Previous /sdg/ thread : >>101738090>Beginner UI local installEasyDiffusion: https://easydiffusion.github.ioFooocus: https://github.com/lllyasviel/fooocusMetastable: https://metastable.studio>Local installAutomatic1111: https://github.com/automatic1111/stable-diffusion-webuiComfyUI: https://github.com/comfyanonymous/ComfyUISD.Next: https://github.com/vladmandic/automaticAMD GPU: https://rentry.org/sdg-link#amd-gpuIntel GPU: https://rentry.org/sdg-link#intel-gpu>Use a VAE if your images look washed outhttps://rentry.org/sdvae>Run cloud hosted instancehttps://rentry.org/sdg-link#run-cloud-hosted-instance>Try online without registrationsd3: https://huggingface.co/spaces/stabilityai/stable-diffusion-3-mediumtxt2img: https://www.mage.spaceimg2img: https://huggingface.co/spaces/huggingface/diffuse-the-rest>Models, LoRAs & upscalinghttps://civitai.comhttps://huggingface.cohttps://aitracker.arthttps://openmodeldb.info>Black Forest Labs: Fluxhttps://huggingface.co/black-forest-labs/FLUX.1-schnellhttps://huggingface.co/spaces/black-forest-labs/FLUX.1-schnellhttps://comfyanonymous.github.io/ComfyUI_examples/flux>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 news08/05/2024>Dimensional Latent Perlin for ComfyUIhttps://github.com/NeuralSamurAI/ComfyUI-Dimensional-Latent-Perlin>Arc2Face ComfyUI Node Libraryhttps://github.com/caleboleary/ComfyUI-Arc2Face>OpenAI co-founder leaves for Anthropic, Others departhttps://techcrunch.com/2024/08/05/openai-co-founder-leaves-for-anthropic/>CogVideoX-2b: 清影 Video Modelhttps://huggingface.co/THUDM/CogVideoX-2b>CogVideoX ComfyUI Wrapperhttps://github.com/kijai/ComfyUI-CogVideoXWrapper>Flux-Magic: LLM-Powered Image Generation with Flexible Optionshttps://github.com/ahgsql/flux-magic>VAR-CLIP: Text-to-Image Generator with Visual Auto-Regressive Modelinghttps://github.com/daixiangzi/VAR-CLIP>AI chip startup Groq lands $640M to challenge Nvidiahttps://techcrunch.com/2024/08/05/ai-chip-startup-groq-lands-640m-to-challenge-nvidia>Flux Dev & Schnell Support Added to Pallaidiumhttps://github.com/tin2tin/Pallaidium#change-log>AIマンガ「LearningToon」無料公開:社会人向け縦読み学習https://prtimes.jp/main/html/rd/p/000000011.000138218.html>Proof of Concept LoRA on Flux Schnellhttps://x.com/ostrisai/status/182046267423005932808/04/2024>SimpleTuner v0.9.8 - flux 24 gig training has entered the chathttps://github.com/bghira/SimpleTuner/discussions/639>FastSD CPU v1.0.0-beta.35: Adds Aura SR v2 supporthttps://github.com/rupeshs/fastsdcpu/releases/tag/v1.0.0-beta.35>SimpleTuner now supports Flux.1 training (LoRA, full)https://github.com/bghira/SimpleTuner#flux1>Civitai Model Manager: CLI tool for managing AI models from CivitAIhttps://github.com/regiellis/civitai_model_manager>Skimmed_CFG: Powerful ComfyUI anti-burn allowing much higher CFGhttps://github.com/Extraltodeus/Skimmed_CFG>ComfyUI-AdvancedLivePortraithttps://github.com/PowerHouseMan/ComfyUI-AdvancedLivePortrait08/03/2024>ComfyUI/Forge Implementation of Smoothed Energy Guidancehttps://github.com/pamparamm/sd-perturbed-attention
>mfw Research news08/05/2024>Conditional LoRA Parameter Generationhttps://arxiv.org/abs/2408.01415>Transformers are Universal In-context Learnershttps://arxiv.org/abs/2408.01367>Hallu-PI: Evaluating Hallucination in Multi-modal LLMs within Perturbed Inputshttps://arxiv.org/abs/2408.01355>TexGen: Text-Guided 3D Texture Generation with Multi-view Sampling and Resamplinghttps://dong-huo.github.io/TexGen>General Framework to Boost 3D GS Initialization for Text-to-3D Generation by Lexical Richnesshttps://arxiv.org/abs/2408.01269>S2TD-Face: Reconstruct a Detailed 3D Face with Controllable Texture from a Single Sketchhttps://arxiv.org/abs/2408.01218>Interpreting Global Perturbation Robustness of Image Models using Axiomatic Spectral