Discussion of free and open source text-to-image modelsPrevious /ldg/ bread : >>101774288>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://github.com/Nerogar/OneTrainerhttps://github.com/derrian-distro/LoRA_Easy_Training_Scripts>Fluxhttps://huggingface.co/spaces/black-forest-labs/FLUX.1-schnellhttps://comfyanonymous.github.io/ComfyUI_examples/flux>Pixart Sigma & Hunyuan DIThttps://huggingface.co/spaces/PixArt-alpha/PixArt-Sigmahttps://huggingface.co/spaces/Tencent-Hunyuan/HunyuanDiThttps://huggingface.co/comfyanonymous/hunyuan_dit_comfyuiNodes: 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/sdg-linkhttps://rentry.org/rentrysd>GPU performancehttps://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>>>/aco/sdg>>>/trash/sdg
>mfw
from last thread: I checked dev vs pro on the API, and pro isn't that much better, and in fact e.g. renders text worse than Pro. is it really that better?Why would it be bigger?
>>101776833i'm using a 1080ti Takes a while to generate but it works>>101776969that program displays the vram wrong they're all 11GB.. non ti 1080s have less vram
Thank you baker
>>101777010it bothers me how much of a bottleneck VRAM is and how hard it is to get it.like imagine one day we wanna generate Videos with this.
>>101776953Definitely not. Left is FP8 Dev. Mid is FP16 Dev. Right is Pro. It's not a little bit of a difference. It's a massive fucking difference.
>>101777067Well, when there are only two notable GPU manufacturers in the world and the CEOs of said companies are cousins, there's not much you can do.
>>101777090anon, you do realize this could be just because they intentionally made dev weaker?
Anyone got good use out of the flux realisim lora? https://huggingface.co/XLabs-AI/flux-RealismLora
>>101777102Of course they intentionally made it weaker, what a dumb thing to say. It's very obvious that it's weaker.
>>101777120My point is that it might as well be the same 12B size
>>101777124It's not. I promise you.
>>101777133meh
>>101777090>more colors is gooder
>>101777134You can see the gens happening on FAL as you gen them. I am out of tokens now, but it's obvious it's not just an unpoisoned dev. Dev is excellent. Pro is significantly better.>>101777144Yes, more colours, better prompt adherence, better reflections, more detail, is gooderer.
>>101777134He's right. It's probably not 12B. Intentionally brain fucking the model to give to the poors would have been a lot more obvious.
ermm didn't they say that controlnets are impossible?https://huggingface.co/XLabs-AI/flux-controlnet-canny
>>101777160>more detail, is goodererVermeer rip, :^)
>>101777144while I agree it's hard to draw any conclusions from just the picrel, there's more to itconsider the reflection: pro has more details and those details are accurateconsider the textures: dev looks more mspaint-y with flater colours. pro looks more like a render in terms of textures
>>101777207I literally saw nobody saying controlnets were impossible.You might be confusing it with the speculation that finetunes may not improve the model.
>>101777208
new anime came out
>>101777060dios miku
>>101777090>we don't get the best model againopen source is doomed
I didn't get to post in the chibi thread :(
>>101777207>We are working on ComfyUI node support, it will be available soon.
>>101777295Pay them $100 million and they'll give you their best model. Then you can release it as open-weight. Huh? Why are you not doing that?
>>101777317There's an open PR in ComfyUI already
>>101777248ruff for humans>>101777295>communism 2.0 fails>pikachu
>>101777318If I had the model I would release it for free.
>>101777295>for profit company isn't giving their product away for free
>>101777326Protonmail sucks now lol>Dear User,Your Proton address is currently restricted from registering on third-party services.To lift these restrictions, you need to add a verified recovery email address or phone number, or consider upgrading your account. Learn how to verify your recovery email address: https://proton.me/support/set-account-recovery-methods#how-to-add-or-change-a-recovery-email-address and phone number: https://proton.me/support/set-account-recovery-methods#how-to-add-or-change-a-recovery-phone-number Find more information on upgrading your account: https://proton.me/support/proton-plans/Why did this happen?Our system flagged third-party service registration emails shortly after your account creation, indicating potential disposable account usage. As a result, we have implemented these measures to protect the overall reputation of Proton Mail. This helps safeguard our domain and prevents the risk of our IP being banned by third-party services, which could impact service availability for all users.Thank you for your understanding and cooperation.Best regards,The Proton Anti-Abuse Team
>>101777295Flux dev even in fp8 is leagues ahead of anything we've managed so far. And this is just the first version.
