General dedicated to local usage of free and open source text-to-image modelsPrevious /ldg/ bread : >>101108459>Beginner UIFooocus: https://github.com/lllyasviel/fooocusEasyDiffusion: https://easydiffusion.github.ioMetastable: https://metastable.studio>Advanced UIAutomatic1111: https://github.com/automatic1111/stable-diffusion-webuiSwarmUI: https://github.com/mcmonkeyprojects/SwarmUIInvokeAI: https://github.com/invoke-ai/InvokeAIComfyUI: https://github.com/comfyanonymous/ComfyUI>Auto1111 forksSD.Next: https://github.com/vladmandic/automaticAnapnoe UX: https://github.com/anapnoe/stable-diffusion-webui-ux>Pixart Sigma & Hunyuan DIThttps://huggingface.co/spaces/PixArt-alpha/PixArt-Sigmahttps://huggingface.co/spaces/Tencent-Hunyuan/HunyuanDiTComfy Nodes: https://github.com/city96/ComfyUI_ExtraModels>Use a VAE if your images look washed outhttps://rentry.org/sdvae>Models, LoRAs & traininghttps://civitai.comhttps://huggingface.cohttps://aitracker.arthttps://github.com/Nerogar/OneTrainerhttps://github.com/derrian-distro/LoRA_Easy_Training_Scripts>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>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>Share image prompt infohttps://rentry.org/hdgcbhttps://catbox.moe>Related boards>>>/h/hdg>>>/e/edg>>>/d/ddg>>>/b/degen>>>/vt/vtai>>>/aco/sdg>>>/trash/sdg
Blessed thread of frenship
>>101118874>he made the collage
>>101118795>>101119007>>101119074https://files.catbox.moe/rzwrqg.pngLet me know if you have any questions
Beware young anon for there be trolls afoot
reposting this early in hopes of finding manga painting Hunyuandit anon: >is the guy who did the "manga painting" still around? >>101095831 is there a specific style prompt to get that style? it's exactly what I've always wanted but sdxl never took to it well, even with loras
>>101119567Read my response in that thread>>101108570Hunyuan in general is a very strong base, by just prompting for anime, you are likely to get a variety of different styles. Just like with Dalle, you will naturally get "manga paintings" with just anime prompt itself, as the model has learned to associate anime and manga on its own (E.G. styles heavily inspired by manga), but you can also force it by asking for a painting that way.
So let's say I had a ton of things I'd genned over on Bing/dall-e which I liked stylistically and wanted to be able to gen using SDXL. I foundthis: https://www.shruggingface.com/blog/blending-artist-styles-together-with-stable-diffusion-and-lora on blending styles, and this: https://www.shruggingface.com/blog/self-portraits-with-stable-diffusion-and-lora on using Replicate to create the LoRA.Does that sound about right?
worthless dead generalwhy keep spamming this shit? just let it go
>>101118874Here's for double the chance to appear in community effort collage. Good luck in the next one!
>>101119222personally i was hoping to see more nostrils
>>101119866>was hoping to see more nostrilsis that a new fetish?
>>101119810please dont ever use one of my gens in your "collages", thank you
>>101119866Meh, image fucked up I'm not going to sit there to gen a perfect test to drop in catbox
>>101119886you're welcome
How do I upscale an image to 1.5x size with a 4x upscaler model?What you're supposed to do: upscale via a model to 4x, then Upscale By 0.375, but you can't enter the value of 0.375 into the node. The node automatically rounds it to 0.38.Even some Float node that I found rounded it to 0.38.
>>101119980Check >>101119222it might have what you seek
>>101120049
>>101120049>>101119222Thanks. Found a spot where it's just a regular nearest-neighbor 1.5x upscale and some other spot based on nodes I don't have.I guess it's just not possible in default Comfy then.Thanks.
>>101120184catbox please
>>101120212See if you have [Upscale Image By] node and try it that way.
>>101120281I did. That's what rounds to 0.38. It's not a big deal, I'm just surprised that it doesn't even allow 3 decimals.
>>101120281I guess I can just have two Upscale By nodes together. One 2x's it, then the next one does a 0.75x.
