Discussion of free and open source text-to-image modelsPrevious /ldg/ bread : >>101923884>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
Bad brain here
>>101927559>>101927570>>101927581She's just really happy to see (you)>>101927580omg it (my) migu
Protip: guidance 1 is ok for schnell, but way too low for dev, and this took 20 minutes to generate
>>101927704>>101927688very nice
>>101927690>20 minutes Jesus anon, are you using a 4gb laptop gpu or something?
nerds
How do i clear these LORA loaders so comfy doesn't try to gen with them? The instructions say to just set their strengths to 0 but comfy will still try to load them anyway.
>>10192769012 steps, 4 minutes. 1.5 sd did the same in 50 seconds
>>101927721but wait, there is more.
>>101927733Right click + Bypass
>>101927750absolutely wrong lmao
>>101927730Hype
>>101927580i want to have sex with every image in this collage
>>101927758guess I was thinking of a different node
>>10192774314 steps,3 minutes for some reason
>>101927733dont use all in one node that you're using.
>>101927733Don't use shitty all-in-one loaders. There's plenty of dedicated nodes that can load as many loras as you can, with toggles.
>give prompt to SD 1.5 to see what would happen
>>101927758just slap a decent lora loader into the model chain, right after your SD1GEN thing. I recommend rgthrees "power lora loader" node. can right click on loras to grab info from civitai, can bypass loras with one click, can also seperate clip and model strength.
>>101927868How many steps you use?
>>101927806>bulge again
>>101927930That's a feminine bulge
>>101927841>>101927849>>101927892thank you
>>101927666please give prompt satan
>>10192791525
>>101928052have you used the Pokimane lora with good results?
>finally get a blowjob lora>its nigger dicks
>>101928016artstyle lora, prompt will give you the generic flux look
>>101928052For some reason my generation goes fast in the last steps, why is that?
And here comes SD 1.5 with the soul.
This is what i have so far, Now what i need is to add the nodes for saving metadata.. wheeere in this setup do i add those nodes?-from this onehttps://github.com/receyuki/comfyui-prompt-reader-node?tab=readme-ov-fileIt just seems to further confuse how i'm understanding comfyui nodes work to begin with, I already have nodes that generate the image, and save the image, but this is supposed to.. override those nodes, i think? I can't just use the SD prompt saver as the last node in the chain to both save the image and metadata to it, as i would've understood.. wtf
>>101928109Is she from Silent Hill?
>>101928048is it worth the wait? again, what are you trying to do? make it follow the prompt more or get rid of the blur?>>101928089cool beans. going to try it out later
>>101928096If you're using adaptive guider in a workflow then it's setting the CFG to 1.0 at the end and no longer taking the negative into account, thus doubling the speed of the final few steps.https://old.reddit.com/r/StableDiffusion/comments/1enxcek/improve_the_inference_speed_by_25_at_cfg_1_for/
>>101928116what's your issue exactly, the default image saver saves the whole workflow with settings and prompts included in the image
>>101928152I need A1111 type metadata, compatible with civitai/anythingnotcomfyUI
any good deepfake video workflows out there, or should i just turn the video to single frames and batch-liveportrait the pictures?
>>101928161but civitai does read, and let you copy, comfyui metadata and if you use standard nodes it can show the prompt and loras used right on the webpage
>>101927580I liek creepy landscapes.
Is everybody using flux?
>>101928188it's not standard, no it does not.https://files.catbox.moe/t7l3mr.png>imagine the filename is no_it_does_not_you_triple_nigger.png
>>101927785Even the cat?
>gen erotic images with flux>inpaint with sd>????>PROFIT
>>101928116first of all, raise your ult sd upscale mask blur to something around 16, or even 32. thank me later. also, use a "get image size" node and route the image into it after the sd&ref, and use those numbers as a tile size for the ult sd upscale. euler is also not really the best sampler for a resample. try euler a or whatever, ddim. also those genatives arent so hot, esp. "bad art". define bad art.might wanna try these for anus1111 compability https://github.com/receyuki/comfyui-prompt-reader-node
>>101928216I know how to make it work but you used a slur so I'm not helping you.
ignore and report
>>101928228I never noticed anything amiss with the upscaler settings but shit ill try it out.i should mention 99% of what you're seeing there is defaults, I didn't touch any of that. I would NEVER prompt like that EVER.>>101928238ok nigger.