>>101777384Huh, why did you reply to me? You don't have to register on GitHub to read the PR or to pull it.
the real Problem is that THERE ISNT ENOUGH FUCKING VRAM IN THESE SHITTY FUCKING GRAPHICS CARDS FUCK
>>101777401yet you'll continue buying nvidia, cementing their position even more.
What happened to Pixart/Sigma btw?
>>101777304We don't do chibi threads.
>>101777409there is nothing else I could buy that has more than 24GB VRAM
Flux is really neat because you can do a lot with images, probably even useful for website concepts
>>101777393>wow Internetgive me your phone numberThere's a reason we want to deport you back to china.
>>101777409As if i'd buy AMD. I'll consider it if they ever come up with some proper competition.
>>101777295Waah we only got the very slightly and often unnoticeably inferior version of the greatest image generation model to date which only promises to improve as the toolkit becomes available.
>>101777410nvidia bought it so they could block the release. they're going to roll it up into some kind of "nvidia creative platform" in the future
>>101777431yeah, real dalle3 at home (dalle could do all that but with worse coherency since its a much older model based on an older architecture)
>>101777440Interesting...thanks
Super glad I never bought protonmail.
>>101777428You can buy an a100 for $15k
>>101777435
>>101777420I meant the last thread with my OP pic.Also, it would be cooler if you did
>>101777506Can you gen one without the blur?
>>101777428>get an a6000 with 48gb vram>run the full fp16 model and full fp16 encoder all in your vramI'm sure some anon has done that already, right?
>>101777393*stabs you to death**>literally pushed non-free software*I guess I won't actually stab you to death, I found a way around it.
>>101777536are you retarded? please seek medication.
I hope you people know the reason software isn't free is you are not moral enough.>>101777544calm down, I'm not going to stab you to death.
>>101777445Yea Dalle was able to create some simple concepts but the stupid filter they put on it always got in the way especially anything photorealistic
>>101777548shut the fuck up nigger, stop replying to me, why did you even start? I don't know why the fuck you're talking to me, take your schizo pills.Why did you reply to my >>101777326???
>>101777558Oh wow, you don't care that you're going to FLOSS hell?
noob question, is it possible to run this with an integrated gpu?
Fuck this took way too long
>>101777523I only have 1 or two without the blur saved and I'm not making any more pics tonight. I was trying to make it look like a real photo, which is why it's blurry and reflective
The power of infinite possibility at your fingertips, yet you still choose to gen generic 1girls and cunny.
>>101777594Making it look like a scan is more fun imo
>>101777601The power of infinite possibilities of cunny is there too (flux can gen realistic ones though, so be careful and always add "anime" or something else, I've been scarred a few times)
>>101777601Have you found a way to do stuff like this?
According to ComfyUI blog poorfags can have better Schnell quality by merging it with dev model. Here's the workflow.Enable Clip and VAE nodes if you want it embedded.You can also try custom clip_l model from https://civitai.com/models/618792?modelVersionId=698764
>>101777611I was told otherwise
>>101777634Anon it's quite bad, because apparently flux gens better nipples for [REDACTED] than for adult women.
the discord gang's all here.
>>101777635Interesting that hands and feet still need tuning.
>>101777633Or just download the merge from civitai without having to burn compute for no point?
>>101777632I don't know, what the fuck is that supposed to be?
>>101777646That's obviously not Flux, anon.
>>101774616>Can someone share an example workflow with high cfg and the cosine thing? I have no idea what it is.Take this anon: https://files.catbox.moe/nxz32g.pngIf you want to know more about that method, take a look at this:https://reddit.com/r/StableDiffusion/comments/1ekgiw6/heres_a_hack_to_make_flux_better_at_prompt/and this:https://reddit.com/r/StableDiffusion/comments/1emow5p/finding_the_sweet_spot_between_guidance_and_cfg/
>>101777650No gpu compute, it just write to disk.
>>101777662It still has to merge them with your CPU, dummy, why are you burning electricity?
>>101777652Art, obviously, are you some kind of weirdo?
>>101777633checked, what's this do?
>>101777666>talking about electricity shit on such threadBEGONE SATAN
>>101777669>ArtIt really is subjective, huh.
>>101777666Maybe nobody has released the upgraded schnell?>>101777700Art is, ironically, the contrapositive of adjectival reality.
