Previous /sdg/ thread : >>101658301>Beginner UI local installEasyDiffusion: https://easydiffusion.github.ioFooocus: https://github.com/lllyasviel/fooocusMetastable: https://metastable.studio>Local installAutomatic1111: https://github.com/automatic1111/stable-diffusion-webuiComfyUI: https://github.com/comfyanonymous/ComfyUISD.Next: https://github.com/vladmandic/automaticAMD 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>Run cloud hosted instancehttps://rentry.org/sdg-link#run-cloud-hosted-instance>SD3 info & downloadhttps://rentry.org/sdg-link#sd3>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://aitracker.arthttps://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
Trani and his followers are incredibly bad people.
>mfw Resource news07/31/2024>Bubble Prompter Extension for Stable Diffusion WebUIhttps://github.com/captainzero93/sd-webui-bubble-prompter>Waifu Diffusion V public testshttps://huggingface.co/waifu-diffusion/wdv-tests>Comfy-Org / ComfyUI_frontend v1.2.7https://github.com/Comfy-Org/ComfyUI_frontend/releases/tag/v1.2.707/30/2024>Canva Acquires Leonardo.Aihttps://www.maginative.com/article/canva-acquires-leonardo-ai>Improving 2D Feature Representations by 3D-Aware Fine-Tuninghttps://ywyue.github.io/FiT3D>ComfyUI-segment-anything-2https://github.com/kijai/ComfyUI-segment-anything-2>SAM 2: Segment Anything in Images and Videoshttps://github.com/facebookresearch/segment-anything-2>ImagiNet: A Multi-Content Dataset for Generalizable Synthetic Image Detection via Contrastive Learninghttps://github.com/delyan-boychev/imaginet>AuraSR-v2: GAN-based Super-Resolutionhttps://huggingface.co/fal/AuraSR-v2>Hugging Face Offers Developers Inference-as-a-Service Powered by NVIDIA NIMhttps://blogs.nvidia.com/blog/hugging-face-inference-nim-microservices-dgx-cloud>Access to NVIDIA NIM Now Available Free to Developer Program Membershttps://developer.nvidia.com/blog/access-to-nvidia-nim-now-available-free-to-developer-program-members>Library Dataset: Image-text matching for large-scale book collectionshttps://github.com/llabres/library-dataset>Garment Animation NeRF with Color Editinghttps://github.com/wrk226/GarmentAnimationNeRF>InternVL2-26B-AWQ: 4bit quantized VLMhttps://huggingface.co/OpenGVLab/InternVL2-26B-AWQ>Efficient Face Super-Resolution via Wavelet-based Feature Enhancement Network https://github.com/PRIS-CV/WFEN>CatVTON: Concatenation Is All You Need for Virtual Try-Onhttps://github.com/Zheng-Chong/CatVTON>Memory-efficient Diffusion Transformers with Quanto and Diffusers https://huggingface.co/blog/quanto-diffusers>Meta's AI safety system defeated by the space barhttps://www.theregister.com/2024/07/29/meta_ai_safety
>mfw Research news07/31/2024>Matting by Generation Iconhttps://lightchaserx.github.io/matting-by-generation/>Add-SD: Rational Generation without Manual Referencehttps://arxiv.org/abs/2407.21016>GABInsight: Exploring Gender-Activity Binding Bias in Vision-Language Modelshttps://arxiv.org/abs/2407.21001>MMTrail: A Multimodal Trailer Video Dataset with Language and Music Descriptionshttps://arxiv.org/abs/2407.20962>UniProcessor: A Text-induced Unified Low-level Image Processorhttps://arxiv.org/abs/2407.20928>What is YOLOv5: A deep look into the internal features of the popular object detectorhttps://arxiv.org/abs/2407.20892>Bayesian Low-Rank LeArning (Bella): A Practical Approach to Bayesian Neural Networkshttps://arxiv.org/abs/2407.20891>Retinex-Diffusion: On Controlling Illumination Conditions in Diffusion Models via Retinex Theoryhttps://arxiv.org/abs/2407.20785>Inverse Problems with Diffusion Models: A MAP Estimation Perspectivehttps://arxiv.org/abs/2407.20784>SceneTeller: Language-to-3D Scene Generationhttps://sceneteller.github.io/>Machine Unlearning in Generative AI: A Surveyhttps://arxiv.org/abs/2407.20516>Learning Feature-Preserving Portrait Editing from Generated Pairshttps://arxiv.org/abs/2407.20455>Contrasting Deepfakes Diffusion via Contrastive Learning and Global-Local Similaritieshttps://arxiv.org/abs/2407.20337>PIP: Prototypes-Injected Prompt for Federated Class Incremental Learninghttps://arxiv.org/abs/2407.20705>Interpreting and Mitigating Hallucination in MLLMs through Multi-agent Debatehttps://arxiv.org/abs/2407.20505>Evolver: Chain-of-Evolution Prompting to Boost Large Multimodal Models for Hateful Meme Detectionhttps://arxiv.org/abs/2407.21004>PIXELMOD: Improving Soft Moderation of Visual Misleading Information on Twitterhttps://arxiv.org/abs/2407.20987>Vulnerabilities in AI-generated Image Detection: The Challenge of Adversarial Attackshttps://arxiv.org/abs/2407.20836
Will StabilityAI be able to save Stable Diffusion 3?also first for (tr)ani is literal human garbage
>>101664246>>101664265he doesn't even post here anymore. at what point do you accept victory and stop this ritual posting?
