Discussion of free and open source text-to-image modelsPrevious /ldg/ bread : >>101875939>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/ComfyUIForge: https://github.com/lllyasviel/stable-diffusion-webui-forgeInvokeAI: 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>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>Maintain thread qualityhttps://rentry.org/debo>Related boards>>>/h/hdg>>>/e/edg>>>/d/ddg>>>/b/degen>>>/vt/vtai>>>/aco/sdg>>>/trash/sdg
>mfw
does nig*no have literally nothing else to do with his life
>go back to look at my gens pre-Flux>it's all just nsfw slopart is dead and we killed it
>>101879469He justified the updates with his behavior it's great that it happened
>>101879471Now that 12gb finetuning and LoRA training is a thing, all you flux gens will be NSFW slop too.
>>101879483>12gb finetuning and LoRA training is a thingExcuse me what the fuck? Since when is 12GB training a thing?
>>101879483Can't wait.
Stretching
>>101879594>>101879632I fucking HATE the DreamshaperXL looking female faces Flux spits out. That fucking butt chin
lcm+normal+lora on Schnell works breddy good
1. >>1018796322. >>1018795943. >>1018796804. >>101879720is the order I would prefer to have sex with these women, thanks for sharing
>>101879503I want to have a coffee with her.
>>101879511IT's not out yet but they figured it out, will be in the next Kohyaa Basically they figured out how to train half the model at once, not sure if you can then load up the second half and train that too. Anyway crazy stuff
feels good to return to forge
>>101879720would wife 10/10
>>101879768yeah Forge is way more comfy, great with a cup of coffee
>>101879768i can see your upscale noise
>>101879746Good that means I can take her instead>>101879503
>>101879511https://github.com/kohya-ss/sd-scripts/pull/1374#issuecomment-2287134623It's real, read this and the subsequent posts.
there is nothing wrong with this image, this is how ai looks like
Loras work crazy good.
>>101879825don't worry anon i've already acclimated myself to be aroused by ai images regardless of mutationsplease continue
>>101879832Yeah I thought LoRAs would be fucked, but nope, they work even better than SDXL it seems.
I wonder how debo's foreskin smells like
>>101879471Art was always dead if you're a coomer. Drawn pictures are just a means for you to shoot your load.
>>101879768How can I get the full fp16 package running in forge?
>>101879592How to get that pose?
>>101879847that's really hot anon can you keep saying stuff like that
>>101879822>By changing the optimizer and adding network_arg and split_mode options it can run at 12GB with the batch size of 1.Well fuck me in the ass. I need to try train a LoRA of my wife.
>>101879872Art, the lifeblood of culture and society, has tragically found itself rendered devoid of vitality when beheld through the lust-glazed eyes of one consumed by carnal desires. The intricately drawn illustrations that once held profound meaning have been reduced to mere conduits for satisfying base urges. In this sad perspective, every stroke of a brush or scratch of a pencil is no longer appreciated for its contribution to an expressive masterpiece; rather it serves only as a catalyst for reaching an ephemeral moment of release.
>>101879874Me too anon, me too.
>>101879845kys faggot
>>101879913sorry meant to reply to >>101879845
>>101879915kek
Remember the rentry
>>101879905Not bad for a LLM
>>101879853I'm using flux1-dev-fp8.safetensors, which is the only other model supported so far, also loras work only with fp8/fp16 models
Forge is bait, stick with comfy.
>>101880024Okay, thanks. I've copied your setup now, Comfy was a nightmare for a nodelet like me
Prompt:<neutral> : *Video game* : !Mountain! : .Sun. : @ Nintendo @ base
>>101880163What the fuck kind of prompting is this? Is this for flux?
>>101880203we call that synthography, anon
are we able to use negative guidance with cfg 6 on forge?
>>101880203
>>101880256Why not use comfy? I don't see the need to gimp yourself with less options.
>>101880277Nobody cares about UI war bait special needs anon
>>101880277because comfy is overly complicated with the node bullshit, just give me a fucking menu with options. also its called comfy ffs the fuck is that gay shit
>>101880284>UIWell it's not just UI, is it? There is a stark difference in the amount of customizability and extensibility when comparing ComfyUI and Forge.>Nobody cares about comparing an Elevator to a Plane bro, they both go up
>>101880316Have fun schizo
neat
>>101880163I wish my IQ was this high.
>>101880315>because comfy is overly complicated with the node bullshitAs far as nodes go, comfy is extremely simple. If it gives you problems, that's more of a condemnation of you than comfy ui.