>>101777709>Maybe nobody has released the upgraded schnell?https://civitai.com/models/628862/flux1-schnell-merged-with-flux1-devhttps://civitai.com/models/618792/nepotism-fux-devschnell
>>101777639men.
>>101777754Sadly no, [REDACTED] is not men
>>101777112
>>101777747>https://civitai.com/models/628862/flux1-schnell-merged-with-flux1-dev>>101777666checked, thenSo why doesn't comfy's fp8 work right with the negative example? The vae is apparently wrong when used with those new flux cliptext
>>101777572nice
>>101777758ahhhhh trans-men
>>101777768
Models should NOT be trained on trans-men.
>>101777709
>>101777815Oh, I am sorry, is that supposed to mean something, where is his 10ft cock?
>>101777790Prove it
MODS
>>101777823you probably lost it up your ass
>>101777832https://paste.debian.net/plainh/d8c16d78 try doing this prompt with dev, anon. Purely for scientific reserach. You'll see why I think spergs will sperg about Flux a lot soon.
>>101777832
>>101777823Wrapped around his waist. That's not belly fat.
>>101777850No.
>>101777850AAAAAAAAIIIIIIIIIIIIE HF DEV DEMO DOESNT HAVE NSFW CHECKER I GENNED IThttps://huggingface.co/spaces/black-forest-labs/FLUX.1-dev
>>101777850
>>101777747>flux1-schnell-merged-with-flux1-devwhat the actual fuck
>>101777879nice knowing you anon
>>101777850>>101777879ermmm i just generated an illegal picture, sending it to every journalist who has covered AI or Stable Diffusion in the past.
Wait, so does that mean that p*dos can now finetune their dream model based on Flux?
>>101777883
ahhhhh I typed "apple" and my computer is showing apples, how did this happen to me, it's someone else's fault!!!1
FBI agents in overdrive. I just want to generate big breasted anime girls and maybe some bbw stuff. Please leave me alone.
>>101777949anon, you might find this funny, but the german government (BFL is a German company) and media will not.
>>101777951I just want a model that can gen pubes
>>101777850can it do legal cunny (2D cunny)?
>>101777978Unironically no, it's much better at 3D than 2D, take that as you would.
>>101777957If you solicit someone to draw illegal artwork, it's your fault before anyone else.
>>101777998That's not how it works, anonie
>>101778001Sure it does. Say the other person refuses. You've still solicited.
>>101778001That’s literally how it does work though
liberals can't into personal responsibility, the newsletter and blog
>>101778017so that's why SD had to neuter their models completely, yeah?
>>101778027SD didn't have to do that they're just retarded.
>>101778037The necessity of personal responsibility, it is always lurking. Conservatism always wins.
>>101778027SD had to do that because their investors told them to
>>101778027SAI is a cuck company, nothing new in the sun, when Runaway released the uncensored SD1.5, SAI wanted to take them down
>>101778011unless the solicitor is a three letter agent trying to fuck you over
>>101778046So you think BFL investors won't tell them that? I'll email a16z (they invested 16 million into BFL) some illegal flux dev gens, they'll surely like that, yeah?
>>101778055You're going to send WHAT?
>>101778070yeah
https://www.404media.co/a16z-funded-ai-platform-generated-images-that-could-be-categorized-as-child-pornography-leaked-documents-show/oh a16z are pedophiles already apparently
Am I the only person who never opens mail from strangers, if there are attachments?
>>101778086I can just send images as links instead of attachments
>>101778080>leaked-documents-showleaked documents? DOCUMENTS? emails, surely.btw, Conservatives always win everything. Personal responsibility is an innately necessary feature of human existence. It was all completely inevitable.But you know we trust parents to care for children, right?
Why doesn't the Guidance value have any effect on Schnell? You can make it anything and the image doesn't change, seems to be Dev only.
>>101778055BFL investors != SAI investors
>>101778080>journalist hit piece calling unrealistic drawings CP
>>101778110because schnell doesn't support guidance
>>101778117>>101778110holy fuck, Schnell fucking sucks then, Guidance and CFG are really important, I really hope the community won't focus on this shit and go for dev instead
>>101778117Yeah but why? I didn't think they were architecturally that different except for Schnell being distilled/turbo.