>>101664238>didn't know if you had any new models or loras you were playing withHaven't used anything than breakdomain in a while, BreakDomain XXXL 3 when
>>101664295Cute purple witch, anon! :]
>>101664286Crawl back into your dark corner. I command you, satan. Power of Christ compels you.
Now's a good time to ask, what's /ldg/ and why is it a separate thread? Don't tell me it's because of that discord this OP has
>>101664344Jak made /ldg/ because he didn't get enough attention here. Coincidentally he has returned back here.His discord minions are also a factor.
>>101664344it depends on who you ask and you'll never get a straight answer. it was a product of the discord cabal though
Trying duos again and changed my proompt up, getting a bit closer!
>>101664344Anons were fed up with the rulebreaking, avatarfaggotry and discordtroonsSchizoanon tried to fix /sdg/ but failed so far
how do I hiresfix on pony models?I can't get it to work and not sure which upscaler to use. Should I just give up and img2img upscale?
There is an easy way for that discord channel. Report with screenshots has been forwarded to Discord's enforcement team. Expect a reply in few days, probably next week.
>>101664499Img2img with sd1.5 is probably the best way. Pony is already fried and lobotomized, it's just luck it's able to do what it does now.
>>101664502you can't destroy the discord. its a powerful containment chamber and must be preserved at all costs
>>101664499What UI are you using? Also for upscaler I like using lollypop! It's always worked really well :]
>>101664499get 4x_NMKD-Siax_200k .. works fine for me, pic related is a direct 3x upscale of a 1280x720 gen, 0.36 noise then redone at 4k runalso pray for peace!
>>101664560>>101664535>>101664586I used to use forge and just updated to reforge.Should I not just use the same base model to upscale?I'm specifically talking about whether to use latent,R-ESGRAN or whatever,since there are many different ones and ones seemingly for anime styles.While img2img works pretty well just reusing the same checkpoint
>>101664286Why would you lie about something easily verifiable? >>101652936>>101647187
>>101664606one post?
>>101664626one reply?
>>101664604same model should be fine, img2img also does a upscale run with one of the selected, it basically works exactly the same as hires fix .. R-ERSGAN is err.. well limited in many cases ge whatt that anone said >>101664586>4x_NMKD-Siax_200kits pretty good, alternativly 4x_foolhardy_Remacri or 4x-AnimeSharp
>>101664604Just try all of them and you'll get a preference. It takes time, you are creating images but you'll learn what fits your own set of goals.
boot program for ai [remix] is such a fire track, holyy fuckhttps://suno.com/song/7d074682-35e2-4bba-85da-829c1b5f4b85
>>101664701One day Sharon Osbourne will call you and ask "are you Debo?" and then...
>>101664733nta debo anon you are talking about .. but wtf is wrong with your obsession with them anon?
Any kind anon cares to share their wildcards collection?
>>101664747Please rephrase and post again, I don't understand zoomerish.
>>101664784sure thats the one I got somewhere from huggingface>https://files.catbox.moe/v9hakg.zipalso pray for peace!
>>101664784Here's mine>{some brain fart | another thing | third one},Try artist names for example.
python dreams, making servers scream
>>101664842nice poetry anon, actually quite inspiringalso pray for peace!
>>101664813Thanks>>101664815Already have lots of artists.txt
>>101664885>Thanksyw, just dont use all that silly camera wildcards in there .. the guy who made that seemed to obsessed to list every single photo camera in existence in that wildcard list, the rest is very useablealso pray for peace!
>>101664139wake me up when there is some new breakthrough for local shit
>>101664865>>101664910Fuck off back to your d-cord please.
>>101664926I am not on any discord.. I am just helping anons. Whats wrong with you?
well I guess someone just wanted attention.. anywho, praying for peace is pointless now .. the city is burning
>>101664942Sorry, I was molested when I was a child.
Is it just me, or did photorealism take a step back with XL? It's better at hands and at prompt adherence, but it is incredibly grainy and the overall quality of image is much lower compared to 1.5.
>>101665072Sad, I'll pray for you. You might not wanna hang on 4chan in that case, it's not a mental health friendly place.>>101665114The finetunes are definitely better at it than any SD15. I'd say SD15 and XL core are both crap, but SDXL finetunes > SD15 finetunes. You have to adapt your prompting tho.