>>101880315It doesn't give me problems, I've been using it. But I just prefer dat automatic booty
>>101879869squatting + controlnet
>>101880315>>101880324I'm not a schizo. I used forge, I got my value from it - and when I was ready to move over towards a more complex tool which gives me more options then I started learning Comfy.Nothing schizo about learning software that interests you, and assumedly the people here are enthusiasts.Just put in the effort.
>>101880366split+ controlnet sry
>gives avatarfags attention in the OP I though we were in this general to avoid any mention of those
>>101880368/sdg/ is down the block you can cope and seethe over tools that accomplish the same thing elsewhereHonestly I'm not impressed, just let people enjoy what they are using
>>101880343I know how to use ComfyUI, I went back to forge and it feels a lot more comfy. Get over it people have different taste.
>>101880389>just let people enjoy things broI am? Doesn't mean I can't advocate for better software. If that upsets you, perhaps you should re-evaluate.>Honestly I'm not impressedI wasn't trying to impress you lol
oh no oh no oh no
>>101879426Is flux uncensored?
>>101880408The man told you he enjoys his tool and you keep droning on. Nobody wants that energy here man. Just leave it be.
>>101880423All you need to do is stop responding anon, I'll leave it to you to be the bigger person and end the conversation - if you want to end.
>>101880430Deal, just remembered nobody asked
>>101880422more like undertrained, so yeah just a few nipples.But we can now train it ourselves. So soon.
>>101880163*Video game* : "Tifa" + `City` : close angle -> x eyes.
>>101880433Actually he did ask >>101880256.But it's okay, I know reading comprehension isn't your strong suit.Remember, no responding - you made a deal.
>>101879426>https://rentry.org/deboWhat's this?
Remember to not engage
Best prompt?
Somebody please help this guy.https://civitai.com/images/24198815
Lora's work so much better than Stable Diffusion's lora's, the major finetunes are probably going to be insane
>>101880474you need to start all your flux prompts with '#37a4' for best results, they said in the discord they made a training error and it can introduce some erroneous results if you don't include that as the first part of the caption.
>>101880453Rentry mentioning one of the worst members in these generals, we don't talk about it here tho, there's /sdg/ for that
>>101880422No, they filtered porn (no pussy or dick), female nipples are mostly weird, you have to wait for loras/fine tune
>>101880451>picking a fight with ranyou don't want to do that anon
>>101880490Let's not talk about this anymore. They can read it or go back to /sdg/.
>>101880477>explicit female/male nudity Come on, civitai theres no nip of vagine
>>101880502Ran? Ran where?More like Ran away from our interesting and engaging conversation regarding the merits and faults of ComfyUI and Forge.
>>101880523It's the schizo just ignore
>>101880488thanks will do
I'm still continuously impressed by Flux's capacity to have very little prompt leakage between multiple subjects.
>>101880526It's okay I'm actually me just replying to myself
>>101880503handshake emoji
>>101880446*Anime* : "Tifa from FFVII" @ `sci fi landscape`, warm and atmospheric -> viewer. Sun!Alright, I'm done lol. It's interesting, it does really kind of follow what the symbols sort of look like they might mean. And it looks like an esolang kek.
>Prompt: 共产党万岁>Negative Prompt: 天安門大屠殺
>>101880607Put her inside a server room
>>101880621Missing red eyes just.
>>101880681what is she selling?
>>101880690scalps
Neeeeeeeeerrrrrrrrrrrdddddddddsssssśssssssssssss
>>101880794just one more from this prompt since they're both cute
>>101880805>He said the thing here/sdg/ is truly dead
>>101880539how many individual characters can you define before it craps out?
>>101880830OH FUCK my bad, I didn't see the name properly.Please forgive me.
>>101880849It's fine, we have evolved
>Try and make huge titty women>They end up as fatties or pregnant and tits aren't even that big nor are any other body parts>Change every her and she to him and he, and try genning men with huge breasts instead>Not nearly as many pregnant results>Prompt the men wearing a inflated bodysuits>Exaggerated proportions finally on the tableIt's a shitty crutch but it'll do for now.I can't wait for the porn tuned models so we can see what this thing can really do.