>>101120328Image will come out trash that way.Do Ksampler -> Latent upscale by 1.5 -> ksampler -> latent upscale by 1.0 -> Ksampler -> output your image.You can dump previews of image to each latent upscale to see what is messing it up.
>>101120363
>>101120217mean!
>>101120406but I love wind instruments
>>101120672glad to see you can relate
>>101120714prompt please
>>101120756a 21 year old beautiful nun, deep cleavage, micro skirt, thighs
how do i get consistent proportions? the seeds have random proportions
>>101120828Each model has only 20 seeds the rest is a variant of the 20.
>>101120363>Image will come out trash that way.If they're both nearest-neighbor then it shouldn't affect the outcome.>Ksampler -> Latent upscale by 1.5 -> ksampler -> latent upscale by 1.0 -> Ksampler -> output your image.Every time I've tried a latent upscale the outcome always looked like trash. This is also two extra 1.5x upscaling steps, so it'll add an extra 4-5 minutes per image instead of the usual 2 minutes for a 1.5x upscale. At that point I might as well just do a 2x upscale and I would still save time.
>>101121344Which model are you using?SD? SDXL? PixArt?
>>101121383Ignore the messed up picture details don't really have time to fine tune it.This image is the first pass through Ksampler.
>>101121439This is the second pass after latent upscale by 1.5
>>101121455after 1.0 latent upscale of last image and through ksampler becomes our final output.
>>101121439>>101121455>>101121470Damn it, it helps if I don't fuck up the workflow and set it up correctly the first time around
>>101121527>>101121439Redo
>>101121527>>101121455Redo
>>101121527>>101121470Redo
lol, the water
another lora failure, too large dataset>>101121588almost jesus
>>101121603it do be like that sometimes
>>101121603
>>101121577
>>101121615I'm too stubborn with loras, I should try find good settings for dylora. On paper it should be much better>>101121616Try parting the sea
>>101121639
>>101121952
>>101122346
>>101122497i am groot
>>101122839
>>101122890
>>101122908
>>101122945
>>101120557The composition is impressive, how did you make it?
>>101119705>Read my response in that threadThe prompt you replied with in the thread wasn't for the manga painting image though..? Am I blind and there's another reply?
>>101123059Nevermind ignore me I'm retarded, I confused the other anons cinderalla girls post for yours and didn't see your actual reply. Thanks anon
>>101123052added some art from Redline to lora dataset and just prompted for sonoshee mclaren in a cockpit. I use fluffyrock-terminal-snr-vpred-e132 v-prediction model so it's very reliable and easy to use
>>101122993
>>101123211I too can post 1girl
>>101123241450 years too old
>>101123241
>>101123251Tick-tock, time is running away from you as well kid. You just wasted your first half of life.
>>101123270
>>101121639Are you blind? The result is far from base image
>>101123329I care about the final image not the progress of it.I was displaying what it does in the background.
Any Udio alternatives that let you use whatever lyrics you want without censorship?
>>101123139Interesting, thanks
>>101123343
>>101123425
>>101123383As long as weights aren't leaked or a decent open model doesn't come out there will never be real alternatives.
ded
>>101123524So are you.
>>101123553in time, we all areenjoy it while it lasts
>>101123555I am
>>101123580
>>101123594
>>101123687
>>101124050
>>101124209
I know this will seem like a basic noob question, but I'm in desperate need in a simple point to a good model to use. I've faded out of using these ai's for a while, and i returned and used some of the sdxl lighting and turbo models and they all seem pretty good, but it feels like none of them are good at more basic stylistic paintings and such, despite older models being much better at grasping those kinds of requests without needing 20 loras stacked on one another to achieve it. I don't know, i'm not getting it. Either it's all photorealistic, or it's pony model shit, which all looks like ass.
>>101124312Use base SDXL
How you gen realistic pic with the pixart Is not censored?
>>101124333actually though? I'd be surprised if it printed out all these good results without these peoples divine "checkpoints" with 6 million images of trained data or whatever the fuck. But shit why not, i'll give it a try.
>>101124312https://civitai.com/models/139562/realvisxl-v40orhttps://civitai.com/models/58431
>>101124515
what is the difference between HunyuanDIT 1.0 and 1.1? 1.1 is the same model but updated, correct? is there any reason to use 1.0 instead?
