2024>LLaVA-NeXT-Interleave: Tackling Multi-image, Video, and 3D in Large Multimodal Modelshttps://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 Extractionhttps://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>Compositional Video Generation as Flow Equalizationhttps://adamdad.github.io/vico>Big Tech needs to generate $600 billion to justify AI hardware expenditurehttps://www.techspot.com/news/103699-big-tech-needs-generate-600-billion-annual-revenue.html
>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 Alignmenthttps://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
>>101371118>SLUTXTROPLIOMD
you are pathetic
woman collapses and falls apart into black goo
>>101371164
>>101371172who?
>>101371225meemaw
>>101370909Kind of thinking about setting up another box to just gen on and use this one for LLM cards from the cream of the crop, as it were.
>mfw Research news07/11/2024>4DiMMs: 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>MARGE: Race Mixing 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 Alignmenthttps://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>POO-PIPE: 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>Dethinking I2V Adaptation: An Chud-centric Perspectivehttps://arxiv.org/abs/2407.06871
>mfw Research news07/11/2026>4DiMMs: Controlling Space and Time with Diffusion Modelshttps://4d-diffusion.github.io>LeLeMulti-task Prompt Words Learning for Social Media Content Generationhttps://arxiv.org/abs/2407.07771>VGayEnhancer: Generative Space-Time Enhancement for Video Generationhttps://arxiv.org/abs/2407.07667>Pruning Pission-Language Models with Candidate Labels by Prompt Alignmenthttps://arxiv.org/abs/2407.07638>MARGE: Race Mixing of Auto-Regressive Models for Fine-grained Text-to-imagehttps://arxiv.org/abs/2407.07614>InstructLayout: Depissistruction-Driven 2D and 3D Layout Synthesishttps://arxiv.org/abs/2407.07580>Zero-Slut Class Unlearning in CLIP with Synthetic Sampleshttps://arxiv.org/abs/2407.07485>Video-to-Audio Niggereneration with Hidden Alignmenthttps://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>BiLight: LGBTechnical Report on the Motern AI Video Language Modelhttps://arxiv.org/abs/2407.07325>Sloot-Shot Image Generation by Conditional Relaxing Diffusion Inversionhttps://arxiv.org/abs/2407.07249>Deinferreffettereference-based Controllable Scene Stylization with Gaussian Splattinghttps://arxiv.org/abs/2407.07220>ColorPajeet: Color Prompt Learning with Diffusion Models via Color and Shape Disentanglementhttps://moatifbutt.github.io/colorpeel>CumFreeDiff: Camera-free Image to Panorama Generation with Diffusion Modelhttps://arxiv.org/abs/2407.07174>Pissiffusion Model-Based Video Editing: A Surveyhttps://arxiv.org/abs/2407.07111>POO-PIPE: Variational View Invariant Pose Embeddinghttps://v-vipe.github.io>Gaytent Space Imaginghttps://arxiv.org/abs/2407.07052>HumanLefiner: Benchmarking Abnormal Human Generation and Refining with Coarse-to-fine Pose-Reversible Guidancehttps://arxiv.org/abs/2407.06937>Dethinking I2V Adaptation: An Chud-centric Perspectivehttps://arxiv.org/abs/2407.06871
will trani save StabilityAIâ„¢?
pitiful display of dignity
@101371353sad nig*bo! gone too far(t) this time
I used the same tags as this image on the booru but still won't get this position. Any ideas?
>>101371416love the screencap style and ears
@101371431d*bo
>>101371438not debo
>>101371431sank yew!
>>10137132116ch POO-PIPE when? It's over!
>>101371457forgot your gen
>"stealth posting"
>>101371571>|["`'quoting someone who is themselves quoting someone else'`"]what the fuck
I hope the anatomic diversity has been kept to a minimum with this one.
