Discussion of free and open source text-to-image modelsPrevious /ldg/ bread : >>101507621>Beginner UIEasyDiffusion: https://easydiffusion.github.ioFooocus: https://github.com/lllyasviel/fooocusMetastable: https://metastable.studio>Advanced UIAutomatic1111: https://github.com/automatic1111/stable-diffusion-webuiComfyUI: https://github.com/comfyanonymous/ComfyUIInvokeAI: https://github.com/invoke-ai/InvokeAISD.Next: https://github.com/vladmandic/automaticSwarmUI: https://github.com/mcmonkeyprojects/SwarmUI >Use a VAE if your images look washed outhttps://rentry.org/sdvae>Model Rankinghttps://imgsys.org/rankings>Models, LoRAs & traininghttps://civitai.comhttps://huggingface.cohttps://aitracker.arthttps://www.modelscope.cn/homehttps://github.com/Nerogar/OneTrainerhttps://github.com/derrian-distro/LoRA_Easy_Training_Scripts>Pixart Sigma & Hunyuan DIThttps://huggingface.co/spaces/PixArt-alpha/PixArt-Sigmahttps://huggingface.co/spaces/Tencent-Hunyuan/HunyuanDiTNodes: https://github.com/city96/ComfyUI_ExtraModels>Kolorshttps://gokaygokay-kolors.hf.spaceNodes: https://github.com/kijai/ComfyUI-KwaiKolorsWrapper>AuraFlowhttps://fal.ai/models/fal-ai/aura-flowhttps://huggingface.co/fal/AuraFlows>Index of guides and other toolshttps://rentry.org/ldg-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>Try online without registrationtxt2img: https://www.mage.spaceimg2img: https://huggingface.co/spaces/huggingface/diffuse-the-restsd3: https://huggingface.co/spaces/stabilityai/stable-diffusion-3-medium>Related boards>>>/h/hdg>>>/e/edg>>>/d/ddg>>>/b/degen>>>/vt/vtai
official pixart bigma and lumina 2 waiting room
sigma cookin........
sigma is so easy to tell anons, it has a very distinct aesthetic, not a fan.
>>101533895nogen is so easy to tell anons, it has a very distinct aesthetic, not a fan.
>>101533895>>101534040To be fair, I can't wait until the Pixart team gets us off the SDXL VAE
Good bye for now anon
blessed bred?
>>101532879>LDG no longer shows when searching SDG Curious...
>>101534730RIP, baker didn't link >>>/g/sdg in the related
>>101534079>I can't wait until the Pixart team gets us off the SDXL VAEthe 16vh VAE would be a better fit, someone needs to train this VAE to work on pixart though
>https://rentry.org/ldg-link >Pub: 22 May 2024 >only now linked
I hope bigma is easier to install since sigma filtered 90% of anons Most still have yet to experience the goods locally
>>101534730where long dick general
mikuslop
llama 3 obese finally dropped, anyone here planning on running it locally as a prompt generator for pixart?
>>101535425The q4 of the 405b is 234GB. There's not going to be many people running it locally in these image gen threads. You'll have more luck over in /lmg/
>>101535425Use dolphin-llama3 8b currently with a modified prompt based on something in the ComfyUI-IF_AI_tools repo and that works pretty well so far.>Act as an objective, emotionless visual prompt maker with the following guidelines: - Describe the visual aspects of the image in detail using simple language. - Use complete sentences or partial sentences. - Provide high-quality, non-verbose, coherent, concise, and not superfluous or emotional descriptions. - Focus solely on the visual elements of the picture; avoid art commentaries or intentions or narratives. - Construct the prompt by describing framing, subjects, scene elements, background, aesthetics. - Limit yourself to three or four paragraphs. - Be varied and creative. - Always reply on the same line. - Do not enumerate or enunciate components. - Do not include any additional syntax in the response. - Feel free to elaborate in several sentences but keep responses to about 3 paragraphs. The following is an illustrative example for you to see how to construct a prompt your prompts should follow this format but always coherent to the subject worldbuilding or setting and consider the elements relationship: 'This is an epic covert art with a dynamic angle. It depicts a demon hunter with glowing eyes and a cybernetic exoskeleton with sleek metallic details and glowing blue accents.' Make a visual prompt for the following Image:Input:>artwork. ink drawing. ink wash. beautiful girl. rainy day. yellow umbrella. reflective city sidewalks. at night.Output:>This artwork, an ink drawing with an ink wash finish, captures the essence of a beautiful young woman on a rainy day. She stands under her bright yellow umbrella, shielding herself from the downpour while observing the reflective city sidewalks around her. The scene takes place at night, with the urban landscape illuminated by streetlights and neon signs casting an atmospheric glow.Picrel is input vs output on pixart900m. Not sure which I prefer sometimes
series of SFW bitches i did with juggernaut xl v9, via the new intel ui, dumping
nope looks like i got cockblocked
>>101535739which arc gpu do you have and how fast was each of these images?