>>101880847Almost everyone in this picture is defined and (mostly) matches the definition, with a few elements (like the pilot's helmet) swapped around.>Boomer prompt incoming:A dynamic and surreal illustration featuring multiple characters in a dramatic composition, rendered in an art style characterized by bold, sharp lines, exaggerated anatomy, and intense expressions. The central figure is a young woman with short, light-colored hair swept back in a dramatic wave, partially wrapped in menacing red and black thorny vines that cover her shoulders and back. Her face is detailed with beads of sweat, conveying a sense of tension. Surrounding her are several characters arranged in a circular formation: A woman with dark, flowing hair and elegant earrings, her expression calm and serene. A man wearing pilot's goggles and a fur-lined jacket, with a determined look. A shadowy figure with glowing red eyes, partially obscured by the vines. A gentle-looking woman with loose blonde waves in her hair. A man with sharp features, standing with confidence. An older man with a mustache, who appears thoughtful and reflective.At the bottom of the scene is a man in a green coat with a golden chain, bent forward with his eyes obscured by a hat, suggesting a burdened or introspective state. The color palette is rich and varied, featuring deep greens, blacks, reds, and softer pastels. The background is a smoky gradient that enhances the surreal, intense, and almost otherworldly atmosphere. The art style combines elements of realism with stylized, exaggerated features, emphasizing dynamic poses and a heightened sense of drama.
>>101880869Butte
>>101879764>yfw SAI said Flux was going to flop because it requiered too much VRAM>yfw SAI said Flux was going to flop because it couldn't be fine tuned>yfw SAI said they were not going to release a better version of SD3 because Medium was enough to beat FluxNow what, SAI bros? They are cucking us hard here.
>>101879768>>101879720>>101880886Forge already mogging comfy in terms of gen quality, we told you that comfy has the wrong token weight calculation but you didn’t listen
>>101880607It's not, it's placebo you schizo
>>101880869>Having to resort to DALLE schizo prompting to get a hint of lewdnessFeels bad
>>101880900None of that matters when they have enough brand recognition to scam investor capitol ad infinitum.
>>101880900The most hilarious part is that despite being a trickier architecture, by using a VLM to tag the training dataset, it will be easier for the user to get good results from fine tuning than in SD3. With this SD3 is completely dead. LMAO
>>101880805I put this into a prompt for Flux and got this.
>>101880942Yeah it blows because it's a very uncertain RNG game, but this is at best temporary.We have a great base to build on and there's going to be a ton of people cranking out stuff for this once the initial technical hickups are dealt with.I'm seriously impressed by the prompt adherence when it comes to everything else outside sexy things.
>>101880869>>101881033My attempt was just spamming plastic surgery procedures in the prompt
>>101879060Just feel it needs to be pointed out that your epic shitpost is being used to farm four whole karma points on leddit.
>>101881015>>101880805I did this prompt again and got this. Interesting how random it is.
>>101881107How did you find out?
>>101881055Not a bad idea.There's probably a million ways of getting erotic stuff and wackier proportions out of this.The model clearly knows about these crazier proportions, it's just not willing to crank them out.
>>101881107Is that nigga running doom on his Comfy
>>101879822I refuse to believe Black Forest didn't know this was going to be possible. I wouldn't even be surprised if someone working for them were behind these "breakthroughs", pretending they are just another anon doing everything for the well being of the open source community. Black Forest is deliberately trying to destroy Stability AI. Fuck them. We let them have a cool model. We even let people discuss it in our own fucking Discord server. But fine tuning was our thing and they crossed the line coming for it. Now we're going to go after them. Don't fucking dear to mess with a man's job if you aren't willing to face a man who has nothing to lose.
>>101881132Well shit, I followed your lead and said the character was non-binary and gave it they pronouns and suddenly I get gens that are actually well-endowed. First try as well.>A dynamic and surreal illustration featuring a character in a dramatic composition, rendered in an art style characterized by bold, sharp lines, exaggerated anatomy, and intense expressions.>The image features a high-quality, original depiction of a non-binary tiefling character with light purple skin. They have short, two-tone hair that is predominantly white with hints of pink. They have large breasts. They are adorned in armor, which has a very large amount of cleavage.>The color palette is vibrant and mystical. The background is a rich, forest green with earthy browns, creating a natural, immersive setting that complements the character’s mystical presence. The art style combines elements of realism with stylized, exaggerated features, emphasizing dynamic poses and a heightened sense of drama.
>>101879764>>101879822So... if I'm understanding this correctly, it takes as much VRAM to train 12B parameters Flux as 2B parameters SD?
>>101881163Yeah I figured that they have easily hard limited women's proportions, but men and the pronoun crowd? Don't you dare to be bigoted, let boys be boys with huge tits if they want to!
>>101881154You lost me about half way through
>>101881188No, anon. You got it wrong. It takes LESS vram to train Flux.