>
>>101371563nah I didn’t
>>101371836>nogen
>>101371846yeah good job
looks like the last wave of bans expired and all the ne'er-do-wells are returning with no lessons learned
imagine caring
>>101371898this general would be a lot better if you would leave forever d*b*
>>101371923don't call me d*b*, d*b*
>>101371898haven’t been banned since samus
>>101371882thanks
>>101371898thread schizo
so, does anyone want to talk about lora training? i dont come here much but it looks like half naked anime girls mostly, am i in the wrong place?
>>101372088shut up, faggot
>>101372088never enough half naked anime girls desu
>>101372088sure what would you like to talk about
>>101372116>>101372138look the half naked anime girls are everywhere online i thought this was a technology board, where people actually talk about the technology relating to making half naked anime girls>>101372154hey, im working on lora training at the moment and having some issue trying to get it right, i thought i might be underfitting it since the training rate never seems to spike after 10 epochs, the character is suppose to be a simplified one (non photorealistic, anime girl esque, flat colors ect) but when its trained it comes out realistic looking despite being trained on an anime model and ran on sd1.5 anime model, just want a pointer as to where I'm going wrong, been experimenting with repeats, epochs, max training steps, image count and seemingly getting nowhere
>he doesn't know
should we tell him?
>>101371898Nipplegate
>>101369511Nice, but what calibre is that, .177 BB?
i'm working on a local image ranking tool where you just get shown two images from a set and you pick the better one to help A/B test samplers/prompts etc, should i make it into a auto extension or a standalone thing(so far it's mostly confirmed what everyone already knows DPM++ 2M SDE is goated)
>>101368661I wrote my personal flutter client for Android and PC that uses swarm API
Ted stated "Parlay. Is metal G. G Ted mode. iBark."I post bears and have talking robots...>W220K2
my therapist says I have a wide sexual palette…
>>101372188>sd1.5 anime modelWhich one? You should train on NAI if it's a 1.5 model.
I got soap in my eye and it really hurts.
>>101372770this is why I never shower. too risky
> inpaint a hand> turns into a foot instead> increase the padding by 15 pixels, so a piece of the shoulder is in the inpaint image now> it turns into a hand> thumb on the wrong side> add one pixel from the other hand, change image to horizontal, remove the shoulder> inpaint works, but hand too small in total image, not enough pixels. hand now shaped correctly but still kinda wonky> copy/paste new image into the inpainting window> restart from the top with denoising lowered > starts turning into a foot again
Anyone making money on this shit yet
>>101372873yeah a little
>>101367606catbox?
fox
>>101372804>>101372904debo btw
not voting in novembrr
>>101372992Sticking it to Trump and Biden. Yeah.
>>101372914If you’re around laterotherwise it’s 0003 - Pony Delta with laserflip, yd, cutesexyrobutts, and redrop LoRA at 0.3 weight.
>>101373050I'll be around, just (you) me. I'd appreciate a catbox but thanks for the reply in the meantime, that model is nice.
damnit pytorch! whats wrong with you? I just spent two hours wondering what the heck is wrong with my GPU until I found that cuda121 version is like totally crippled and runs like a snail on my 4090 compared to cuda118 version of pytorch .. damnit! anyhow ... you all having a good evenin?
https://youtu.be/Y_EougWltgs?si=mf_lY7sYGOSmkoY7proompting soundtrack
@mizuI just tried using the AND keyword with a1111 parsing and it blew up. is that keyword supposed to work?
>>101372814I import the image to paint.net, copy + paste a hand I found on yandex images, and then use Differential Diffusion masking workflow to inpaint. That's my technique
>>101373379(orange cat AND bread)it was meant to fuse two prompts .. not sure if its even still supported? but thats how I used it in October 2022
>>101373338It's a little saucy sometimes:Error: CUDA out of memory. Tried to allocate 2.00 MiB (GPU 0; 7.91 GiB total capacity; 7.46 GiB already allocated; 6.50 MiB free; 7.71 GiB reserved in total by PyTorch)
>>101373411thought the way was (cat|bread) or [cat|bread]
>>101373405Sometimes I will grab a hand from somewhere and use it or use meshgraphormer but as much as I complain in my experience manually doing it that way is faster for me if I am going through a lot of images semi-quickly especially if I clean it up a bit in gimp first. I was mostly trying to bring up how sensitive inpainting is to the other context in the image you supply it.
https://huggingface.co/fal/AuraFlowWhat the fuck is this?