>>101535927a770 and a few seconds for 50 steps
why the fuck does every character lora i train result in the generated character having smaller breasts than the training images?
>>101535939damn, that's faster than I was expecting for that res. Seems like Intel have caught up a bit, albeit at the expense of being limited to their software.
>>101536062I ran a timer for a more accurate answer. The 50 steps actually take 30 seconds or so. The jailbreak prompt that got fixed on Bing works fine here and I'm getting a lot of booba and occasional pusey
I think I make the OP collage maybe 90% of the time. Its so nice seeing my gens in there
How would running local on an RTX 3080 compare to NovellAI's image generation? Trying to decide if it will be worth the time setting it up
>>1015364773080 wouldn't be too bad desu worth it IMO
>none of the women peeing LORAs I try actually make them pee, instead they just mutate them to have extra legs or be squashed no matter what strength I select
>>101537108I'd be pissed
>>101536281>and I'm getting a lot of booba and occasional puseyGodspeed, anon.
sports car on the road
>>101538394CLOSER! Hype!
>>101538394thats a sports car alright
>>101538394>256x256
>>101538521You're welcome to make a model from scratch using 10+ million uncensored images.
>>101538521>he doesnt know
>>101538543I prefer to do nothing rather than making a shit model no one will use, you're wasting your time anon
>>101538781I'll use the model and I probably won't even release it :)
>>101536477would be good, go for high vram.
do artists only get mad if you are trying to make money off models trained off their art?
>>101538806>I probably won't even release it :)
>>101538839Weird because you seem to care even though you say you don't. Anyways... enjoy your censored slop models.
>>101538838Artists are low IQ and lazy, AI existing is an existential threat so they will be against all AI models and will move goalposts along the way and use every possible argument against them.
>>101534413Do you have the un-Adetailered image? Can you load it, VAE encode it and pass it to Adetailer? (I don't use Adetailer, I just refine manually selected regions in Krita).
>>101538848>Anyways... enjoy your censored slop models.anon, I'm a old fag and I enjoyed myself on old ass 360p videos back then, but c'mon, I can't fap on 256x256 pictures, you're asking too much of me ;_;
>>101538543>uncensoredI don't see a pussy on that sports car.
>>101538891I'm training up to 1024px. But you have to train low res first otherwise you waste a lot of time. 256px means less decisions for the AI. But the learnings it gets from 256px can be stretched to 512px then to 1024px. Technically it can generate 384px images because of the ratio bucketing.
>>101538935>But the learnings it gets from 256px can be stretched to 512px then to 1024pxit's not gonna look good, if you want an AI to be good at 1024px it needs to be trained on that resolution
>>101538960Is that what I said? I know you're dumb but take a second to use critical thinking.
>>101538982who cares anyway you're not gonna release that shit, you're just wasting everyone's time with your retarded 256x256 pictures, this place isn't your blog get the fuck out nigger
>>101539023Just me training a model is relevant to /g/.Don't you have /degen/ or somewhere to be? Still wanting to kill yourself at work centaur anon? Not enough 1girls being posted?
>>101539056>Just me training a model is relevant to /g/.no it's not, it's /ldg/ here it means LOCAL, your model won't be local so like I said, get the fuck out with your offtopic piece of shit, kill yourself
>>101539084Runs locally on my machine lmao
>>101539119and dalle3 runs locally on Microsoft's machine lmao
>>101539133That's not even close to right, are you technologically illiterate? Really living up to the boomer stereotype lmao.
>>101539119>lmao>>101539148>lmaoOk Zoomer, keep eating Tides or whatever
>>101539180You don't even know what local means.
>>101538806>probably won't even release it:d pls anon
>>101538394it starts to look like a sports car, yes
>>101538781>>101538839>>101538891>>101538960>>101539023>>101539133>>101539180Absolute retardation
>>101538872>Artists are low IQ and lazy,*bad artists desu
>>101539570If I release it it will be a dead file drop somewhere because this model certainly isn't going to make copyright holders happy because trained on stuff like the full Alice in Wonderland animated movie.
>>101539945>full Alice in Wonderland animated movie.kino
Just when I start losing confidence in Dreamshaper, I find a way for it to deliver on what I want. (or damn close to)
>>101538838I've seen them celebrate people who've trained loras on their own art completely overlooking the fact that base SD has loads of protected art in it already They simply do not understand how any of this works
>>101535668In my case the Demon hunter usually leaked and mixed into output so I made a simpler one
too much pink
>>101540597dreamshaper dreamshaping
Where there is a dream, there is a way.