Importance Decompositionhttps://arxiv.org/abs/2408.01139>CoLoRA: Contribution-based Low-Rank Adaptation with Pre-training Model for Real Image Restorationhttps://janeyeon.github.io/colora/>EIUP: A Training-Free Approach to Erase Non-Compliant Concepts Conditioned on Implicit Unsafe Promptshttps://arxiv.org/abs/2408.01014>FBSDiff: Plug-and-Play Frequency Band Substitution of Diffusion Features for Highly Controllable Text-Driven Image Translationhttps://arxiv.org/abs/2408.00998>Reclaiming Residual Knowledge: A Novel Paradigm to Low-Bit Quantizationhttps://arxiv.org/abs/2408.00923>Toward Automatic Relevance Judgment using Vision--Language Models for Image--Text Retrieval Evaluationhttps://arxiv.org/abs/2408.0136308/04/2024>Comprehensive Survey of Complex-Valued Neural Networks: Insights into Backpropagation and Activation Functionshttps://arxiv.org/abs/2407.19258>Detached and Interactive Multimodal Learninghttps://arxiv.org/abs/2407.19514>Take A Step Back: Rethinking the Two Stages in Visual Reasoninghttps://arxiv.org/abs/2407.19666>ActivityCLIP: Enhancing Group Activity Recognition by Mining Complementary Information from Text to Supplement Image Modalityhttps://arxiv.org/abs/2407.19820
I sure hope we get anime models on the level of Flux soon.I would love to have more complex scenes like this and have them make sense.
>>101745004anime website>>90482933>>88149343>>84755745>>74846612
>>101744979thanks anon for using my gen
how crunchy is the install process for flux?
>>101745080comfy put out an all-in-one modelhttps://comfyanonymous.github.io/ComfyUI_examples/flux/#flux-dev-1
>>101745080if you're using python you need to use the git interface.if you're using comfy just download the srt and the ae
1.5 still king
>>101745031Lol. Imagine putting in as much effort as the last guy only to be btfo by jannies in the end.
>>101745101Still putting in a 12 hour shift making absolute garbage?How do you manage to fuck up hands even on lux?
>>101745101>>101745102sick, ty
>>101745136Cut him some slack. He doesn't try to get better, he only gens to post and be a "thread celebrity".
>>101745156How can one be a celebrity when nobody wants you around?
>>101745164Good question.
>>101745136He makes shit gens on purpose to farm (you)s, thanks for playing newfriend
>>101745164That's why anon said celebrity in quotations.
wait is g cursed with the log spammer? is that where he learns to inseam go figure
>>101745136>>101745156>>101745164>>101745172>>101745173https://suno.com/song/71b244e0-6a38-467f-9471-f49192cb3701
>>101745172But for 15 hours a day?
desu i never went to g as it came off as just linux fags posting in other boards but this thread topic has been kino
Not clicking that but I assume it's a song written in the third person about how great he thinks he is.
>>101745143>>101745101really bad idea, the text encoder (t5xxl) is also into its 8bit version, you'll get shit gens your model won't understand shit, it should always be at fp16
>>101745197Still trying to get bmp and fran attention I see
Burbujas de cristal
>>101745223your mask must be crazy i dont want to see it, well two masks yeah
>>101745223Another variant
>>101745215no one converts the clips, only fluxwho converts clips?
>>101745218its sad to think about how many posters he pushed away but then i close the thread and forget about him
>>101745250everything is converted on that file, clip_l, t5xxl and flux, you should never quantize the text encoders that's a golden rule
>>101745275no most convert flux to fp8 but leave clip alone, even the lazy default workflow doesn't reduce clip
dear jannies, I triple checked and they ALL have tops on
>>101745264He made a suno song about BMP ignoring him. Sad is an understatement, at this point it's stalker territory now.