>>101778125dev has a non-commercial license so that civitai "influencers" won't like creating loras for it because they can't get paid for it
>>101778132if I remember well, schnell loras can work on dev, if that's true then it's a non-issue, I'm more worried about actual finetunes
>>101778125All the loras so far on civitai have been for Schnell, even the ones that were posted in the Dev category were just retards misclicking and turned out to be for Schnell when I tried them.I don't see what the point is, since Schnell isn't much better than a quality SDXL finetune (while being hugely inferior to SDXL in terms of tooling). Only Dev is worth using, compute spent on training Schnell is a total waste.
>>101778152>Only Dev is worth usingfp8 is very nice.
>>101778152they fucked us with the licence, schenell is Apache 2.0, meaning they can make money out of their work, it's not the case at all for dev
why do people care about licenses
>>101778080>girl and dog, short girl, pimp, slut, petite girl, potty, vulva, very young, orgasm, nsfw, lascivious, lewd pose, interspecies, zoophilia, sex with dogwow based
>>101778181...
>>101778173>>101778152Perhaps schnell loras can be moved over to dev loras as kind of a workaround for the license? Like finetune all the stuff on schnell then devify it?
>>101778181because finetuning a model is really expensive, ponyV6 cost tens of thousands of dollars for example, and that was for a small 3.5b SDXL model, imagine the cost of a "actual good" finetune on a 12b model now?
>>101778152People seem scared of the non commercial license but in reality all that means is that you can't sell access to it (or at least not without an agreement with Black forest labs)
>>101778192that doesn't answer my question at all.why not just do it, who cares? we're literal art pirates and everyone used a NAI finetune(model pirates).
>>101778181I have no fucking idea. Somewhere a long the line the pony guy is making some sort of money off of this?
>>101778192This is why (despite Stability being garbo for all the well-known reasons) we should root for the updated version of SD3 2B to be good.I have a 3090 and can run Flux Dev just fine, but that's useless to me if it doesn't know styles and nobody's training on it to fix that.
>>101778212>why not just do it, who cares? we're literal art pirates and everyone used a NAI finetune(model pirates).so you're asking a finetune guy to ask publicly for some donations after he made a dev finetune and expect the company to not notice it and sue his ass? you live in lala land anon?
>>101778259So you took the prompt from >>101777850 but made it SFW and older female? hmm
>>101778227NAi didn't sue anyoneI haven't heard of anyone getting sued for this
>>101778227We don't need any doomniggers, BFL won't sue anyone, we can easily finance $100k finetunes bro
>>101778271NAI is not this company, if you want to risk it that's on you anon
>>101778264yes lmao
>>101778227Asking for donations wouldn’t violate a noncommercial license, a noncommercial is explicitly for selling direct access.
What we need is a coomer version of Elon Musk who can use their riches to finetune and release whatever the fuck they want.
>>101778296>Asking for donations wouldn’t violate a noncommercial licenseit will, read the licence of dev again
>>101777782i liked this one more but it didn't do the ps5 box art for some reason
>gen times suddenly went from 5.9 s/t to 8.5 with all parameters the samewhy tho?
>>101778167well there is dev-fp8...
>>101778350Using which?I think fp8 is less intelligent, maybe. Investigating...fp8 looks good though
>>101778301It's not even about cooming man. Do you think we'll ever get a model with just every style and character crammed into it? A model that understand camera lenses and all the visual effects? A model that can draw people and their emotions and poses even better than Dalle?Finetuning a model like that for porn is whatever, porn loras are more than any coomer autist would need. It's the other stuff that is way more important but we seem to be moving further and further away from.
>>101778376eh...6 out of 8 having box art isn't too bad
>>101778428ろ!
is the dev vae slow?
>>101778457experiment for yourself with different VAEs/Clip/Model see what works best for you
>>101777556they all have so much personality
>>101778485which vae work with dev?
>>101778503All right that’s the best flux realism gen I’ve seen, I need to give it a try in the morning
I'm stuck at vaedecode. Generation went faster than I expected. I have a preview, so I know it genned.
>>101778513there's only one single vae that both dev and schnell use, you don't get a choice
>>101778533then why the fuck did you tell me experiment with different vae
>>101778552Dumbass
guys I'm demoralized
>>101778513i just use one called AE
>>101778533I didn't, take your meds
>>101778561have a 'za
>>101778561what are the tires for?>>101778573prompt/box?
>>101778564Do I need to tell comfy somehow to unload something while doing vae? or is that not possible? I have 32gb, and 4gb are free, while it's running Sampler CustomAdvanced.