>>101665114Just because you are genning the sameface Instagram portrait with some retarded prompt doesn't mean sdxl is worse. It does what you want to do with it.
>>101665139you say that yet there are never ever any photorealistic posts that manage to clear the bar of mediocre, let alone good. I certainly hope you weren't attempting to pass that image of as photorealism.>>101665151You could just say, "I am wrong but I have an emotional attachment due to estrogen supplements so I need to create a insane delusion to cope rather than deal with reality" instead of wasting your time typing that textual diarrhea
>>101665185Jak, go back to your dark corner.
>>101665185>you say that yet there are never ever any photorealistic posts that manage to clear the bar of mediocre, let alone good. I certainly hope you weren't attempting to pass that image of as photorealism.I am not aiming at photorealism with these .. target is naive depiction of angels in half 2d/3d
>>101665207The thing about schizophrenics is you always assume everyone knows about their delusion phantoms
>>101664865yeah its a nice line. suno is fun desu. here's some of my favourites that ive generatedhttps://suno.com/song/8f195193-b4c1-4806-b583-8370aaa71531https://suno.com/song/ccc19fcc-34f7-474f-8c6c-9c8f13659008https://suno.com/song/9b7811b7-a615-439f-ad1f-bbae0ec7d42a
>>101665221>target is naive depiction of angels in half 2d/3dI'm sorry, but you have no business proffering an opinion on quality if you are actively trying to achieve 2.5d.
>>101665238cant vibe with the first two at all, second barely passes, overall AI music leaves me feeling cold .. lyrics are fun tho>>101665251I don't care.
>>101665267I can tell
>>101665231that's more autistics. they lack theory of mind so they think everyone knows everything they know>>101665267they're not really what i wanted desu. needs more data. it's still impressive though from a technological standpoint and very fun
fuck adetailer hands,every time I've tried it it just makes it even worse.It seems like it's completely pointless to try and improve hand anatomy wtih img2img, you just need to get good RNG at txt2img and hope for the best.
>>101665267I believe in you anon! You are too good for these pigs.
>>101665316Schizos do to. You go to /x/ and you'll see them posting the most insane shit as if it's completely obvious and everyone one already knows that the benjoran's have a peace treaty with the reptoids to avoid conflict in the Arizona tobacco market during the month of october because that's when bigfoots are the most powerful without explaining any of that shit.
>>101665331inpainting can fix hands. Adetailer is bad at fixing them because it doesn't give rerolls
>>101665347Thank you. But I really don't care about some random anons opinion on an image board I post some pictures for fun for a bit. Even less if they think they are art critics lmao.
>>101665353Nah, these folkloric things stem from the fact humanity was seeded on this planet and the creators are still observing this mess. Most likely some gorm of biological automatons but still. But it's good, just keep fighting over politics and arguing about some archeological nonsense...
>>101665408Thats why He reasonably often wipes His creation .
>>101665383You do care, though. You seek approval by posting gens publicly. You are just a bitch about criticism and double down on being bad. If you had actual confidence you wouldn't bother addressing someone denigrating the style you use. Have some balls, man. If me saying I dislike the style in general shakes you that much, develop the masculinity to brush aside a pointless criticism rather than dwelling on it.
>>101665443Yeah, last one was about 12k years ago.
>>101665448>talking about estrogen>multiple mentions of masculinity and ballsAre you in denial? Such an aggressive stance.
>>101665465Don't bother in my decades on 4chan I met them type so often. It is not worth any attention.
>>101665465Perhaps you should turn that focus inward and consider why you hyperfixated on those specific insults>>101665484this was the right stance to take five posts ago when it doesn't come off hollow
>>101665502Very funny post.
>>101665517likewise
>>101665502You could just say, "I am wrong but I have an emotional attachment due to estrogen supplements so I need to create a insane delusion to cope rather than deal with reality" instead of wasting your time typing that textual diarrhea
>>101665530Look, that was pathetic, but I feal this conversation is a dead end so I'm going to let it go. Don't let people like me get under your skin because I don't like the style you use. It's pointless to argue that. Have some confidence in your work.
>>101665625like your pixel art, would so have played any game with that kinda art in my youth
>>101665657This is genuinely the best gen I have seen in few days.
https://civitai.com/images/9850532lmao this prompt>body status Bar Sprite in form of woman
>>101665674Aww shucks :3
>>101665573Empire States building in the background! kino
I finally got around to downloading Pony and it's all cartoony and can't do photoreal at all. Pic unrelated from 1.5.
>>101665976try ponyRealism or Godiva, those are photo realistic pony derivates, 2dnPony for 2d/3d, and autismMix for better anime .. core pony is pretty pointless
>>101664139That's nice; needs a cat
>>101665340goddamnis that dithering post-processed?
