prev >>101349321>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 news07/10/2024>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>Minutes to Seconds: Speeded-up DDPM-based Image Inpainting with Coarse-to-Fine Samplinghttps://github.com/linghuyuhangyuan/M2S>GenArtist: Multimodal LLM as an Agent for Unified Image Generation and Editinghttps://zhenyuw16.github.io/GenArtist_page>AID-AppEAL: Automatic Image Dataset and Algorithm for Content Appeal Enhancement and Assessment Labelinghttps://github.com/SherryXTChen/AID-Appeal>Align and Merge Deep Learning Modelshttps://github.com/shoroi/align-n-merge>UltraEdit: Instruction-based Fine-Grained Image Editing at Scalehttps://ultra-editing.github.io07/08/2024>JeDi: Joint-image Diffusion Models for Finetuning-free Personalized T2I Generationhttps://research.nvidia.com/labs/dir/jedi>PartCraft: Crafting Creative Objects by Partshttps://github.com/kamwoh/partcraft
>mfw Research news07/10/2024>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 Image-to-Video Adaptation: An Object-centric Perspectivehttps://arxiv.org/abs/2407.06871>Beyond Aesthetics: Cultural Competence in Text-to-Image Modelshttps://arxiv.org/abs/2407.06863>Chat Edit 3D: Interactive 3D Scene Editing via Text Promptshttps://sk-fun.fun/CE3D/>Powerful and Flexible: Personalized Text-to-Image Generation via Reinforcement Learninghttps://arxiv.org/abs/2407.06642>Mobius: An High Efficient Spatial-Temporal Parallel Training Paradigm for T2V Generation Taskhttps://arxiv.org/abs/2407.06617>Iteratively Refined Image Reconstruction with Learned Attentive Regularizershttps://arxiv.org/abs/2407.06608>Vision language models are blindhttps://arxiv.org/abs/2407.06581>Sketch-Guided Scene Image Generationhttps://arxiv.org/abs/2407.06469>VIMI: Grounding Video Generation through Multi-modal Instructionhttps://arxiv.org/abs/2407.06304>A Survey on Mixture of Expertshttps://arxiv.org/abs/2407.06204>Poetry2Image: An Iterative Correction Framework for Images Generated from Chinese Classical Poetryhttps://arxiv.org/abs/2407.06196>More Distinctively Black and Feminine Faces Lead to Increased Stereotyping in Vision-Language Modelshttps://arxiv.org/abs/2407.06194>VideoEval: Comprehensive Benchmark Suite for Low-Cost Evaluation of Video Foundation Modelhttps://arxiv.org/abs/2407.06491>MiraData: A Large-Scale Video Dataset with Long Durations and Structured Captionshttps://arxiv.org/abs/2407.06358>EMBANet: A Flexible Efffcient Multi-branch Attention Networkhttps://arxiv.org/abs/2407.05418>Learning to Adapt Category Consistent Meta-Feature of CLIP for Few-Shot Classificationhttps://arxiv.org/abs/2407.05647
>>101360320imagine the plapping
>>101360368>plapping!
>>101360409is she an angel or is she ded or does she just have wings for fun
>>101360468chief angel in charge of the angel slime congregation! stressful job
"slime hat" is a good tag I haven't thought about before
>>101359723Kek'd!>>101359691>>101359844I might post later on but I will need to pay up to get the higher res version which makes it harder to get accused of being the poster. However, even on my own I can see way too many red flags in what she's produced, some stuff just gives that untrained AI generated feel but not all of it. I should've listened to my gut feeling and taken the small loss when it was barely commissioned now I'm stuck paying a lot more.>>101359973If I was making this stuff and getting paid for it, I'd increase the price if he wants it made faster. Since you find him annoying just increase the price by a larger amount. If he accepts you just got a big pay day if he follows through.
>>101360541>>101360476Nice tits, faggot.