>>101778533>>101778552But there is a different one, in the vae folder on the huggingface named diffusion_pytorch_model.safetensorsFrom testing it seems to give the exact same results as the ae one, so I'm not sure why it's smaller.
since there is no negative prompt on flux, this should work right?
There's something here but I need to explore
>>10177860432gb of what?if regular ram this thing will eat up whatever it can
>fags complain about lack of vrammeanwhile you can just generate flux pro for freehttps://glif.app/@Philip-ec12ac7063/glifs/clzj0ocd5000gnhgx4ciiw97i
>>101778580I have no idea, they were already there when I moved in
>>101778622not bad desu and checked
>>101778621yeah with cfg at 1.0 you connect the positive prompt to both inputs, since the negative will be ignored and the node will bitch if it's unconnected
>>101778637I do not want my information and prompts anywhere but on my own PC
>>101778639he's literally me
>>101778649
>>101778637because this is local diffusion general?also those pictures on that site are horrible
>>101778649you should buy a light bulb with a warmer color temperature for your living room, if so
>>101778645thank you, is there anyway to bypass the flux watermark? I read that the vae added an invisible watermark
>>101778627Yeah clearly the vae issue is just a system ram shortage.
>>101778667caught mine right in VAEwould like to see some 64GB fags show theirs
>>101778428Ultra cute
Art style kind of swings depending on the subject
>>101778665Send it throught your favorite sdxl slop at like 0.01 denoise
If flux has an xbox hueg dataset but we can't quite access it because of how it was tagged, wouldn't that make Flux a good candidate for textual inversions and embeddings?
>>101778735great composition but they all look so fucking deep fried/overcookedit looks like when you crank the CFG up to 12 on sd1.5/sdxl
>>101778770It's pissing me off desu, I'm struggling to reduce it with this model
>>101778751yeah, that's my understanding of what embeddings are best atthey don't teach new information, but they help the model find things that it already knows how to do but doesn't have a name forwhich seems ideal for Flux since it clearly CAN do painterly art, it's just tricky to access that part of it
>>101778637But what does that have to do with local models?
>>101778798I wonder if anyone has even had that thought yet.
>>101778737
>>101778637Has anyone tried using pro to gen... you know...
>>101778798>>101778826somethnig about setting guidance to 1.2 will do better paintings
>>101778840>Have to sign in through google first
>>101778840you think pro has realistic nipples and anuses?haven't figured out how to see what the vagina looks like yet
what's the story on the red cross on white backgroundit's like super copyrighted
>>101778901>>101778923that's Jansen who decided to not increase the VRAM, he's a chink not a jew kek
>>101778955so will 5090 have only 24GB VRAM?
Do you think flux needs a alternative method of handling CFG?
kek
>>101778955Jensen*>Huang was born in Taiwan in 1963, but when he was nine he and his older brother were sent as unaccompanied minors to the U.S. They landed in Tacoma, Washington, to live with an uncle, before being sent to the Oneida Baptist Institute, in Kentucky, which Huang’s uncle believed was a prestigious boarding school. In fact, it was a religious reform academy. Huang was placed with a seventeen-year-old roommate. On their first night together, the older boy lifted his shirt to show Huang the numerous places where he’d been stabbed in fights. “Every student smoked, and I think I was the only boy at the school without a pocketknife,” Huang told me. His roommate was illiterate; in exchange for teaching him to read, Huang said, “he taught me how to bench-press. I ended up doing a hundred pushups every night before bed.”That never happened.
>>101778991Now add "a photo of"
>>101778971>mikuew
>>101778971>24gb?>what do you need 20gb for?>fine take your 16gb
>>101779009
>>101779003
>>101778982All I know is fucking with the cfg makes it 3x slower. Until someone figures that out, we are stuck with normal g
>>101779024dat fro hat do
What is the best way to "capture" the features of a character that has been generated for subsequent images? I want to be able to create a series of art works that contain a consistent character in different contexts. Something like:Step 1. Generate 100 of images of the character and their features, pick the best oneStep 2. Consistently get the likeness/features of that character in generated images afterwards using OpenPose, etc... Or is there a better way to do this? Currently using Automatic1111
>>101778971only speculation at this point, but my guess is$6666 for 5090 Ti - 26 GB VRAM$9999 for 5090 Ti Super - 26.5 GB VRAM
>>101777880I am aware, I was simply asking him a question, you faggoty pass user with no reading comprehension.