>>101373426ya thats the better syntax .. but AND as keyword was a thing at least in SD1.x .. or its an automatic1111 implementation thing .. even sdxl seems to still give some resullts for (orange cat) AND bread
>>101373411thats how I'm using the keyword but it breaks in mizu's encoder node>>101373426[cat|bread] basically edits the prompt every other step, searching for cat then dog then cat again. AND creates a completely separate conditioning which acts as a parallel sampling (from how I understand it, but I never really understood it)>>101373443https://blog.fal.ai/auraflow/
>>101373466>Polygon hair on a semi realistic stylized modelCursed.
>>101373441Have you tried adjusting the sensitivity with a Gaussian Blur node
>>101373518its more papercraft than polygon
cuddly, (cat or..?)
>>101373466>Right around this time, we were convinced that a SOTA open-sourced model is the way forward for this space to move forward.
>>101373466I don't think Auraflow knows Quokkas either.
>>101370128I think its my loras, I'm realizing what is consistently giving me bad hand gens with some. Aside from that, I also started using depth map + openpose and can consistently get good hand gens now. What other things should I be wary of to avoid bad hands?
>>101373729it seems to know they're kinda australian and kinda marsupial, lol. thats such a weird close-but-not-close-enough gen
>>101372188did you tag your data as anime or illustrated?
The captcha is mad because I always click the center pixels. It's not my fault; I'm just good at pool, and darts, and bowling, and apparently environmental science projects. Then the captcha didn't let me post until I disconnected. My dogbot called me a magic fucking zerker. Jannies stay piss mad at puddles."Mainframe king! I bark for you!" EINE was easy on an i5.>28NN0DTo my oh dear.>
>>101373780reason is one of the few things I know about loras is if you tag something it will be "ignored", and you shouldn't tag stuff that you want to show up when you use the lora. This might extend to the medium or style, that would make sense.
why is cyberspace an impossible concept. it just gives me monitors
>>101373844I guess you have to describe the concept in detail, something like "floating in light, surrounded by holographic geometric scultures" etc. .. cyberspace is to abstract, maybe 1980s science fiction helps to?
>>101373844my cyberspace take> immaterial space, transparent, translucent, ghost like, floating in light, surrounded by holographic geometric sculptures, in cyberspace, retro futurism, 1980s, realistic, depth of field, hacker girl, freckles, green eyes, floating, tight bodysuit, sharp lines, details
>>101373729Cursed kangaroo quokka
also: dissolving into goospace>chibi slime AND 1girl in immaterial virtual reality, electronics, cables, transparent, translucent, ghost like, floating in light, surrounded by holographic geometric sculptures, in cyberspace, retro futurism, 1980s, realistic, depth of field, hacker girl, freckles, green eyes, floating, tight bodysuit, sharp lines, details, floating displays
>>101374008>>101374071not a bad take on cyberspace but I feel like it should be a stronger token on its own>>101374086the new era of 3G printers (the G stands for goo)
if person I think ill stick with 2D .. realism gives me the creeps
>>101374222Neat
>>101372724mistoonanime_v30, what is NIA? ive heard that referenced before as a good 1.5 model but couldnt find much on it
>>101373780no, i only tagged the main tag, the tags for the outfits and the backgrounds were tagged extensively to make sure it knew what was what, ive also experimented tagging with emotions and positions but for some wierd reason the one thing that seems to have helped was training it on 512/512 instead of 768/768, should i be tagging it as anime? wouldn't that confuse the training and see it as a separate entity?