So it seems like image models have come a long way compared to a year ago, but what about local voice models? I want to make my own audiobooks but don't want to pay eleven labs 20 a month.
>>101360600>I might post later on but I will need to pay up to get the higher res version which makes it harder to get accused of being the poster.but you just said you'll be one of the guys that paid for it and your post is sitting on numerous 4chan archivesYou should've provided less information about yourself and more about the artist
>>101360867lol thats so good. was this inpainted or did it come out like that?
>>101360890<3 just a happy accident.. just gotta love ancestral samplers
>>101360890>.png (2.81 MB>>101360867>.png (3.9 MBYou scumbags realize m.jpg doesn't work on blue boards, right?
>>101360909give all your bandwidth to me
I wish the backgrounds didn't suck ass on this model
>>101360909ow dear.. you still on dial up? I am so sorry for you!
>>101360822>>101360867imagine the plapping
Animagine is not good
>>101360782Fair point. It's simply enough time for it to spread online and then who knows maybe some "random anti AI art person" comes here to ask if it's AI made. Fuck I think I screwed up.
>>101360964!>>101360909I lovered the limit when JPGs will be saved to 1MB now, just for you, you happy now?
>>101361011thems fightin words
>>101361124Wizard
>>101361161
image to prompt is needed for each (every) model, right?
>>101361173img to prompt? what ya mean? CLIP test? its very inaccurate I would not rely on that at all
>>101361242
>>101361257
>>101361266
>>101361279
another doppelganger!
>>101360320plap>>101360409plap>>101360483plap>>101360550plap>>101360646plap>>101360822plap>>101360867plap>>101360905plap>>101360958plap>>101361032plap>>101361234plap>>101361361plap
>>101361394!!
>prompt (analog film:2.0)>gen looks like a crappy cgi render>shit reads "ANAL FILM"
>>101361483cute
>>101361535Thanks. I think it's getting confused by the chibi tag and also adding a chibi?
>>101361585very common for character art to have both a full render and chibi version so you're going to get that from the training as it will be taggedput solo if you haven't
>>101361637>put solo if you haven'tI thought maybe I hadn't, but I did
>>101360320ugh, crimvael is so cute, i wanna fuck him so bad bros
>>101361737I feel like devil slimes should be red or orange
>>101361749the masha lora turns any angel into her>>101361779you are right, I tried to prompt "red slime" just doesnt trigger in 10 tries.. orange slime does tho .. probably no red slimes in any source material (looks to much like blood?)
>>101361348This is jungle fevertesting stealth PAPA NOVEMBER GOLF info transmission on channel Alphaplease acknowledge
>>101361811fucking hnnnnng
>>101361818open up them pits gurl
>>101361827!! ! (also https://files.catbox.moe/981vpl.jpg is probably to lewd for blue board)
>>101361856damn I love silly
>>101361843
>>101361892what model is this?
>>101361922autismix-confetti, masha lora, kotoyama lora
thinking about a good source of images with transparency to try training a vae with transparency desurun rembg on images sure but the results arent always perfect which would probably affect the model
>>1013622592D? -> danbooru, gelbooru3D? good luck
Kolors is so good!
>can't recreate my own image with the exact same prompt and seedi fucking hate when this happens
>>101362326you using ancestral sampler? then its working like intended, each time you restart the webui/comfy it will get a new results.. if you want 100% consistency use a non-ancestral sampler (anything without the small a!) .. also old xformers were non-deterministic, but that should be fixed now
>>101362353I didn't even restart. This doesn't happen all the time, or even often. Most of the time when I recreate with the exact same settings, it comes out exactly the same. Maaaybe a few tiny pixels different, I assume due to some slight fluctuations in the parallel computations. But occasionally, rarely, I plug in the same values I just rolled and something completely different pops out- and any further attempts to regen that produce identical results to the latter attempts.
>>101362392what sampler and UI you using?