>>101666092AI is fantastic with pixel art when done correctly.
Is the CUDA toolkit necessary if I only generate images with SD? Would it boost performance?I can generate images just fine, but this message is kinda bothering me.
>>101666342Torch etc already uses cuda and that's all what you'll need. No need to install anything new.Use --cuda-malloc --cuda-stream as launch parameters.
>>101666390Got it, thank you!
I cant tell whether something went wrong or right
new products and cleaned it up
acquiring dreamsone byte at a time
>>101666794very cool background, could turn it green like the matrix
>>101666682gucci lora? I want it .. or what am I am looking at?
>>101666682Gucci Gang! I can finally construct my autumn wear because of you.
>>101666820It is big data and he just wants a like. Not trolling. This could be used for lora but my own gpu is too tiny.
>>101666844sadface .. databases are ancient, nothing to brag about, I want a lora of that!
I think trani is shit
>>101666923I think he is a pretty cool dude because I'm not a depressed troglodyte
>t. d*bo
>>101666923flerpa is canon proper dragonborn
>>101666879I would work on one because I'm a big fan of Chanel and some designers. I love Karl Lagerfeld, because he was truly autistic but in a cool way.
emotional bytes, encrypted in sighsa database of tears when the day is done>>101666811thanks!>>101666820>>101666879why cant i just do the data acquisition. training is not supposed to be my department as well but fine ill add it to my list>>101666826gucci gang gucci gang gucci gang >>101666844i'd really appreciate the likes
>>101666988cool, how do you get the numbers?
>>101666988you surely can do! good job anon, I just was sad it was no lora>>101666811>>101666988>gucci gang gucci gang gucci gangpic related
>>101666988I like you! Aren't you concerned about huggingface deleting the entries btw? Ultimately this doesn't affect you of course but as a hosting place?
>>101667018he tells dalle to add vertical floating binary numbers
>>101667006>>101667059
>>101667068>>101667059
>>101667070>>101667068>>101667017>>101666988
i really like this one desuhttps://suno.com/song/26817194-cf0f-4d03-82a9-c81c863f3a1a>>101666987i'm a big fan of jean paul gaultier and vivienne westwood desu>>101667006>dreamy space background, binary, 1s and 0s, green and black>>101667017have you tried any of the clothing adapter models? that's what i had in mind for these fashion sets. the models ive seen are all trained on the same data, viton-hd and another one>>101667018not really, basically everything on hf is copyright>>101667059not my fault it's the best model
>>101667089>not my fault it's the best modelit's good, niji has its moments tho
>>101667089Did you watch any documentaries about her? She met her soul mate bf at some point and while the guy seemed fairly gay on the outside I do think they had some deep connection.
>>101667089>have you tried any of the clothing adapter models? that's what i had in mind for these fashion sets. the models ive seen are all trained on the same data, viton-hd and another oneno I am just RNGing with loras in stable-diffusion .. kinda like it <3
>>101666820>>101666794>>101666436>>101666481Who Dat?
>>101667231Who you?
>>101667240Sensei LA Dew
>>101667240They call me the Proctor.
>>101667301and your buddy Gamble.
>>101667231Bunch of OCs
I have forwarded the proof of trans hate to Discord's moderation team. Basically I'm a Lady Boy from Finland and 6wUwtcJsr2 staff was very condescending and didn't want to even try finding my posts.
>>101667127that's very sweet, ill check it out>>101667174i dont think they're integrated with any uis anyway
>>101667466Don't remember which one but supposedly a recent one.
>>101667472I actually found one for comfy>https://github.com/cozymantis/clothes-swap-salvton-comfyui-workflowbut I am uncomfy with comfy
>>101667487Looks fairly clean but I do think it could be simplified more. It's using ipadapter. Don't start straight from this if you switch software.
>>101667487hadn't seen that model, swapping clothing is interesting too. looks like its a paid node though which is lame af, its a free modelthis model takes an image of clothing as conditioninghttps://github.com/ShineChen1024/MagicClothing
>>101667534that looks pretty cool, need to look into it
>ponyv6 in comfyui>cfg 13>euler>normal>desired tag weighted at 2>still only get it half the timeWhat the FUCK, what am I doing wrong? It's just a headpiece, it makes them very high quality when it does too so I know it knows how to make it. Is CFG 13 too low or something?
>>101667585what kinda headpiece? also cmon its pony, its trained to do horse pussy not headpieces
>>101666436>>101667231my wifeyou mirin bro?
>>101667585Cfg will not force anything. You should keep it at some nice 6 to 7 or so.Only way to force subjects is via repeating similar tokens in the prompt and playing around with weights. Also keep in mind these models are pretty braindead anyway. Less = more (until it's not).
Finally getting somewhere with backgrounds and fine details.