>>101124561
rip
>>1011245831.1 is supposed to be improved but I found it's a bit more censored (with no tune, meaning the RLHF on the base model is at least more censored in my testing).Some more info-https://desuarchive.org/g/thread/100983399/#100984008In addition to this I ran my bathtub prompts-https://files.catbox.moe/aibty0.pnghttps://files.catbox.moe/hq1rdt.pnghttps://files.catbox.moe/mac2o5.pnghttps://files.catbox.moe/fty92s.pngand I couldn't get nipples at all, so it definitely is a bit more cucked.Though after testing 1.1 I found it's a bit more prompt adherent and also improved rate of hands for some prompts, which is already good enough on 1.0.
threads usually last 24 hours. this one is only 8.
>>101124810Though I must say that my assessment on fake skin filter for 1.1 has changed a bit. It's capable of it, that is, you can prompt for selfies as usual, it's just less likely to succeed compared to 1.0 and I noticed more boring or plain/fake looking backgrounds too when it's trying to censor it. The distilled 1.1 version is 100% not capable of giving you real skin, which is where this is present the most. I keep both versions around.
Can't wait till Sigma 2.0 comes out
>>101125165
>>101125151release date when
>>101124387Not censored
ded thred
quiet, anon sleep
sigma 2.0 waiting room
>>101120660Moody
>>101118874 can someone instruct me why the install bats for comfyui manager is failing constantly? I dont even get a chance to see why it fails. I'm trying to use the portable version. I have no idea what i am doing.
>>101126922>>101126921these are cool>>101126933fight scene with hunyan?
If I want to make little fantasy jrpg game monsters/creatures, what model should I be eyeing up? I have no issues training loras, I imagine I could try my hand at a small sdxl fintetune if I really had to. not sure if I should be considering pixart or something?
>>101126933nice fighting
>>101126936>fight scene with hunyan?yep, Chinese prompt is the only way to get it to understand what a high kick is
>>101126939>monsters/creaturesI like to use Pony + i2i with some other model to give it unique look. From 1.5 to sdxl there's lots of models you could use>pixartToo early for it>>101126953Does it understand Hong Kong action movies? Hard Boiled by John Woo
>>101126939I've had a lot of luck with Pony models. https://civitai.com/models/288584/autismmix-sdxl as an example to get you started.You probably want a style lora that will suit your needs best. Iit might lean towards monster girls, but you can always tard wrangle it with img2img or inpainting.
>>101126982>Does it understand Hong Kong action movies? Hard Boiled by John WooThis is what I got for "Cinematic screenshot, Hard Boiled by John Woo", no idea if it's from that movie
>>101127039looks little bit like Chow Yun-fat. In chinese it's 辣手神探 according to wiki
>>101127066
>>101127066>>101127099At the end of the day it's probably seen it somewhere in its material but it's hallucinating as it doesn't remember precisely, but it's pretty good with plenty of video game characters and anime. For instance pic rel is supposed to be a video game character and it's a pretty good rendition of her even if not 100% accurate.
https://www.reddit.com/r/StableDiffusion/comments/1dn6yjp/stable_cascade_weights_were_actually_mit_licensed/
>>101127127>SD3 and Cascade have the same license guise i swear
>>101127127Interesting, but I'm not sure if that alone would make people gravitate towards it. For what little we know about castade, it lackedany noteworthy breakthroughs, didn't it?
>>101127160maybe, people skipped it because of the license and because SD3 was on the horizon. with SD3 being DOA and the license problem gone, maybe we will see something with cascade.
>>101127127Interesting, too bad Cascade is not a DiT model and even if it had potential it was never properly pretrained by SAI.
>>101127124>pretty good with plenty of video game characters and animeYeah I've noticed. Does it know hopping chinese vampires? 僵尸 or 殭屍. These are extremely silly
Can someone explain what DiT is to the layman?
>>101127127doesn't that mean that it technically stays a MIT licence? because that's how it should work, but I'm not sure
>>101127198The implication is that whatever version they first released, still remains MIT, even if later they changed it.