>>101362408Forge, Euler A SGMUniform.
>>101362392could be anything .. seems like a miracle this whole stuff works at all considering how many different software packages are stuck together to make it run .. also ya.. Euler A is ancestral, that has some probabilistic calculations in it, that will give different results, if you are up for some science trials run 100 Euler A and 100 Euler without A and check if they both behave erratic or just Euler A
>>101362303i found a few sites to scrape with random transparent images like effects, animals and people, i suppose the content and quality don't matter too much as long as the transparency is goodwondering how much variety/quantity training a vae needs too
>>101362437ooowww you mean like alpha channel transparency? I think this will be a tough one, not sure if that even can be done effectively, for SD thats just one other color and if you zoom in onto any "uniform" color of a gen (pic related from >>101362353 zoomed) you will notice its a blurry mess, having a whole area come out in exactly one color is kinda exactly how SD doesnt work .. but its worth a try! go for it!
>>101362478yeah it might not work because of that but worth a try, maybe it could be fixed in postprocessing like if the transparent area in the gen came out with values just off from fully transparent they could be rounded or something. i also remember seeing some paper or something about the color thing but not sure if its integrated into any uis or what it was called
>>101362535yea I would just try to prompt something like "black background" "simple background" and then use a somewhat capable image editor and cut the image part out you want .. but who knows? maybe the AI can understand what you want
>>101362571my intuition is that training the vae will be ok and training unet/dit will be the hard part, "transparency" or "transparent" is going to interact with other tokens, maybe an added special token would be bestwould be really cool if it works though, lots of applications
>>101362632yes, a different token is a must, this prompt right here uses transparent slime .. all models I know will handle transparent as something like glass, thin cloth etc.
>>101360890found a way to produce doppelgangers consistently .. add "symmetrical" to the prompt
Hawk tuah that thang
>>101362326>>101362392do you mean the image is very different or just a little different?Had this happen once and uninstalled add-ons the generation didnt use
>>101363129Also I think what was causing my issue was the prompt UI addon. It could undo changes you made in the text box when you hit generate
i miss schizo anon
>>101362996
Is there a way to get the robot to stop saying shit like tableau, or is it too stupid to do that?
Neeeeeeeeeerrrrrrrrrrrrdddddddddsssssssssssssssss
>>101362779its a useful token
Pretty gud for an AMD card I'd say (6700 XT)
oh no>PX2YN2I will not post my weiner.
might be a retarded question, but is there any way to extract parameters from an image with no metadata? like from an 'artist' that got their stuff posted on kemono
>SVJAY0
>brap
>>101364733Not really, clip similarity is bullshit and nobody bothered training a image->prompt model, the caption models we do have just produce gpt slop essays
um guys
>>101365829yea wassup?
?
fox
hello fellow prompterwhat do you think about this ponyXL derivative?https://civitai.com/models/282341/4th-tail-animehentai
>>101366076>>101365967shit is goofy, new fetish
>>101366301ya welcome!
where is everyone?
>>101366360it's too hot to gen 1girls in this weather
>>101366360ya >>101366374 is kinda right .. considering to undervolting my 4090 right now
>>101366360schizo won
>>101366386based
>>101366360in the pool
>>101366384you should be undervolting even if it wasn't hot https://github.com/LunarPSD/NvidiaOverclocking/blob/main/Nvidia%20Overclocking.md#undervoltingThe amount of wattage saved vs the minor FPS hit in games, or generation speed is worth it. Just record some figures before and after and see for yourself.
>>101366428She floats! Must be very light..>>101366459yaaaa.. guess you are right *sigh*
>>101366560Youve been up for at least 12 hours spamming this stuff
>>101366604oh no the important image slots ...
why does ai always insist tails come out the ass, there is definitely enough fox/cat girl image training data
>>101366360d*b* had a meltdown because schizo is seriously breaking himd*b* gone => thread has 80% less of his spam and "stealth posting"
>>101366660if you into anatomically correct tails you might consider ponydiffusion or a mix of it .. just sure to put source_pony into negatives if you dont want to have furry gens .. most other models think you want a tail butt plug
>>101366664it mostly just seems like they went to sleep anon
>>101366664>d*b* gone => thread has 80% less of his spam and "stealth posting"or he lives on the west coast and has been asleep for the last 8hrs or so.