>>101667472bing/dalle immediately blocks any prompt with vivienne westwood which is annoying. i wanted the orb. other designer names work>>101667564there's no training code though. should be easy enough to make, the clothing adapter is a second trainable unet, the base is frozen during training, i think its going to use masked loss. in the end its using the hidden states inputed to the attentions from the cloth adapter unet concat to the base model's hidden states, there's an extra input for the latent of the clothing and an extra part for cfg calculation
>>101666092yesI use SD to generate the image and an extension to make it into pixelart using another neural networkthe extension is only for A1111: https://github.com/luhovei/stable-diffusion-webui-pixelization
>>101667752waah.. to much for me, maybe some smart person will make it eventually
very demoralised by the recent releases of mj 6.1 and gen 3opengen is so far behind that the gap seems unbridgeablesd gens have defined the aesthetics of povertyit's good for coomer stuff and nothing else
>>101667752>Glivenchy
>>101667848>mj 6.1 and gen 3I didn't see anything of note really, just FOTM models then everyone just uses xl after they are bored
>>101667848https://blackforestlabs.ai/announcing-black-forest-labs/good news for you then
>>101668020it's 12b, can't run it. this is also from the mind that gave us sd3 medium so I don't really care
>>101667733187.3% improvement on the background, this is a solid summer muppet. good one!>>101667924better, the other one with the blue hair was weird>>101667848so demoralized, tech available (for free) that was completely unthinkable of 10 years ago, sniff, sob
>>101668020I'd apply if I wasn't senile.
>>101668028>Those grapes were sour anyways!
>>101668033You're a cute muppet.
>>101667798maybe ill do it myself desu>>101667873balenciaga did this, its meta af i love ithttps://www.balenciaga.com/en-us/ai-generated-hoodie-medium-fit-black-white-767877TQVF99034.html>>101668020>We are hiring exceptionally strong machine learning and backend engineers.>not hiring for dataDOA
>>101667734this is nice>>101668033thanks
>>101668050Exceptionally bold.
if i drop that data you gotta like ithttps://suno.com/song/4dd2d1c4-9210-4fc6-b822-c87c0e4a0948
>>101668004it's over, you're in denial>>101668020looking forward to everyone running the commercially locked down 'open' model on their $4000 RTX 5000's>>101668033best stuff is getting permanently paywalledbig vram on consumer cards is never comingyou will pay for access and you will be happy
>>101667059this is one hell of a gen dude,don't let anyone tell you otherwise lol ! did you ipadapter old japanese robots?>>101668050THE SHOES lol<<img limit reached/nogen>>
>>101668106videos like this suck. make something interesting.
>>101668106porn vids please
>>101668117how about you generate a temporally coherent video at all>>101668138>>>/gif/
>>101668100I already liked you.
>>101668149>how about you generate a temporally coherent video at allhow about you? even these videos have problems>>>/gif/where is the thread for ai porn? it's just trannies
>>101668020awesomeanother DoA model trained on synthetic dataI looooove that glossy, plasticy lookyum yum
>>101668106then so be it. I've had a good run. too much of a punk to pay for it. sorry. also, happy with my "used" 3090 which I got for 700 bucks. this is a beautiful animation btw, did you make it? you know how high you gotta be for your pupils to close like that? whee.
>>101668060thank you
>>101668154>even these videos have problemsminor issues, solved by gen 4.how're those animatediffs going though?>>101668154>ai pornnot an interesting use case>>101668172>did you make itof course not, i have better things to do than to making this sort of slop
Have small models advanced in any meaningful manner since the beginning of the year? Base SD was pretty already on a 1060 but SDXL takes super long per gen, or at least it used to.Do we have any better alternatives now?
>>101668114the shoes are sickgonna scrape balenciaga. i think clothing adapter training needs model + product images rather than just product>>101668151like the datasets pls>>101668164rated by elo score, there was some guy at sai who wanted that kind of rating, like an arbitrary scale of yes/no ratings, but imo that only works for chess and tinder because the other side is playing too and like there's more to whether an image is good than yes/noit's also too easy to cheat, assuming they didnt just make it up and actually got some number of people to rate the images they'll have had to cherry pick images to use in the first place either manually or using one of the real rating models
>>101668225oh all high and mighty, well then fuck off.
>>101668225>minor issues, solved by gen 4the issue here is you make boring slop. do better>not an interesting use caseit's 90% of imagen. if it's not interesting then get the fuck out of the community>>did you make it>of course notholy cope, go get a sub and come back with your own shit
>>101668263just here to remind everyone just how far behind the SOTA these threads are>>101668265>90% of imagenyou should spend more time outside coomer circles>come back with your own shitsorry, my stuff is NDA'd
I made a booru tag selectoreasy to copy the prompt, has multiple tabs for pos/neg and saves everything in cookiesI made it for myself but I thought I'd share anywayshttps://boorutags.netlify.app
>>101668297>ust here to remind everyone just how far behind the SOTA these threads areSOTA is universally shit with botched big datasets and retarded captioning methods. not to mention they are getting in shit for scraping copyrighted material AGAIN. DiT is even obsolete so I don't really care for these models anymore. fuck this gay earth>sorry, my stuff is NDA'dani at least shows what the fuck is happening behind closed doors.