>>101127197DiT = Diffusion Transformers, that's basically a new architecture that is objectively better than the old Diffusion architecture we had on SD1.5 and SDXL, Sora is a DiT for example
>>101127194Prompt is just 僵尸 Though you can probably get what you want out of a more detailed prompt
>>101127207So it means that Stable Cascade can be saved? That's a good news no?
>>101127256>So it means that Stable Cascade can be saved?Possibly.>That's a good news no?If it proves to be of use. Better to have the option than not.
>>101126982>>101126989Thanks friends, I'll give it a shot with pony/i2i
>>101127254looks like anime version perhaps>>101127282good luck
>>101127282Just make sure to include the score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up prefix. Unfortunately, it really makes a difference. If you use something like Automatic1111 or forge, you can put it into a style, so that it doesn't get in the way of prompting.
Gen quality of this thread has hit absolute rock bottom. We need to force some of the spammers back into /sdg/
>>101127356Unfortunately this is 4chan, meaning no janitorial powers to us. I am almost tempted to make a discord server, but that would kind of defeat the purpose of this general existing in the first place.
>>101127362Yeah but what exactly are we supposed to do then? Low quality, low effort super blurry gens getting spammed by one anon and nobody even reacts? We have to make a change /ldg/
>>101127381report batch spamming>>101127362sdg made one and the quality suffered even more, it's now a place where they air out their discord drama.imo the lack of an AI board meant this was bound to happen
>>101127398It's not about reporting though, is it? He was forced out of sdg several times by coordinated action and now he's here and nobody wants to put in the effort to improve the thread? Wtf?
>>101127409we would need to move to another board, the same group "running" sdg is also "running" ldg.
holy b8, Batman
>>101127418No, we have the power to run this thread, and we should absolutely exercise it by coordinating abuse towards those who are not a good fit. I don't want to see this devolve into another /sdg/ and if we don't take a stand against this shit, we will also become known for being a shit thread full of blurry sloppy gens
>>101127424which gens are a problem? thread seems fine
>>101127429NTA but he's referring to these:>>101127267>>101127216>>101127150Every gen from this anon is exactly the same style, exceptionally low quality and super blurry. We essentially have a second debo on our hands and absolutely nobody wants to do the thing that needs to be done
>>101127439not the original anon but now that you mention it there is something very debo about this anon, what can we all do to help?
>>101127439nogen fag
are there any other blue boards for ai imggen? there is a vtuber one in the op, but i didn't find anything else.
Stable Cascade is an MIT Modelhttps://huggingface.co/stabilityai/stable-cascade/blob/e16780e1f9d126709c096233d96bd816874abef4/README.mdhttps://github.com/Stability-AI/StableCascade/commit/209a52600f35dfe2a205daef54c0ff4068e86bc7
>>101127421Indeed>>101127454right here in /g/>>101127455>>101127127
>>101127452Sorry, but can you actually open up those gens I've posted and tell me you're okay with them rather than having a knee jerk sperg out?
>
>>101127469ok you have a point, but what can be done?
>>101127356>>101127362>>101127381>>101127398>>101127409>>101127418>>101127424>>101127439>>101127451>>101127469>>101127496Breathe in, gen out
>>101124312https://civitai.com/models/175298/one-mix-xlone mix you need \o/
>>101127381>nobody even reacts?Because that's the best you can do on a platform without major moderation, or giving it's users the ability to moderate themself, if at least on a thread level.That's partially why I started making those collages, so as to encourage creativity and reward anons for the community effort. It ain't much, but it's something.>>101127398>report batch spammingI'm not sure that's viable. I'd sooner get warned for a risque gen, then have someone clean up arguable spam that I've reported.>sdg made one and the quality suffered even more, it's now a place where they air out their discord drama.Very fair point. Then again, it's the same users bringing their shit over to the neutral grounds of their 4chan threads. That kind of behaviour could be penalized by banning their ass for propagating drama, were it to repeat in a better handled community. Unfortunately that doesn't solve the issue of trying to run this general here, as supposed to.>imo the lack of an AI board meant this was bound to happenPartially, but shitposting and trolling is pretty much bound to happen anywhere without proper moderation and a relatively toxic culture.>>101127429>which gens are a problem?Examples I'd point to would be >>101120538 >>101120517 >>101120449Tl;dr I'm afraid the best we can do is report spam, ostracize shitposting by refusing to engage with it, similarly to how we refuse to engage with bullshit over at the other general, and remind others to try and not bite the ragebait either.Other than that, positive reinforcement, as in focus on actually worthwhile things. Praise creativity, engage in meaningful discussion, help out anons by sharing insight, and nurturing the anonymous community by collaboration, whether in forms of events, collages, or any other small scale projects.Breathe in, gen out indeed.