>>101366604nah I slept 5 hours in between
>>101366699>>101366664Imagine actively bringing up the person who brings the general down when he isnt even here
>>101366736>the person who brings the general down seems to very much be here now
2dnPony_V10 is the best thing I've ever used.
>>101366774Nothing I said was refuted
>>101366774>you can't just make an accurate observation of irony, this means you are bad
>>101366798>>101366798>>101366798
>>101366801?
oops its replies not images.. dont wet yourself gentlemen Ill delete it again
>>101366801wtf are you even doing?image limit not reachedtext reply limit not reached
>>101366809Chaos!
>>101364379nice one
>>101366807>oops its replies not images.. dont wet yourself gentlemen Ill delete it againjust so you know since it appears you new to 4chan.text reply limit is 311image reply limit is 151
>>101366780is pretty good yeah
>>101366801>>101366725>gets called out for being an attention whore>denies it>creates next thread 50 images earlyYou're pretty desperate, anon.
>>101366878you can post your pictures too
>>101366818Thanks! I am actually not .. but as >>101366604said I am probably online to long with to little sleep .. bzzz>>101366878loool
Morning anons
>>101356172>>101366889>"gooanon" doesnt know reply or img limit hm...........
I've been out of it since the initial 1.5 hype. What has changed? Can I still run it on a 1050ti? Any improvements? Do we still use auto1111?
>>101366939we've actually all been waiting on you to move forward
>>101366939I couldn't use auto or the others besides easy diffusion, something with python versions and manjaro updates and system python and so on and so forth.easy diffusion is pretty great and apparently self containedif you're going for porn 2DNPony_V10 on civit is amazing
>>101366939I use invoke for inpainting>>101366894
Walking on water:1.2
>>101366939use comfyui desusearch filter civitai for pony models if you want lewd
Got a question: Should I get 3060 12gb or a 4060 8gb for both SD and vidya? Due to budget, these two are my choices.I only play at 1080p (would like to go to 1440p at some point) and for SD, I use Forge WebUI (for those with lower VRAM).I have a 1660 Super so either way, it's an upgrade.
>>10136705512gb vram wins
>>101366976
>>1013670553060 all day
>>1013670552 things, one; are you planning on playing GTA5, and two; is your monitor 60hz
>>101367072I thought so, I just wasn't sure if newer architecture would beat out higher vram.
pony.. little real control on style, 1 pic takes 2 minutes.sd 1.5, throbbing with style options, can gen a batch of like 14 in 2 minutes..conflicted
>>101367084Don't care about GTA5, I play a variety of old and new, like cyberpunk and TF2. My monitor is 100(+)hz, I just forget the exact number.
>>10136705512GB VRAM wins for AI
>>1013670972 minutes?! what gpu?
>>1013671261080, not fi. the pony pics are fine i guess, i suppose i just wish i had my 1.5 loras to use with it
>>101367157My 8gb 1080 takes 1min 3s for a 1024 with 2dnpony.
>>101367164let me try again, often the gen time changes for some reason idk
>>101367136clean!>>101367164ya okay. then it probably swaps to cpu ram, my 3070 8gb struggled hard on any XL model to, with 24gb vram now I barely notice the difference between an sd15 or xl model, also pic related is 2dnpony .. loras get ignored abit, so not the best model for style flexibility, has a nice core style tho
>>101367164that feels very slow?? it is 14 seconds on my 8gb 3070 ti
>>101367076
>>101367198The settings affect it a lot it seems. Adding a lora makes it double for example and some of the samplers do crazy shit.This was as fast as I can get it and load the better models.