>>101539148HYYYYYPE!!! Best person so far>>101539728kek. Boomers and their gotchas
>>101537577Carlos, I will feed your spic ass through a 4B LLM I swear to God
But can dreamshaper shape my dreams ?
>>101542094Maybe, but it can't shape mine. I can't even get the same character consistently with the same prompt.
>>101542120 1girl, big boobs, {celebrity1}, {celebrity2} simpleas desu
1girl, big boobs, {celebrity1}, {celebrity2}
>>101542154Do you use the celebs with weak weights to keep it from actually looking like them or something?
>>101542175depends on how well represented the celeb is in the model. some i will have to lower, some raise. some ill omit entirely and find a replacement. just throw a bunch in there to begin with and adjust based on the output.
also, defining hair color/style, body size, etc. all things that will help in garnering consistency from any model
>>101542245Fuck, now I have to recover from my extreme bordering on amnesia apathy towards celebrities and remember which ones I liked the appearance of and their names>>101542277I do like hair color it seems to lock into some features when combined with clothing for some reason, just sample bias I guessAlso SD loves to ignore clothing fully or partly
>>101539148Was the prompt for this one just >man standing ?
>>101542289>Fuck, now I have to recover from my extreme bordering on amnesia apathy towards celebrities and remember which ones I liked the appearance of and their namesNo, you don't. Just look up names and experiment.
You can experiment with celebs, cartoon characters, anything, really. You just need names. A given model may not even respond to certain names, so trial and error is the only way forward.
One day you will be able to enter your own name and receive your own face without the need for loras.
>>101543096that would be messed up as all heck, /b/ro
>>101543040>top list of names of celebritieslolcan I just have a list of names of ones that work?>>101543069I used to put in chun-li a lot but she's almost too strong until a certain point of toned down and then she just ceases to influence at all it seems like
>>101543155>can I just have a list of names of ones that work?no one has a thorough and complete list because anons don't cooperate on making oneGood ol' Trial and Error(TM) is inescapable.
>>101543304Can I have YOURS, my friend in cooperation?
>>101542543>1boy wearing jeans and a t-shirt
>>101540597Plebbit model
>>101539056>centaur anonkek. glossed over this part at first
>>101533895Your'e eyes have fully adjusted to SDismsA full removal is in order
>open old anon catbox >notice he chose width and height by connecting any two of 15 float nodes
>>101526015v0.9give me a sec
>>101535759made this qt too
fix inpainting for fucks sake
Good morning anon
Bump
>>101538394Hey man it's starting to look great. How many epochs/days to go?
>>101549106Nice
>>101549111Many days, there's still a lot of prompts that aren't even close.
>>101549642Could become killer model for finetunes. I still have to train everything for 1.5
>>101549497Thanks. And back at you : nice stuff.
>>101549866ty>>101549868ty https://files.catbox.moe/fr7qxr.png
>>101549631Windows XP wallpaper vibes. Nice.
>>101549631>>101549932long dick general has 1girl at home so this is what you get
Are you kidding me Sigma? Mind blown on this one
i last checked on this stuff back when cumstain and co were attempting a mass exodus from sai to make their own model. checking back in, seems like little has changed and people are now even saying sd3 was a 'beta' and to wait around longer for the fixed version. is it still just an eternal waiting room for something actually good to come out?
>>101550041There's other models, but no base models that beat years-later finetunes of SDXL. You can keep waiting or make your own
Good morning genners
>>101550532Thank you :)
>>101550495Good morning>>101550532>i like how the mountains lookAlso thisAnd good bye for now
big ma
>>101551241Very cool
>>101551241nice
>>101551871>imgonline-com-ua-dexifK1RIw450o0BP.jpg
>>101551241Nice
>>101544000Plebbit touristshoo
Later
>>101553136Later
Is Sigma (or bigma) SOTA? Why hasn't anon switched?
>>101554308brap
jannies smoked anon
>>101554308what is she doing?>>101554430cool gen
>>101554654Contemplating life.
Yo
gen
>>101556776>>>/g/ldg right now
>>101534730s'why dead
>>101534730>>101557810Only 52 gens to go
>>101557844Go !
>>101557844Forgot pic
>>101557860Choo choo
Chugga chugga chugga
sigma cooking
Come out anon, I don't bite
kissa me, kiiiss me
>>101558033>>101558250Very good details
>>101558265ty
Been a while since I had a good one of these. I wonder how much carbon they release vs driving a car
>>101558791always wanted to eat the charcoal as a kid
>>101558818Now is your time to bloom. Don't be held back by your childhood fears. People unironically put charcoal in drinks thinking it cleanses them instead of stealing the nutrients.
It's that time again, come and get your fresh...>>101558992>>101558992>>101558992
>>101559098Nice.. filling up before I go
>>101558992>>101558992>>101558992