>>101127127>>101127455We're so back
ooooooo the next collage is right around the corner! :3
>>101127455somehow I doubt that it will become popular because of this
>>101127654hot whats the prompt for the bikini ?
>>101127530You've ignored the absolute worst gen poster in the thread and in doing so have massively understated the problem
>>101127530Agree with everything you said. I think that the core of the issue is that every gen that comes out from an XL tune looks the same, so it's bound to look like low effort especially across similar prompts. We need to encourage anons to break out of these habits with more creative gens, perhaps different themes across threads E.G. pixel art, and maybe some detailed rentry guides for pixart and hunyuan in case they are struggling to move on from XL.That is only possible assuming they are not deliberate trolls, otherwise just reporting repeat spam is the only option.
>>101127713He is a deliberate troll, he's absolutely spamming /sdg/ full of this shit, and he's doing the same here. His gens are almost painfully intentionally blurry and poorly composed, and something must be done.
>>101127713>That is only possible assuming they are not deliberate trollsI love 1.5, can't help it
How do I view the actual raw tokens used to generate an image, rather than just the prompt
>>101127772print(*whatever the fuck the token is called*)
>>101127689You can't expect me to take note of every single shitposter, anon. What would you have me do? I'm not a janitor. As said, I can make a meme or a collage at best, remind others not to bother wasting their time and effort on shit's worth bait, or try and drown out the noise by posting something more worthwhile.>>101127713Time will tell. Wouldn't rely on rentries though. Encouragement of creativity needs to be more approachable, and so it's best left to inspiration. Most folks learn through osmosis, meaning the more creative gens you see around, the more likely anons are to take notes. XL in of itself isn't the issue, though I do agree there are models and finetunes that are better suited for artistic purposes, just as there are UIs that favour a more creative workflow.
>>101127672catbox - https://files.catbox.moe/68t305.png
>>101127808No I'm sorry this particular shitposter is different and I think it's pretty important the community do something rather than just ignore and hope he goes away
>>101127713Other than that, we can share tips at most. My most valued one is recommend anons to try and put some meaning, symbolism or narrative in their gens. Try and have your images convey something, or include meaningful references, other than just being aesthetically pleasing. This is also why I always recommend anons to start playing around with inpainting and img2img. Devil's in the details.
>1gen :|>gen that can by analysed for interpretation :O
>>101127831>https://files.catbox.moe/68t305.pngnice lora
>>101127941which one?
>>101127808you've ignored almost the entirety of the problem in your response - the issue is this thread has no barrier to entry for shitty prompters, which is absolutely and completely required to keep a place like this runningotherwise, just like that anon said, you end up with 15 absolutely blurry as fuck zero creativity gens like that one anon is spamming and other prompters dont want to post actual good gens alongside itbasically i think the general is far too accepting of absolute slop to be taken seriously and until some kind of coordinated action is taken, /sdg/ will remain far more active
>>101127946i guess the Kenva giving it the comic look
We need organised action from a group of anons who have this generals best interests at heart
>>101128022i don't think the issue is quality, but rather the spam that harms the thread. posting only 1 image from a batch or making a collage should solve most of these issues. there seems to be no avatars here, which also stops the usual image spam to blogpost / mass reply
>>101128022If the problem is just one anon and his SDXL spam can't we ban SDXL gens and move that to /sdg/? If it results in increasing the overall gen quality it's not a bad idea
>>101128023Yeah, found it in a pony styles pack
>>101128072that won't solve the problem at all, and it will limit other anons that actually want to contribute.