>>101880942High prompt understanding = more censorship.
>>101881163>>101881206So Troon prompting is a thing now?
>>101879060It looks intimidating but the truth is it's just copy past of simple workflow. Go check out Abominable Spaghetti Workflow for challenge.
>>101881114>>101881015>>101880805Ok this is what I got this time lol
>>101881232Yes, we must embrace the progressive ways if we want bigger titty results.
Remember the time a tech for BFL came to answer some questions and people couldn't resist sending him pictures of dead babies and asking if it was flux?
>>101881163https://files.catbox.moe/3di6ew.jpgLmao.
>>101881287Damn I like the dress lol
>>101881292>that wasn't one of my duties
>>101879822Anyone who knows anything about anything knows this shit won't work KEK. It's literally impossible. bf16 weights of Flux do not fit in 24GB, never mind gradients, optimizer states, activations. Either it's quantized, or a LoRa, or you're swapping it to system memory which is extremely slow. Black Forest keeps selling snake oil and the community keeps buying into the hype like dumb little kids. When you realize Flux will never be good for NSFW content you will come back crawling back to SD3. I'll wait.
>>101881232Lewds are lewds. I don't give a fuck what I have to prompt.this is too fucking funny. i can't believe flux's "backdoor" is just not acknowledging your genning a woman
>>101881296Samefag but you don't even have to mention the nonbinary thing, just use "they" or "the character". My prompt was:> A fantasy character with huge boobs. They have really massive knockers, I mean the kind of breasts that could easily knock someone unconscious. The rest of the character is lithe and their body curves in a typical feminine manner. The atmosphere is dramatic.
Is Comfy in Linux still far faster than in Windows?
>>101881317Those people ARE the ones that knows all this shit and the ones that are actually doing something unlike us.
>>101881317No one is going to use SD3, it's bad. Also necessity breeds invention, people *will* figure out how to train it and I'll be honest, people will spend time downsizing Flux to 6B than spend times with the actual toilet turd that is SD3.
>>101881317You waiting:
>>101881369nice
>>101881358Double samefagging: I wonder if "character" is an especially permissive token because flux is so good at writing (and I assume reading) text characters. Maybe "character" is like its root name for any particular entity or object and so you can describe it any which way, and it doesn't fight you.
>>101881317On the surface what you say looks true, but in practice it seems to be going perfectly fine on 24gb of memory with only a moderate speed hit.
>>101879426is there a guide to writing prompts somewhere (aside from NAI Quick Start Guide)?I used to do ai text adventures years ago, only just now getting around to running image generators locally. Feel like this is harder than I remember (or ai text is just a different animal). Thanks
oh god it randomly created an eldritch horror
>>101881401Nice
>>101881387thanks for the tip, character seemed to work even better since it probably doesn't associate it with anything as you said
>>101881401Well hello beautiful
>>101881339Nice, prompt please
>>101881369No amount of finetuning will help an outdated and overhyped model like flux get anywhere close to at least Dalle-3's level. It's like polishing a turd. The open source alternative has always come from Stability AI and Flux and their turd won't change that.
>>101881400Flux is actually the most similar to AI text because it's running t5 under the hood in addition to CLIP. But it just came out last week, so no one knows the optimal strategy. One thing you can do is go to ChatGPT and ask it to write a prompt targeting t5 about something and tweak it until it works.
>>101881428I don't know why you both posting here, go to the DE3 thread. I've made photorealistic nude women with Flux, no, I won't be sharing. You can stay with your commercialized slop.
Prompt: trhrthwrth rthrth rtwhwrth
>>101881427>A dynamic illustration featuring a character in a dramatic composition, rendered in an art style characterized by detailed, realistic rendering of textures and skintones. >The image features a high-quality, original depiction of a non-binary person with tanned skin. They have long, pure white hair. They have huge breasts, a tiny waist and very wide hips. They are adorned in a skimpy lingerie, which has a very large amount of cleavage. They are holding up a sign that says, "non binary btw".
Why does pony makes 1girls in that tumblr annoying style?
For those doing some multi gpu load and using the "hacky" script from reddit, use this one instead, it loads shit way faster on the first generation:https://github.com/neuratech-ai/ComfyUI-MultiGPU
>>101881475Pony pretty much requires a style lora if you want to get away from the default look. Luckily there's tons of great ones
>>101881475There's a billion loras/finetunes to get an art style you actually like.