>>101668339>DiT is even obsoleteOh? What's new, then?
>>101668297>sorry, my stuff is NDA'dthat's a funny way of saying your outputs are shit
>>101668371https://arxiv.org/abs/2407.04620
Black Forest Labs just came out of nowhere and open-sourced a 12B parameter rectified flow transformer, there are some big names on their team. Thoughts? Is it trash or are we back? I haven't used it yet and you need to log in to gen, but some of the examples are looking pretty nice.https://blackforestlabs.ai/announcing-black-forest-labs/
>>101668247Anyone? :(
Anyone able to replicate the olympic surfer's amazing photo only with prompting?
>>101668395Here are some of the examples that were posted, it can do much longer text as well
>>101668407you can scourge civitai for some good civitai for some good sd15 models, but it is where it is, most finetuning efforts have moved to sdxl, sorry it will not much iimprove with a 1060
>>101668247Most people are not used to rendering anything so therefore any image which takes more than 30 seconds is>omg my brain this is superrrrr slow i can't breatheAnything under 5 minutes is acceptable especially for NON PRODUCTION WORK. Use TCD lora, optimize your shit.
>>101668458kek.. I am used to an image downloading in 5 minutes back in the day, you young ones are way to comfy in your hyperspeed internet
>>101668407upgrade your ish. a used 2080, anything better than a 1060. get a used 3090. that is what I did. if you like image generation you gotta invest a little in hardware. I found sdxl unbearable with an rtx 2080. and no, small models have not advanced in any way. other things have tho. the little things that make things go faster and shinier. PAG, hi diffusion, new samplers and schedulers, etc.
>>101668458Do you get paid per obnoxious post you make in this thread or something?>Most people are not used to renderingStupid argument, have YOU ever done any rendering work before? Do you keep rendering the final result every time you make any minor adjustment? So fucking stupid.>>101668442>>101668499Ah that's alright then, thanks anons.>buy a new rigThere's nothing else I do on my computer that requires a 3090 and I don't gen enough to warrant buying one.
I fucking love brainrot
>>101668544>Do you get paid per obnoxious postconsidering he is saying his own garbage is nda he is probably a runway shill
>>101668407>>101668458right, actual speed increase through lighting/hyper/LCM tech, I forgot to mention that. oops.>>101668454HELLO>>101668497(brutal grey fog:2)(hazy:1.7)(mist:3.1) nice gen, somewhere under that fog
>>101668544>posting Maya screenshot with a random bs rigBut it's quite obvious you are not a professional user at all. I am not going to outline this here because you are quite too stupid for this.
>>101668576Random picture I picked off google in order to illustrate my point jackass, now answer my question, do you get paid per obnoxious post or is that the only way you have left to feel anything resembling a human emotion?
>>101668563don't confuse me, a houdinigod, with a mayalet.
>>101668544Yes, I would submit prman renders and later on mantra to the farm. Most layers would get rendered overnight.
>>101668608I know you are stupid, your aggressive tone explains everything.
>>101668610Houdini = Heaven A coincidence? I don't think so.
>>101668651give him slack, his xanax probably ran out
>>101668676Opioid or bentzo withdrawal.. This is why they are human trash.
>>101668317github?
>>101668913>hot windowsheh
the shoes hahaha
>>101668949Give it a few years and the fashion dictators will tell everyone this is trendy *must buy, your friends will be soooo jealous* fashion.
>>101668949That's Balenciaga!
>>101668802>>101668943nice muppets, first one tho, her left hand. if you haven't tried it yet, there is a powerful clean-up tool called "lama cleaner". can try it online here: https://huggingface.co/spaces/Sanster/Lama-Cleaner-lamaor, if you like it (and its a great tool for the removal of unwanted things, so I really recommend it), you can also run it locally. could also just paint over the hand in the bright cyan from the dress and resample.>>101668949lol
>>101668996this is so nice, I think I am just too damn rotten to gen something like this. is that autism/confetti/dpo mix? love it>>101669016balenciaga creepy af
>>101669090It's NovelAI, sadly. I've been waiting for local models to finally catch up for months now, I want to ditch this.
>>101669053You're a real digital muppet
Can't find an up to date tutorial on training a Lora with Automatic1111 and SDXL, is there an easier way.Ip-Adapter is kinda meh, wasting a lot of time tuning the parameters
>>101669164Switch to cumrag, it's as efficient as the python interpreter goes.
>>101669111ugh, you could start genning "realism", thats what I do. can't complain. what do they want from you? 25/month? it is uncensored too, right?