>>101745288that's not possible to have 16gb at the end, flux fp8 is 11gb and t5xxl fp16 is 9gb, you can see on the civitai that they quantized everythinghttps://civitai.com/models/623224/flux1s-16gb?modelVersionId=696732>Conversion of Unet to Checkpoint including T5 fp8, Clip L and VAE, which gives a model of 16GB.
How about you knobs learn how to ignore debo already and filter him?
>>101745290for you
>>101745304that was a good onehttps://suno.com/song/2eb1d12b-2241-4f0c-9c01-46b6d85eda2eI still think BMP and I can be friends
low effort trolling like this should be bannable: >>101745088
Is it me or hires fix and upscaling are not viable in Flux? It takes so damn long to run 25 steps on my RTX 3090.I'm using Flux Dev, by the way. Is it better in Schnell?
>>101745338guy was using a 256x256 he should be banned as no model can work with that latent spacebut sure i should be banned
>>101745346Schnell has lower quality but requires less steps so it should gen faster than dev. If you're fine with your current speed I'd stick with dev.
>>101745346>Is it better in Schnell?yeah, schnell is a more distilled version designed for fast convergence so its generally lower quality but you can run it at 4 sample steps
>>101745373schnell is the version that filters most words, it cannot gen even 80% of dev
>>101745332>I still think BMP and I can be friendsThat’s just sad dude
>>101745356It was genned at 1024x1024 and downscaled. You either know this and are pretending to be stupid, or you are stupid.
>>101745373>>101745372But do you guys use Upscaling and Hires Fix? I can gen one image in 40s without them. But it takes 10~15min, if I use a workflow with upscaling.
neat, it can even make japanese characters.
>>101745373>Blah blah blah I'm the thread schizo here's some regurgitated shit I read on Twitter
>>101745395or i dont know who you are as i said many times this is day two of my genningas if i'd know some random on 4chan when i'm writing code to submit
>>101745393I'd rather live my life looking to be friendly than filling every day with hate :)
>>101745346try fp8_e5m2 weight for the main flux file, and leave fp16 for clipname 1. fp16 weight takes longer I find, but im only on 16gb VRAM (4080)still, with these settings it's pretty quick.
post the pastebin kek he has no rebuttle
>>101745402it can do jewish characters aswell :^)
>>101745433fp8_e5m2 lets the model load in a 4090 easy and you have about 3GB for other apps
reminder open source always wins>no bing rejecting prompts>no paying for tokens
>>101745433>try fp8_e5m2 weight for the main flux fileI thought fp8_e4 was better?
>>101745473not for memory
>>101745473some people posting have said e5m2 is closer to fp16 results, but you'd have to generate with both to confirm.
>>101745428Being friendly doesn’t mean you need to actively look to be friends with others, much less on 4chan of all fucking places
>>101745482it is better i paid the price in heat
>>101745482>e5m2sorry I had to do it kek
>>101745482which one of these uses less memory?
>>101745304Kek did he really?
>>101745509i do not register quake 2 as a real game
https://suno.com/song/cb497707-9d01-4c60-87d3-a1054ea12975I kinda like it
>>101745511tp8_e4m3fn is the best for consumer grade
>>101745517Be friends with Debo already
the text encoding is what makes flux special, not just the prompt understanding. I can make nice girls with sdxl/ponyxl and loras, but I can't make signs or text easily or consistently.
>>101745552the CLIP is very nice on it, just talk english no tags and it works
>>101745552the anatomy is nice too, there's not a lot of fucks ups, the hands are good, the poses are good, you don't notice any weird stuff when you look at the body of people during a scene
>>101745552
>>101745584the anime mode of flux is crazy i literally can write prompts with random and it gives what i want like the last one.It is crazy good that CLIP
>>101745552>>101745466>>101745402>>101745373boring movie poster gens all look the same, training models with so many synthetic data cannot be good
>>101745618>movie posterthey're supposed to be 90s video game magazine ads. like, powerglove type shit
>>101745534I would rather try to clear Control last difficultly on Goldeneye 64, which I have putting off for months now. If Deffo came over and cleared it for me I might consider it
Miku Hatsune appears on a piece of 1940 war propaganda. Miku is holding up some vegetables. Text that says "When you don't eat vegetables, HITLER wins!". The text "po-pi-po" is at the bottom.im dying, what a time to be alive. it missed a "you" but it's still good.
its over
>>101745660
>>10174569220 so a pony model?