>>101128065Meh even just one gen from 0anon is enough to make me close out the tab and swap to the disc or do some workflow tweaking. There is just no reason for me to engage with debo tier shit gen spamming
>>101128085How will it not stop the problem? Every gen is identical so just ban that kind of gen and move it to the containment thread
>>101128126>so just banhow exactly are you going to enforce this beyond quoting them and telling them to gtfo?
report spamming is the usual technique but its ineffective on /g/take my advice, only anon can make a change and that is by being more openly hostile to this particular brand of blurry gen spam
>>101128126ban an entire set of models because of 1 anon? what stops said anon from using another model? are you going to ban that too?
>>101126616what install bat? have you tried running it directly from cmd? both cpu and gpu?
>>101128147If we put it in the general OP we can report for off topic and the jannies will be forced to help us, basically
>>101128152for >>101126616sorry
>>101128152>what stops anon from using a other model?Every gen he has poster for 12 months is essentially identical, yes. There was even a compilation going around proving this the last time he was taken off the general
>>101128174just disregard everything lol im tired
>>101128167that's not how 4chan rules work. the offtopic rule is board wide and attempts to punish/warn against non tech discussion. jannies are not enforcing "rules" someone puts at the start of a thread. I would love it if they did but they don't.
>>101128206They absolutely will dish out temp bans if you spam a general full of posts against the generals rules and they get mass reported. We are not as defenceless as you're making out
>>101128224it's worth a shot if it has any chance to lower the amount of blurry slop
>>101128186then what? you ban the model, anon switches to another model and does the same. now this other model also needs to be banned, until we ban all models? isn't this giving power to the spammers? "hey guys just spam the shit out of a model and we will ban it"
>>101128261He wont switch because he cant. Maybe you're a newfriend but this is not a new issue
If it helps improve the thread quality i would support it
>>101128273how do you know that? and that still doesn't address the question i made
>>101128326Because it's been exactly the same situation for almost a year. The rest of your points don't apply because you're using the slippery slope fallacy - just burning out this one anon isnt going to have the downstream effect you're asserting it will
>>101128338hmm points taken
>>101128338>just burning out this one anon isnt going to have the downstream effect you're asserting it willnta but I'm with you on this. There's always going to be someone new to image generating and just posting everything that comes out of their queue because they alone are "blown away" by it being their first time.
>>101128338it isn't a fallacy it is a slippery slope, but sure go ahead, guess this place isn't for me either.
We dealt with this anon before the thread split without having to ban any models. I think they are a brat and coordinated systematic abuse doesnt rely on the jannies to work. It's a better tactic, and will likely force them back to sdg
>>101128397It isnt a slippery slope until it is one, taking action that are necessary to prevent the enshittification of our community shouldnt be avoided just because we are afraid of one single anon. If you want to run away to the sdg safespace and enjoy blurry shit gens and constant drama spam, fine, but I am at least willing to say when something isn't right and needs to be fixed
>>101128421taking actions*
is local prompting or whatever its called possible with ponyXL? im talking about using more than one character lora and producing two different characters. im using comfyui
>>101128489>>101128489>>101128489>>101128489
>>101121288catbox??
>>101128498I cannot endorse this bake. I'd rather doom /sdg/ to oblivion, than lower to it's standards by remarks such as:>For discussion and images relating to SDXL and SD3 please refer to our sister-generalThey've done something like this recently, and they're the last place I'd like imitate. The /l/ in /ldg/ stands for local models, and SDXL is one such model, for better or worse. SD3 is more questionable, but discussion should be allowed, if to learn from it's mistakes.If anyone can agree with me, I'd like you to consider this bake instead:>>101128616>>101128616>>101128616
>>101128629Why is there always so much retarded drama in these threads?
>>101128663it's probably the sdg schizo that they sent our way, i won't be surprised if he is the one behind the "lets ban XL/SD3" he was always against the ldg so this might be his way of forcing anons back into sdg
>>101128663I honestly don't know. Can only guess that /sdg/ has gathered an infamous following that stirs shit wherever it goes. The more a palce looks like a cesspool, the more people act accordingly to it's atmosphere.
>>101124209awesome aesthetic, reminds me of old school scifi/fantasy cartoons