Is this shit heavy on your machines?Is the processing needed for this "hobby" gonna crash my shit out
>>101669239>cumragto what??
>>101669322Cumrag=discord group's insulting lingo for ComfyUI.I'm sorry you needed to even think about this.
>>101669322these guys are psychopaths.
only need another 50m or so and im at 1B generations
>>101669424I liked your data.>>101669350
does a1111 support flux?
>>101669482probably not, but comfy doeshttps://comfyanonymous.github.io/ComfyUI_examples/flux/
3D objects and animation (time series data) are the future because they're modular, consistent and can be moved and rotated. Calculation costs will be drastically reduced.However, the current video generation AI remains an extension/mix of 2D.They should focus on 3D object generation if they want to see the future.
>>101669518thanks i'll check it out
babe wake up, a 12b model has been releasedhttps://huggingface.co/black-forest-labs/FLUX.1-devhttps://blackforestlabs.ai/announcing-black-forest-labs/
>>101669464no you didn't :( why you gotta do me like that. smash that fkn like button
>>101669576Ehh honey you are bit late, SideFx did this with Prisms in the 80s and Houdini in 90s. Procedural graphics is nothing new. SideFx will have the world's best AI toolbox in few years but not immediately.
>>101669590how much vram
>>101669482>>101669518So i also got the email of Flux and tried the demo, logged in on my GitHub demo and after trying it for around 10 times I realized that the request took $0.025 and found the analytics tab which says I have 33s of billable time which translates to $0.07Do i really have to pay for trying the demo wtf?
>>101669590yeahyeah seen it, those images look dope (minus that robot writing on the backboard) also +1 for that company name.>>101669623money well spent huh
I'm used to Automatic1111 but it's been a while. What's SDNext all about?
The power of modular 3d objects (assets) is that they can be copied and shared as is. We will not be generating anything but a few original objects in the near future. Just search and download.
>>101669693fork with some new features and some less, mostly design overhaul.. it crashed on me all the time so I stick with a1111
>>101669648>how much vramhttps://huggingface.co/black-forest-labs/FLUX.1-dev/discussions/1#66ab9dc4fd4ae9a7c49be855>I have a 3090 with 24gb vram. But 12b parameters in float16 format are still ~24GB and this does not include the two text encoders nor the internal state of the model.A lot, a 12b model is giant anon, maybe it will be quantized well at 8bit, it's a DiT and transformers models can work well quantized
>>101669111>It's NovelAI, sadly. I've been waiting for local models to finally catch up for months now, I want to ditch this.NovelAI is unbeatable for sex scenes, but your example of a wife in a relatively neutral context is still a solved problem for common weeb models/mixes like basedv66 (don't use the higher numbers), anyv5, anylora, some ponys, orange.You just need to use a high step count if you do it on one step (like phase), 80 or more. Or you can a low step count (30 or something) for candidate images and then img2img them to something actually good.
>>101669660Nickel and dime, son. Sure you don't need to pay anything unless your card was used when making the initial account. Surely zoomers can't be this stupid?
>>101669697These are called point data.
>>101669770Gyatoo!
>>101669727>check out this new cutting edge model>can I run it>no
>>101669739I'll check out the models you've mentioned, thanks anon.What I've found is that NovelAI is pretty much unbeatable when it comes to knowledge of characters and artstyles. Unless one of these models represents a breakthrough and isn't just some merge, I doubt it will be able to match it.
Now that forge is relegated to experimental bullshit and ruined my mobile goon sessions I think I will finally try out the other front ends. I know A1111 is serviceable on mobile, SD next seems to have mobile compatibility as well. ComfyUI doesn't seem to be very mobile friendly unless I'm wrong?
>>101669590https://github.com/black-forest-labs/fluxWtf the picture you're seeing is from the "schnell" model, that looks so good!
>trying to generate only one naked person on a white background>succeeded and change the size>grows creepy objects faaaaaccckckkckk
>Segment Anything 2: New SOTA on segmentation, by combining synthetic data with human feedbackanti-synthetic schizo on suicide watchdata and reinforcement is all you need
>>101669852N-nooo! But muh moat! Muh OpenAI empire!!!
>>101669813ehh wipe your forge and install ReForge (don't forget to add the path to your python exe in the user-bat file before running it 1st time). that is what I did, smooth operation. don't switch to a1111, its just not fast enough.
>>101669852>It's SOTA it means it's the best method and that only human feedback would've never reached that levelImagine saying that unironically
>>101669865Did you patch ReForge for torch zero vulnerability?
>>101669590>>101668789 >>101668964 >>101669042>>101669149 This is insane!
>>101669766where is the 3D booru of my ideal Hatsune Miku?I believe I need >10000 different free models of her to search/get it.
>>101669883can you try saying that again in english
>>101669900>too retarded to understand englishNot surprising
>>101669833its actually that good, just tried it
>>101669929>just tried itlocally? it works on a 24gb vram card?