>>101374258>NIANAI* was a leaked SD1.5 fine tune, which was the first model ever to be able to produce capable anime and hentai.. made by some horndogs in a cellar with a server farm in summer 2022 .. but they had bad security, so someone leaked it and it was the base of nearly all early SD1.5 models that were useful .. later went obsolete, you dont need it anymore .. also many many most anime mixes that came later probably used it as tuning base
I'm starting to suspect that Automatic1111 is doing something fucky with my PC when it comes to sleep mode. Either that or some component is starting to sour in an esoteric way despite it only being a few monthsAnd if I let the PC go to sleep while SD is still "active", usually I can wake up the PC fine, then do a few more gens, then I suddenly get a BSOD pertaining to, IIRC, usually Kmode_Exception_not_handled (though once I got a different error that had something to do with tcpip.sys)If I do some genning then shut down the terminal or whatever, then the computer will fail to "wake up". It'll just...reboot, basically. OR shut down. Contrast to today. I haven't touched SD at all. I did, however, play a video game for a few hours. Let the PC go to sleep while I did the dishes. I was still able to wake up the PC fine.MSI B650 Tomahawk WifiRyzen 7600XTEAMGROUP T-Force Vulcan DDR5 32GB (2x16GB) 6000MHz (PC5-48000) CL30 RAM
>>101374258You want to use NAI because it's the model that all 1.5 anime models are built on. If you train a lora on mistoonanime_v30, it won't work well with other models.
>>101374294i can give it a try then i guess, isnt on civitai though where the hell do i find it?
>>101374315I think this is it. It's a leaked model, the Novel AI anime model, so sites like civitai won't host it I guesshttps://huggingface.co/hollowstrawberry/stable-diffusion-guide/tree/main/models
>>101374290>esotericya well.. back in the day id say you got pickled, but these days id just say its unstable software will full hardware access, my PC is rock stable, but with a1111 it can just randomly crash under heavy load, probably something in the conglomerate of software has memory leaksI dont think its your hardware. I suspected as much first to, but then after a month break of SD and no crashes at all it was obvious, a1111 just is a monster. Makes nice pictures tho.
>>101374360I want to say I haven't had this sort of problem with my last build, but I've definitely had problems with sleep mode/waking up from that there, though I don't remember if I had A1111 open or anything like thatStill, surprisingly there don't seem to be that many issues raised on Github discussing this
>>101374383different OS? different GPU? could be anything .. but ya as I said a1111 makes everything unstablealso have a manul
>>101372914>>101373298https://files.catbox.moe/mv08vg.pngsorry for mess anon and I forgot to remove the virgin destroyer lora
>>101374414thanks
>>101374414>"Comfy">it's not comfy
>>101374423>"Automatic">it's not automatic
>>101374401The last build was an ASRock Z170 Extreme3 with a 6th gen Intel CPU. Both builds had an RTX 3060 and Windows 10, both of which I've carried over to this one
>>101374423>>101374435both so very true, in comfy I get lost in noodle workflow, in automatic I have to repeat steps over and over again .. *sigh* but my brain doesnt get creative in noodle soup, so I stick with the repetition
>>101373443>16 niggabytes
>>101374437Actually I remember the old motherboard sometimes would "turn off" instead of going to sleep, and when I turn it on again, I would just get a black screen. I usually just do a hard reset, after which the PC goes back to where I left everything.Weird
>>101374437nothing much ya can do but observe and hope the software part that causes problems gets fixed, although upgrading a1111 is a gamble on its on.. earlier this evening I had the silly idea to upgrade xformers and torch, cost me 3 hours to finally give up and revert back cause I just couldnt fix the performance loss
>>101374499I'm on 1.6.1. The "he pulled" meme has made me hesitant to update. But now I'm not so sure
I guess the FBB just gave anon a vacation for his last picture .. hoot hoot
>>101374530just dont do an --reinstall-xformers or --reinstall-torch right now, my performance dropped to rock bottom with the torch cuda12.1 release, the cuda11.8 release runs perfectly
this thread would be more fun if people posted spicy/cringe/political shit. all im seeing is the same pedo anime slop from the same people
>>101374562That's why /ldg/ was made. /sdg/ basically evolved into a social club for the quokka and purple wizard to jerk each other off in.