>>101745753BMP NO
you heard her /sdg/
>>101745632thats worse then, what i dont like about flux, is the little control you have, so you get very similar gens, no matter the prompt, SD mogs flux in aesthetic, while flux mogs SD in hands and text
fun i guess no one giving prompts anymore this thread is boring as fuck now
>>101745772we are on day 2, img2img workflows already exist, loras can be made, give people time to make cool shit
>>101745782That means less than you think it does.
>>101745780>no one giving prompts anymoreNo one asks anymore.
>>101745552>he doesn't know those are the same text encoders with SD3 uses
>>101745782day 6*,aka 2 weeks in AI development, with SD at the same time length already had dreamboothsome anons are getting to overhyped for something they have little control of
it even knows what a "ww2 era Ford" is
>>101745854>day 6*,aka 2 weeks in AI development,still a lot less than 2 whole year of SD models
>>101745860even better:
>>101745830i know how to do it ;)
>>101745860and it did draw that car perfectly, this model is also insane at overall architectures, SD models always put some random shit in it it's uncanny as fuck, there's no way I can go back to SDXL or whatever now, I've tasted the forbidden fruit, I fucking hope there will be some serious finetunes out of it
and the infamous propaganda poster:
>>101745928
>>101745866meh, what a newfag>>101745894also it totally missed the 40s propaganda poster aestheticfuck flux, and fuck you all
>>101745951>fuck you allD:
>>101745753oh bmp, you stupid bitch
>>101745951
>>101745995the lack of proper documentation, real cfg, negative prompt, makes all gens look similar how can people be so blind, almost all gens have the same composition, the only good things about flux are hands and textso keep genning the same flux gens over and over, eventually you will get bored :)
Good evening, anons! I hope everyone is doing well :]
For the life of me I can't get the lipstick mark to look like a lipstick mark and not a cartoon blob. One thing 1.5 will seemingly always struggle with is makeup.
>>101746021>the only good things about flux are hands and textyou forgot:>great anatomy>can do multiple resolutions without duplications like on SD models>great at house and car architectures>insane realism, some gens could easily fool people into believing they're real picturesdon't argue in bad faith please
>>101746022hi pw, happy weekend. awesome style
>>101746041why are you putting the model name in the prompt? put in the filename like a normal ass nigga
>>101746021>eventually you will get bored :)you think we'll magically go back to SD3 after having tried flux or something? AHAHAHAHAHAHAHAH
Miku Hatsune appears on a 1950 comic book. Miku is in a green ballroom dress, holding up some carrots. Text at the top of the comic says "Miku Hatsune" in bold text. The top right part of the image has a Marvel comics logo. She is standing in front of a building in New York.this is why flux is fun, creativity and text which is a big part of that. the prompt understanding is essential, but without clarity it means nothing.
No nipples jannies
>>101746045I thought it would be fun to make 90s video game magazine style advertisements for flux. I was right
>>101746054>but those aren't carrots!
funny
>>101746072hmmmm, i'm not so sure.
flux supports small resolutions...
>>101746041Heya Debo! Thanks so much! You as well :]I'm glad the week is over haha i'm so tired
>>101746021>makes all gens look similarwhat's this then? >>101746054
i did not realize how realize the flux CLIP let me shit postit is a crazy power only andy had kek
>>101746091Which you can then upscale using flux
>>101746054too bad it's cropped on the top, that's a nice gen anon
>>101746103rendering at 256 common you actin real
i'm extremely jealous of the fluxchads. god i wish that were me.
>>101746119no one will render your eris desire and i wont waste the heat
>>101746103now show the output of base 1024x1024 with the same prompt and seed
>>101746114What?
>>101746126everything you said makes perfect sense. but not to me.