>>101669940still downloading the weights for local useyou can try it for free on replicate and falhttps://replicate.com/black-forest-labs/flux-dev
>>101669899Means are there. Autodesk has implemented some rudimentary topology generation model but I haven't followed Maya in many years.
>>101669833Wow. SD3 BTFO.
>>101669927>NO local should NOT do the same as every legit ai company
>>101669968Sorry, but I'm a post-AI human, so I only consider anything that takes only a few steps and a few minutes to generate to be a tool.
Morning anons,got the news for Flux on my email today, i wasn't disappointed when I found the Hugginface demo
I saw this paper's title and thought that a "potato tuber" was some kind of vtuber at first. my brain is brokenHigh-throughput 3D shape completion of potato tubers on a harvesterhttps://arxiv.org/abs/2407.21341>>101670056gm
flux can do anime
>>101670072Does it know Touhou characters?
>>101669929>>101670072>>101669833What the hell...are we back?
>>101670072Neat
>>101670072is this the released model or the api only version
>>101670111https://huggingface.co/black-forest-labs/FLUX.1-dev
>>101670041I understand and I apologize.
>>101670093outputs looking good so far. hoping it wont be too hard to finetune>>101670111its flux dev from >>101669956
>>101670094Bober kurwa
>>101668033I also thought it was weird. Eventually found out I accidentally had CFG set at 11
No way
>>101670143what's the resource usage like
>>101670085>>101670085>touhou, remilia scarlet, 1girl, blue hair, demon wings, dress, fingernails, hair between eyes, hat, looking at viewer, mob cap, puffy short sleeves, puffy sleeves, red eyes, short hair, short sleeves, simple background, smile, solo, teeth, upper teeth only, wavy hair, white background, white dress, wings, wrist cuffs
Really loving the flux interpretation of the crayon style
>>101670188Nice. Seriously.Does it know the less popular ones as well? What about Kyouko (Kasodani Kyouko)?Almost no model knows her.
>>101670153nice. now the hands. looks like autismmix. you on comfy?>>101670157TOP>>101670205wow
>>101670220I'm using auto1111>now the handsthe ever present problem.I dont know of a good way to fix them that doesn't involve actually knowing how to draw. I think that's when pure prompting hits a plateau.
>>101670209im not familiar with the characterjust try it at: https://replicate.com/black-forest-labs/flux-devhttps://replicate.com/black-forest-labs/flux-pro
>>101669590>https://blackforestlabs.ai/announcing-black-forest-labs/nice, now buy my an commercial RTX .. damnit my 4090 is worthless
>>101670252It, uh, doesn't. That's a shame.
>>101670249yeah on comfy you have a strong option called "mesh graphormer", automatically draws a depth map of the hands (a correct one) and makes inpainting a whole lot easier. could try different samplers like the smea dy one. or install comfy and use it to inpaint hands lol. can just point it to your models folder(s), not a big installation otherwise.
Mechs are back!
>>101670341Baphomet told me not to.
>>101670341well, there are similar tools for inpainting hands in auto1111, but I never managed to properly use them.
Enough with the amonguses, gonna try to make something not retarded this time
>artistic nude photographyfluxhttps://files.catbox.moe/x7vehl.jpg
>>101670375I know, been there, tried it, meh. its worth installing comfy just for the mesh stuff, trust me. muahaha (seriously). btw why not forge/reforge? a1111 not cutting it speed wise. switching from a1111 to reforge is easy peasy. kinda.
>>101670391flux devhttps://files.catbox.moe/ooq10r.jpg
well even 12b parameters doesnt seem to solve holding rifles for anime style
>>101670391>>101670424Pretty good for a base model.
>>101670391>>101670424thats decent.
>>101670391Would berry my dick so far in her ass
>>101670408It's easier to iterate with different settings without messing with spaghetti.I tried comfy once, it's interesting for creating tons of pictures in a line of production-like thing. But I dont create many of the same pictures so it's not really worth it for me.I use inpainting a lot, especially in the face for changing expressions, but hands continue to elude me.
>>101670391doesnt allow it for anime tho, prompt:>nude woman anime style
>>101670471Someone prompt that please. I'm at work.>>101670492Yummy tits.
>>101670503>Yummy tits.correct, but no nipples in Blackforest anime land>nude cyborg woman anime style
>>101670482>>101670482>>101670482
also cant tan lines or micro bikini>anime girl at the beach in the sun, shiny glistening skin with tan lines wearing a micro bikini
>>101670424>>101670391We are so back
Filling
Yay
>>101670424Torrent link pls?
>24gb vramaaaaaaaaaa
>>101669852>anti-synthetic schizo on suicide watchI didn't know seg was an imagen model
>>101671615right, it has to be posted in segmentation general
>>101671792I think you misunderstand what anti synth data anon is talking about so maybe sit in the corner. visual models != imagen models