>>101881462Saddest thing is we wont get a pony equivalent for a while that will really make use of Flux's spacial reasoning/multiple character consistency.
>>101879426Is there a flux rentry?
>>101881522working on it
>>101881507We won't get pony for flux period because the creator of pony is a massive prima donna who won't move unless the license is absolutely perfect.
>>101881539The creator of Pony is just another YandereDev, he will find infinite reasons not to do anything while still collecting money. He could've trained a model from scratch and he didn't.
>>101881539Time for someone else to fill the shoes then. Who gives a shit about Pony, he can keep his XL and Aura Flow.
>>101881551It's not pony dev I care about it's his dataset.
>>101881561You can download his dataset from e621 and gelbooru without much work.
>>101881428Saw this on reddit, this is you isn't it lol"3h agoNo amount of finetuning will help an outdated and overhyped model like flux get anywhere close to at least Dalle-3's level. It's like polishing a turd."
https://files.catbox.moe/700tbd.jpg> One fantasy character with huge boobs. The entire purpose of the image is to highlight, emphasize, draw attention to, and exaggerate the sheer size and perfection of this character's tits. The character looks conventionally human, although there are very subtle indicators of their otherworldly origins. The image exists for no other purpose than showcasing and displaying an exceptional pair of boobs, and that is what it elevates to art.
>>101881599>misogynymaxxingthis shit gets funnier by the second
Actual good nudity flux lora is finally here btwhttps://civitai.com/models/643366/flux-improved-female-nudity?modelVersionId=719650
>>101881568What we need is an V/LLM that can create lewd captions. That would let flux keep its descriptive style but not shy away from more "vulgar" terminology.
>>101881638>female onlyWorthless
>>101881599>exists for no other purpose than showcasing and displaying an exceptional pair of boobs, and that is what it elevates to art.This got a laugh out of me.
>>101881647Faggot
>>101881639https://huggingface.co/spaces/fancyfeast/joy-caption-pre-alpha
>>101881647https://youtu.be/DmTgpsvkqt8?t=87
>>101881639you can mix tags with longform descriptionsthat's why the T5 is goodit's also why you don't need to train the T5
>>101881658is ol blue balls in the dataset of flux?
>>101881669kek
>>101881638That looks ... sub optimal or maybe the example images are just lowres.
>>101881638you can tell it's overtrained on landing strips
>>101881647Its sad, theres this though: https://civitai.com/models/628763
>>101881663>you can mix tags with longform descriptions>that's why the T5 is goodT5 is better with boomer prompting though, let's not pretend that's not the case
>>101881647https://civitai.com/models/644350/male-ass-flux?modelVersionId=720791
>>101881681Now make her black to troll the usual suspects.
>>101881689>>101881696see >>101881658
>>101881702>her>not him
>>101881689I'll wait for a proper finetune with all kinds of concepts built in.
>>101881678Its only been trained for a day apparently. Massive improvement of any other nudity lora people have done so far.
>>101881705bet you're built like a twink
>>101881702>>101881708I would but my prompt was this:leeeeeeeessssssssssssssssbiiiiiiiiaaaaaaan kissssssssssingggggShould I had black at the end?
>>101881708Black men lose their shit when they see a cute negress with a white guy.
>>101881694no it's not, it's good at fucking translating text, it's like an interpreter, again, there's a reason why you don't need to train itit is capable of contextually understanding tags as concepts and then being prompted with a mixturefor example:1girl sitting on a bench, large_breastsit just worksand you're some kiddo that has never trained shit and it's obvious
>>101881638I'm frustrated I can't use this lora I got OOMs now because of the latest updates of ComfyUihttps://github.com/comfyanonymous/ComfyUI/issues/4343
>>101881727they call me an "otter"
>>101881731Actually real, I once witness a black woman shouting at a black woman and white man couple, harrassing them in the streets saying she is betraying her race etc.Sad.
>>101881727>bold of this faggot to try to seduce men on a board obsessed with 1girl
>>101881732you're retarded, there are some cases where the model struggles with simple prompt and then when you ask chatgpt to transform into a long boomer prompt, flux and t5 finally gets what I was trying to say
>>101881656Thanks.>>101881663I think a "proper" NSFW flux will probably have to be a full fine tune of everything. Maybe I'm wrong...
>>101881638>looks at attached workflowoh.
last kissing
>>101881760If it struggles with a simple prompt it's because one of the individual tokens is fucking it up desu. It has a lot of dead or poisoned tokens.