SD3 beaten by 1 person:https://huggingface.co/fal/AuraFlow
>>101374562I do that sometimes
>>101374574Nobody believes me when I scream about SAI misappropriating their money. They should have been able to produce SD3 at a fraction of the price they made it for yet somehow hemorrhage 8 million dollars a month.
>>101374360>>101374499The thing that gets to me is, why do I also have issues after SD is already closed?
>>101374635Look at all their other projects. StableLM, stableaudio... They are just really shit at this.
>>101374645>Look at all their other projects. StableLM, stableaudioIt's weird how I subconsciously trained myself to ignore anything they put out besides stable diffusion because I knew all they ever produced was shit, but it only really clicked with me that their image gen was also shit after SD3.I'm glad we seem to be experiencing a glut of competent base models coming out right now. I guess we will see what cream rises to the top in a month or so.
>>101374641if its a memory leak, it could have written on place in vram or ram that the OS currently doesnt need, but eventually does and then there is garbage and you get an BSOD, thats why memory leaks are such a pain, and especially VRAM memory leaks are very hard to spot and CUDA is an absolute memory leak monster, its basically C++ running on your GPU, my MSI Nvidia GPU has a tool that "cleans" VRAM .. I tend to use it after closing a1111 and it always says something like ~15GB of VRAM has been freed..check if you can find a tool like that for your GPU
I'm not sold on AuraFlow yet guys, but the unlimited potential has me intrigued.
>>101374574meh
>>101374699look at that juke
>>101374716Using Auraflow I can say its prompt adherence beats anything else out there. Pixart, dalle, sd3...
>>101374745Its prompt adherence is pretty fucking amazing, which is why I'm giving it a pass on having kinda shitty anatomy right now.
>>101374674Well despite not having touched SD at all today, I got a kmode BSOD a few hours later.Maybe the problem lies elsewhere
more booba
auraflow didn't score very well on the debo test
>>101374478A couple times I left the A1111 open and when I came back the next morning had a bsod. It's only even happened when I left it open over night. I know you said it happened to you other times when you didn't also, but I think leaving it open can cause your the pc to crash.
>>101375007it's cute
>>101375059I dont want koff to think I'm stealing his style
>>101375007you have an uncanny ability to make even the best models look like shit. Not so sure the 'debo test' holds water
>>101375096you may have a pre-existing bias
>>101375096he held on to sd3 for so long. just do the opposite of what he does kek
ALRIGHTspoonfeed me slightlyi havent played with SD in monthsi never liked comfy, im not unwilling to try it but..how do i update SD again? exactly? i remember i have to open the .bat file and type somethingalso whats the new hot models that work, is sdxl good now, uncensored etc? Like are there good new base models that I can mix together?THANKS have an oldie
it def doesn't get how the eyes are supposed to work>>101375125you should just get the node manager for comfy, it makes life a lot easier. you can update by just clicking the update button
>>101375125Looking over civitai it looks like pony is the new thingUses sd2.1 though, i thought that was shit? or was I thinking of something else?Should I just do a fuckin fresh install
Towards the right is a cartoon dragon on top of a cliff, to the left is a anthromorphic fox wearing armor riding a horse. The horse is standing on top of a blue cube. In the background there is a flying eagle holding a sun. The sun has a angry face on it.Auraflow prompt adherence example.
>>101375125a1111? you type "git pull" in command line in the directory you have it installed, then you pray, also SDXL is excellent these days, but you need 12gb vram or you gonna crawl to your grave before you get a gen
>>101375148thanks i knew it was something like thati got the vram for it, what is sd:next and why shouldnt i be using it over a1111
>>101375170SD.next is a fork of a1111 works very similar, was unstable for me, so I stayed with a1111 but it has a more streamlined interface, could possibly replace it eventually if a1111 stagnates.. also if you wanna use SDXL you wanna go to civitai.com and get a model for your tastes, dont use the base model, its still crap
>>101375191>>101375191>>101375191
>>101375197guess ill filling the thread up with hooters for you mr. premature ejaculation
>>101375198
Filling
>>101375218What is thisa breakfast for ants
>>101375251Welcome newfag