>>101746021>makes all gens look similarnot as badly as pixart or stable cascade or even Dall-E in some cases. But worse than SD3 in this regard, no question.>almost all gens have the same composition, the only good things about flux are handsThese are related.
>>101746128It's different, because if you change the resolution you get a different output. I am not implying you would get something the same at high resolutions. I am saying you can gen at a very low resolution 8 times faster, and hunt for a good pic to upscale.
>>101746040>great anatomy => already solved in many SD models from 1.5 there is also controlnet>can do multiple resolutions without duplications like on SD models => What? composition is totally doable with SD, also cant you upscale a proper gen? you can also impaint you lazy fuck>great at house and car architectures => already can be done in SD, also there are many controlnet models that help with houses, architectures, anything really>insane realism, some gens could easily fool people into believing they're real pictures => thats not a feat. at all, many people are fooled even with the shittiest ai gendon't argue in bad faith please => I dont care if you think that really, im just stating FACTS
>>101746092you missed BMP. he committed suicide by janny :(>>101746119you can just use the free version on huggingface until you run out of gpu
>>101746130
wtf is up with stability matrix?ERROR: Cannot install -r requirements_versions.txt (line 10), -r requirements_versions.txt (line 11) and pydantic==1.10.15 because these package versions have conflicting dependencies.The conflict is caused by: The user requested pydantic==1.10.15 fastapi 0.104.1 depends on pydantic!=1.8, !=1.8.1, !=2.0.0, !=2.0.1, !=2.1.0, <3.0.0 and >=1.7.4 gradio 4.40.0 depends on pydantic>=2.0To fix this you could try to:1. loosen the range of package versions you've specified2. remove package versions to allow pip to attempt to solve the dependency conflictERROR: ResolutionImpossible: for help visit https://pip.pypa.io/en/latest/topics/dependency-resolution/#dealing-with-dependency-conflicts
>>101746164rude to be such
>>101746147>great anatomy => already solved in many SD models from 1.5
>>101746165if you are using flux on comfy use the standalone/portable filesif you get it with git, your auto1111/sdxl stuff may conflict with the python stuff you need for comfy, it shouldnt but idk
>>101746160that sounds like a lot of work
>>101746147>you can also impaint you lazy fuckthe fuck is this argument? you can also draw the whole picture with your pencil but we use image model because WE ARE LAZY FUCKS, you fucking retard
>>101746160LOOL what do you mean?
>>101746194I'm not using flux, this is a new error, everything worked fine a few days ago then I updated now it wont launch
>>101746049no one uses SD3, but even SD3 mogs Flux in terms of aesthetic, sadly the model is fucked up anatomical wise, if SD3 had good anatomy no one would be talking about Flux now>>101746093Doesnt look 1950 at all, its the same sharp colors, same aesthetic, repetitive composition that every flux gen has>>101746137I agree, dall-e gens look similar, which is why Flux gens also look similar, because it was trained on other gens, Flux looks like a mix between sdxl, midjourney and dalleDont believe me?, just use this prompt in flux, 9 of 10 gens will have the same composition, then use the same prompt in SD and see the difference"2024, erotic thriller poster, inspired by Tinto Brass vision, staring anya t joy, exploring lust themes and forbidden desires, captivating scenes of desire and passion, subtle hints of sensuality, provocative poses, intimate and suggestive settings, a delicate balance between light and shadow, evocative and sensual typography, focusing on the emotional and psychological aspects of eroticism, exploring the darker side of human connection and hidden desires, while maintaining a sense of elegance and refinement."Tining models with such inbred data will hinder new models more than help them
>>101746242>if SD3 had good anatomy no one would be talking about Flux nowhttps://www.youtube.com/watch?v=gdQGHwD2bHs
>>101745955Beautiful
>>101746196not entirely wrong. it outputs to webp so you have to convert it. but otherwise, you just type the prompt and click gen. don't even need an accounthttps://huggingface.co/spaces/black-forest-labs/FLUX.1-schnell>>101746208he posted nsfw on purpose
>>101746147Generate the following in SD:A potograph of a japanese woman wearing a red dress and an orange hat and green shoes sitting on the hood of a blue car with a yellow bird on her hat. She is holding her fingers up in a peace sign. The background is lava and the inside of a volcano.Tell me how many CNet steps you used, how much inpainting you did, and the total time you took to create this image. Because for me in Flux it took 60.88 seconds.