>>101367220was curious and continuedthis was including the initial model load, with 5 lora, 30 steps / 5 cfg euler a30.33 seconds, don't care much for the bed.
sorry, i was fake-newsing, takes like 54 seconds. i guess first gen or so after switching to model takes longer or..going back to 1.5 anyway
>>101367322Yeah it does and you are correct as well. subsequent testing with heunpp2 karras and dpmpp2 karras resulted in similar times for me.
gm
>>101367322same prompt/image size, etc, just with vae enabled, but with a 1.5 model, 17 seconds. the model mix is partly that zoot model. i think ponyfied 1.5 models are best bet rn
>>101367397Nick stop making kids
>>101367422n.g.f.h.
>>101366939holy fuck anon is back hows it going bro
>>101367482Pedos gonna pedo
>>101367055I would lurk ebay's buy it now sorted by price and get something better
What the best thing for Warhammer 40 000 art?
>>101367552>when your thermostat comes to life and is DTF
>>101360320whys it so good
I teach bears math and feed them to dragons. Doctor, I think my robots are getting lost in a machine.>4X0HXP
>>101367524mfwhey little buddy, ill gen whatever you want,(for free!), anything at all, just, politely and respectfully, ask
>>101367675you winning in a fist fight against trani, please?
mfw no maid waiting at home
>>101367752why do you make these semi toddler face abominations?
>>101367775probably not that anons fault, 90% of all photo realistic models gen that korean plastic doll face when making asians
arghMY ROBOT SAYS thANKS FORF COOKING >YY6DJN
>>101367798best picture today
>>101367862gm
I actually like the AI asian face
>>101367524pervert detected
>>101367917did you choose the wrong meme?
>>101367978nta but you may be a lil slowthat's okay they're also wrong
>>101367917So when you saw the word pedo you immediately thought of yourself and replied? We'll be more careful next time
baka
anyone have advice for adetailer? Should I bother with hand, person, world? Does the detection model matter? I did hands and it made them worse, do I need to crank up the denoise?
>>101367978You see a depiction of a (clothed) girl doing nothing out of the ordinary and you start calling people pedos over it. The implication is that your mind went sexual after seeing a girl but you're calling other people pedos.I'd appreciate it if you didn't make these assertions because it's not good for any of us.>they're generating cp with these tools, look guys!>non-sexual image of a cartoon.jpg
>>101368112Generally faces are the most useful. Hands may be useful in some future case or with complex workflows but if you're just starting don't worry. For models, the most important thing is whether or not it's a segmentation model or not (has an s at the end of the model name). If it is segmentation, you will only denoise the face, whereas if it is not yiu will denoise the entire cropped upscaled square. This was true as of a year ago: for all I know someone solved the hand problem.
>>101367862Morning
>>101368088I can't figure out what you mean by this. You think I replied to my own post?
its almost news time
>>101368153So theres:face_yolov8s (segmented)face_yolov8nAnd mediapipe, judging by the names isolates parts of the faceIs it situational which is better? depending on the size of the area that has problems?
>>101368207>>101368260how are your gens still this terrible
>>101368376>>101368376>>101368376
>>101368143koff is a "no limits" type of guy
>>101368264>Is it situational which is better? depending on the size of the area that has problems?That I honestly don't know. I never got into it, since my current setup just werks. If you're starting out, just grabbing the basic yolo is probably fine.
>>101368378
>>101368264iirc mediapipe is for photos etc yolov for everything
>>101368378he gens only to post to convenient replies or to autistic fill the thread no other reason
language is fun
>>101368525esl nogen
>>101368549yep>GNW8W8
>>101368589(you) know anon is right
>>101368619nogen
>101368658good talk
Filing
>>101368666meant >>101368658
What's the best checkpoint and controlnet for inpainting hands?