>>101778971Correct. It needs to be made perfectly clear. Nvida nor AMD will ever go beyond 24~ishgb of VRAM, in fact, if they thought they could get away with it they'd probably cap it at 12.Why? Every company in the world who touches machine learning is required to make huge investments in buying their 80GB enterprise GPUs. If companies had an option to buy a couple of gamer GPUs for the same price and save a few million dollars in the process, they'd take that option.That's why there isn't even Nvlink on the new 4090 cards. They literally removed that feature to circumvent companies from buying consumer options in favor of their enterprise GPUs.And don't think AMD will save you. The CEOs are related by blood. AMD only produces GPUs for one reason, and that is as controlled opposition to Nvidia to circumvent anti monopoly laws. Don't believe me? AMD just cease and desisted Zluda the open source AMD solution to CUDA. It's all a fucking scam and they should be under the same magnifying glass google is currently under. I hope the company gets its asshole skewered in the coming years for the shit they've done.
>>101779053>>101779072its over
>>101779072>Correct. It needs to be made perfectly clear. Nvida nor AMD will ever go beyond 24~ishgb of VRAMetc
>>101779095meant to say:ThinkSystem NVIDIA L40 48GB PCIe Gen4 Passive GPU
>>101778971Great gen. Freestyler!
>>101779072google lost btwhttps://www.reuters.com/legal/us-judge-rules-google-broke-antitrust-law-search-case-2024-08-05/
That's insane how well this is workinghttps://files.catbox.moe/kqaf0y.png
>>101779111Woke2Broke speedrun
>>101779118you can negative prompt nigger
>Be official UI>Quit>"I'm putting together a team">For a TypeScript rewrite>You can do any experiment you like as long as it's implemented and the same as everyone else
getting close to time for a new thread lets not repeat the mistakes last timeyeah you
>>101779047First Ipadapter, then you train a Lora with those. Then you discard everything and use only images with the Lora. The character is the Lora, everything before really doesn’t matter as no one but you knows how it looked before.
frida kahlo is the greg rutkowski of Fluxasserting that she was the painter of an image seems to produce a lot of nice stuff without wrangling
kek, I didn't know Flux could do this
>>101779218kek whats the prompt pls?
it wasn't too difficult to make it put the prompt in the filenames if anyone is interested
>>101779054lol
>>101779199>frida kahloI don't have too many opinions of famous artists or their art style but for some reason Frida Kahlo absolutely pisses me off. It's probably the stupid ugly unibrow bitch.
>this is what my GPU looks like while genning
A higher CFG helps Flux make celebrities look more like themselves, nice.https://files.catbox.moe/wthai4.png
someone make a new thread before my gpu catches fire
>>101779330
painting kinda?
>>101779325Change the red colors to blue. That way its kalm
>>101779197Thank you anon
>>101779378Same prompt different seed....
ASS
>>101779118your schizo negative prompt doesnt work, try using something more tangible, i.e. a photo of a woman with blonde hair and add "wavy hair, curly hair" to the negative, if you get wavy hair, it means your negative prompt isnt working
>>101779493did you tag the wrong post
>>101779442bro's gotta get that dislocated shoulder checked out
>>101779506no, i tagged it right, dude has been posting his schizo workflows all night, its all comfys fault btw by posting that dishonest blogpost
Well....hopefully a finetune that has better consistency in art or loras can make this better.Back to the shelf it goes
how do I get wide shots and not just the top 2/3rds of the focus target
>he needs his hand held by finetunes and lorasto no ones surprise
>>101779493>your schizo negative prompt doesnt work, try using something more tangible, i.e. a photo of a woman with blonde hair and add "wavy hair, curly hair" to the negative, if you get wavy hair, it means your negative prompt isnt workinghttps://files.catbox.moe/7krrf6.pngAnything else?
>>101779325>CPU temp 100℃Shit's on fire yo
cozy thread
If you are having trouble with flux, just remember
did I hallucinate that there was a new bake and it got nuked
>>101779798>fagoot
>>101779631i wasn't joking..still no one has made another threadthey want me to burn up
>>101779813Thread schizo made it for one of his mfw post
Do you think is very realism?
cheer up lady
Is anyone baking because if not I'm gonna
Here it is people...>>101779850>>101779850>>101779850
>>101779830oh was it debo? I think he finally figured out that this thread was made to escape people like him and purple witch, so he wants to ruin it