>>101746165got it to launch
>>101746261>Tell me how many CNet steps you used, how much inpainting you did, and the total time you took to create this image. Because for me in Flux it took 60.88 seconds.this, I don't know why this retard want us to go back in a time we had to spend hours fixing multiple fuck ups from the SAI outputs, flux just works and gives something coherent and good out of the box
>>101746198yeah, its an argument, AI gen is still a tool, not a definitive solution, so you still need user input to get the shit you like, sadly for you, you lazy fuckkk
>>101746254Oh LOOL
>>101746278>you lazy fuckkkhow ironic, from someone who makes pictures with an AI within minutes instead of spending 10 hours drawing his pencil, the real artists will genuinely call you a lazy fuck, and they will be right. Don't pretend you are some kind of hard working man because that's not the case at all
>>101746261no one cares about schizo item placement in a gen, thats for retarded people and has no real use valueif i wanted to do something like that i would use https://github.com/huchenlei/ComfyUI_densediffusion and mask all the items, its no biggie either
>>101746318>no one cares about schizo item placement in a genI care, fucker
>>101746092Damn bitch, you live like this?
>>101746119I'm running it on a 2060 12GB
>>101746318is Quokka Theft Auto still in stock?Quokka Kart 7?boomer shooter Doom clone Q2 or Q3?
>>101746318the text is attrocious, never comming back from flux that's for sure, SD models are history
Quokknite
>>101746356LOOOL!
nah, we're all about "same slop, different day". we don't need vram, or fancy bespoke models. we are legion. 1.5 ought to be enough for everyone.>>101746368well la-de-dah! aren't you the vram-rich niggerfaggot? here's your vram cookie, vramchad! you are super amazing for your vram capabilities! salud!p.s. i'm not even the slightest bit mad about anything. no. couldn't be me.
>>101746393A 2060 12GB is only $300
>>101746393btw it's pointless now but I never did like you
>>101746295Flux version, you can play with the prompt" image of Full-Page Catalog with prices, Macy's Christmas, Game-Consoles Quokka Catalog, Kids section, (quokka game consoles), quokkas game and cartridges . a very quokka christmas">>101746379yeah its attrocious, Flux has very good text but SD has the better aesthetics, hopefully both models will be compatible in the future, or at least flux can be made more lighter you can mix them in a workflow, i would love to fix hands and text in a SD with flux, I just feel that flux devs intentionally made the dev model really heavy for stopping people of doing this
>>101746404aww
>>101746395change that 2 to a 3 and you'll be getting a better deal
>>101746395thats 100 store-brand frozen pizzas. an entire years worth of pizza, lost, just to gen picture of anime girls
>>101746318>>101746405oops, i quotted myself wrong, this is another flux gen
hi, havent used sd in forever but I'm trying to do something, is there an extension I can use to animate the clouds in something like this image? I want them blowing really slowly over a few minutesthanks
>>101746428i could ask ani to chime in, but what's in it for me?
>>10174641114 hours so far thread schizo
>>101746428>blowing really slowly over a few minutesDebo's mom should be able to help you out there
>>101746470
>>101746484any Quokka 64 screenshots? or better yet playing on the TV in a 90s teen room?
What guindance does in Flux?
>>101746318
>>101746499and the quokka saturday morning cartoon too. it was crazy they got robin williams to voice the quokka>>101746518cfg, sort of. it affects how much the prompt affects the gen but it can be unintuitive sometimes
Can this flux thing do legos?
>>101746530I see. I'll bump the guidance a little.
>>101746541
>>101746555nice. Also castles and stuff?
мы вce дeбaты в этoт блaгocлoвeнный дeнь
>>101746569sure. it does pretty much everything but nipples
фиoлeтoвый нeгp мoжeт ecть дepьмo и yмepeть
flux just inserts a half bra or somethingdefinitely clothed
>>101746416Aaah so the real issue is Flux won't let you be an avatar faggot and your in pure cope mode. It's even more based than I thought.