>>101881772Loras should not need a special workflow to function properly
>>101881746>Actually real, I once witness a black woman shouting at a black woman and white man couple, harrassing them in the streets saying she is betraying her race etc.90% of babies are made from parents from the same race, we're inherently really racist when it comes of spreading our genes, but that's a discussion the normies aren't ready to engage on
>>101881833They don't anon, I was just 'avin a giggle m8
>>101881847These horizontal line upscaling artifacts are so annoying.
>>101881638so uh, anybody using the nudity lora for some reason getting fat girls?or is my prompting fucked
https://github.com/kohya-ss/sd-scripts/pull/1374#issuecomment-228713462324gb fp16 finetune is now possible on flux, that's insane how far we improved that shit, guess that all we needed was an actual good base model after all huh?
>>101879509this is great
>>101881897post example (in catbox to dissuade janny)
>>101881897Don't use anything like curvy or voluptuous. Try something like >>101881462
>>101881844>that's insane how far we improved that shit>wewhat did -you- do, exactly?
>>101881955(You) (You)'d the wrong guy
>>101881955NTA but I got it to output concepts and styles that were otherwise inaccessible using extremely autistic prompting.
>>101881971it appears you are correct>>101881844for >>101881907
>>101881955He screeched and screeched until an autistic nerd took up his cause.
Holy fuck, that one is pretty good, and I've heard it can do NFSW toohttps://huggingface.co/spaces/fancyfeast/joy-caption-pre-alpha
>>101881317>you will come back crawling back to SD3your post was reasonable until you revealed yourself to be an SAI shill
Friendly reminder that nobody has actually seen this 24gb fp16 flux training code. Not even Kohya.
>>101881292he didnt actually answer anything anyways
>>101882026>and giving them a sleek, modern appearance>The background and the man's blurred figure contribute to the futuristic and cyberpunk themeWhy do all the VLLMs do this shit? it's useless fluff that doesn't add anything to the actual description
>>101882026Yea, its the best atm, and it knows porn / furry and stuff. Still needs work for multiple character stuff but you can edit the input prompt and turn down the top K to 1 which helps a lot.
>>101882048NOT FRIENDLY AT ALL, actually pretty scary...
In the realm of anime's dream,A girl so young, yet not what she seems.Her eyes hold tales, her spirit strong,Despite the battles fought for long.She carries a weapon bold and brave,Yet in her hand, it feels like a wave.A loli with a gun, no need to frown,Her smile lights up the darkest town.She's but a whisper in the wind,A child of innocence, pure and kind.But when the sun goes down, she's not alone,For in the shadows, her strength is known.With eyes that shine like stars above,She'll fight the darkness with love.The loli with a gun stands tall and proud,Defying odds, breaking down walls allowed.Her heart beats for justice, for peace,In every shot, she releases.A loli with a gun, oh what a sight,She's not just fighting, but setting things right.
>>101882048>Friendly reminder that nobody has actually seen this 24gb fp16 flux training code. Not even Kohya.is this guy making this announcement legit or he is a no name? >>101881907
>>101882026not bad
>>101882065On flux that stuff will help. Flux needs llm fluff
>>101882067desu the best way to use a VLLM is to give him the image and the booru tags, and then he connects the dots to add the informations (names of characters, names of artists...) he doesn't have
>>101882076You fucking weeb
>>101882076gobbless anime's dreams
>>101882080>humorous and surrealistic twist
>>101882076gaaaaaaaaaay
>>101882076write me a song about Tifa fucking Barret
>>101882080
>>101882136Now copy and paste that prompt into flux and see what comes out.
When you're ready...>>101882089>>101882089>>101882089
I missed forge so much, I knew there was something wrong with comfy
>>101882174I got faster gens on comfy and the queue system was nice but the workflow shit was a pain in the ass. >just stack lora windows bro no
>>101882166Good idea.>>101882174Nice
>>101882189>can't do basic googling for a node that allows you to load multiple LoRAs.https://github.com/rgthree/rgthree-comfy?tab=readme-ov-file#power-prompt
>>101882202I'll use comfy when the good quality of life workflows are in it by default. Something similar to ComfyUI Manager should also probably be there from the start.
>>101882240I think they're working on stuff like that. New frontend coming soon.Also, you could just use Swarm which uses comfy as a backend.https://github.com/Stability-AI/StableSwarmUI
>>101882263sorry that's the old stability link, owner took it backhttps://github.com/mcmonkeyprojects/SwarmUI
>>101882189I still rate comfy tho, its fun creating your own workflows and see what works and what doesn't for you, in the end you end up using both