>>101746564Awesome
>>101746697LOL
>>101745031>okachan, miro ! I posted my dead links again !
Are we still using automatic1111 webui or is there a new version for non computer scientists?
Any tips to unblurry backgroung in Flux gens?I tried wide depth of field but it's not working.
>>101746907look at the Beginner section in OP, those are supposed to be easier to use.
>>101746907comfyui is pretty easy just steal someone elses configuration and type in the prompt window its still a webui i guess but i think thats the easiest way to make it multiplatform
you can do so many different styles, it's pretty cool
>>101746960this one turned out pretty good
>>101746916HEEELP
>>101746969I like how her breasts look
>>101746916>>101746989You can't. It's pretty much baked in. You can sometimes gacha it but it's just something you have to accept with the model
>>101746989doesn't seem to matter what i do its blurryi do like that you can give it instructions in plain speak and it actually follows them, unfortunately that doesn't include it having a sharp, in focus, clear, legible signs, etc backgroundhopefully it will be figured out soon
>>101746969marvel also works:
>>101747063yeah
The biggest issue I have with flux is it will at times just decide to ignore specified art style or the subject.
>>101747063miku-rine:
For example it made about 10 generic naruto covers it decided to follow the prompt.
>>101747114I used to have this issue too, but it rarely if ever happens now. Just getting used to how to drive it.
>>101747147>Just getting used to how to drive it.how?
>>101747197By switching entirely to boomer prompting.
>>101744979>>Discord>6wUwtcJsr2Should I join?
>>101747222this
>>101747147how do you get it to obey art styles? I will find some art style it can do in a basic prompt, then as soon as I insert a character or details it ignores the art style.For example if I take the prompt I used for this and tell it to draw a person it just reverts to a generic anime style..
>>101747229>as soon as I insert a character or details it ignores the art style.yeah, the model has a huge bias problem, it seems to focus on the character's usual style rather than going for what you've actually asked for, it's like when you're prompting "Hatsune Miku", you're not just prompting the character, but also the style that comes with it
>>101747032>>101747011We gotta wait for LoRAs, then.
>>101747229
debiru
>>101747229Here's 10 in a row. What is your prompt?
>>101747223why not?
believe it
>>101747423naruto manga cover featuring donald trump.But then take this miku image >>101747341If I apply the same art style description to a character I described it ignores it and reverts to a generic comic book style.
>>101747496I ask again. What is your prompt?
>>101747496>naruto manga cover featuring donald trump.Wait is this your prompt?You are giving it fuck all to work with, of course it's going to do random shit. Explain the image better.
we're gonna make the space colonies great again.
>>101747747>>101747805Generally the more details I put in prompts the further it ignores art style prompts and trends towards a generic art style.
>>101748049First of all it doesn't know who yoji shinkawa is. Second it doesn't know what a risograph alcohol painting is, picrel is what I get when prompting for that. I think you fluked the Miku and the left prompt is more accurate.
>>101748101It seems to have very little knowledge of artists or art styles compared to SD, midjourney and dalle? I have been largely recycling prompts from other things, but it doesn't really appear to have many stylized options.
>>101748101Hatsune Miku forcefully overwrites nearly all style prompts unless you undo anime with the negatives hack
>>101748140Because the image it was trained on were described by a VLM, not from image tags. The VLM just describes the image. That's why you can do shit like this without naming the character or artist
>>101748178>Because the image it was trained on were described by a VLM, not from image tags. The VLM just describes the image. That's why you can do shit like this without naming the character or artistHow do you describe specific art styles to it then? If I wanted to make this dalle3 jojo beaver for example.
>>101748267Probably with a LoRA
>>101748178Why couldn't they combine natural tags and VLM at least to some extent?
Try out Flux pro, dev, schnell, limited offer:https://grain-compare-packard-anime.trycloudflare.com/
>>101748504They probably can, but didn't for "safety" sake. It supports tags just fine. They may have done exactly that, but sanitised the tags.
>>> total_rows1231173261Diffusion1B dropping real soon:tm:
>>> total_rows1231173261
>>101748143>>101748143>>101748143
>"I won."
Gmornin Anons, have a great day!
>>101748984Thx