>>101928128your workflow is clearly doing something. this is the trash i genned without it
https://www.youtube.com/watch?v=tned5bYOC08Finally a flux analysis by a non-retard
>>101928318>hello sirs, this is pateland closed
>>101928246
Using Flux GGUF. It was great then suddenly everything is taking twice as long. Persists through restart. wtf?
>>101928361You launched some unholy programs that gobble up more VRAM forcing more offloads.
>>101928264ok codelet, have fun being unable to make a machine work as you wish
>use quant that fits entirely in VRAM like Q4_1>4 s/it>use literally anything that doesn't fit, whether it's Q8 or full fp16>always 6 s/itwhy does it just max out at 6 even when I double the model size? not complaining but wtf
>>101928369>>101928369>>101928392>>101928408>>101928416Were these gened with flux? If yes then I have some questions
>>10192813820 minutes for this slop
Hey frens, got a couple questions if anyone is feeling generous.Coming from Voldy w/ a ton of loras/models/embeddings/styles/settings/etc: 1) If I want to use Flux should I switch to ComfyUI? Do I need to?2) Is it fine to just point ComfyUI to use the same directories for loras/models/embeddings/etc that I already have with sd-voldy?3) Seems like the dev checkpoint the right call if I've got 12gb vram (4080) and 64gb ram, but should I use schnell to find seeds quickly and then swap out to dev after? Or just go straight for the good generation?
>>101928444>>>/b/
>>101928433No - Autismmix Confetti.I just checked Civitai and there are FINALLY some anime girl LORAs for it, so I may be switching soon.It's pretty useless to me without anime girl LORAs and I've been waiting for them to switch to it.Also, bros, does Controlnet work with Flux? I use Canny too much to do without it.
>>101928222It's a bit of a slow workflow, but it works.
>>101928438Voldy can't into Flux so you have to use Comfy or Forge, there are other interfaces with Flux support but they are rarely talked about hereyou can use the same directories, Comfy has a file where you can define multiple places to find checkpoints/encoders/loras etc so no symlinking necessary thereyou have 16GB of VRAM, not 12. you want to use FP16 T5 on the CPU and either the FP8 diffusion model or one of the Q8 quants which are closer to the full FP16 in qualitythe seeds will produce different results on dev and schnell
>>101928458Well that model is pretty versatile so I guess My question still applies: when will anti ai people finally shut up about "muh ethics" and "muh 'copyright infringement' " and appreciate how powerful and useful this tech is? Your gens look nearly indistinguishable from an anime a screen cap or fanart and I bet 90% would be non the wiser it was AI if they weren't told
>>101928458>I just checked Civitai and there are FINALLY some anime girl LORAs for it, so I may be switching soon.what the fuck are you talking about, most PonyXL loras work with Autismmix and any lora for a version of Autismmix works with the other versions with little difference
>>101928504>Your gens look nearly indistinguishable from an anime a screen cap or fanart and I bet 90% would be non the wiser it was AI if they weren't toldBecause I shoop and inpaint the shit out of them kek. The original gens have a lot of AI-ish flaws.Anyways, thanks.>when will anti ai people finally shut up about "muh ethics" and "muh 'copyright infringement' " and appreciate how powerful and useful this tech is?Some people need something to feel superior about and virtue signal about.Luddites gonna Ludd.
>>101928521I meant Flux, hence "so I may be switching soon."I could have been clearer.
>>101928528Post originals
>>101928548I could have interpreted you more generously.
>>101928318wow that was a very good video. thanks anon>>101928358obviously italian dumb amerigolem and probably a scientist or researcher at that
>>101928551Already deleted.
I don't get the T5 bullshit. What am I supposed to download to make flux work on forge?
> 2010 screencap of old blurry snapchat selfie with arm extended out, blown out whitebalance, desaturated, low contrast, #nofilter #instagood #yolo
Has anyone else noticed how the inpainting in Forge is kinda bad?Automatic handles things well, but regardless of the models I've used in Forge, the results are just plain worse in comparison bordering on bad.It's pretty much the only thing Automatic does better at this point.
Why is there two flux variants on civitai? What's the difference?
>>101928612
>>101928607the Flux diffusion model, the Flux VAE, Google's T5 1.1 XXL and OpenAI's CLIP Vit-L/14 (just the text encoder)
>>101928625schnell and devturbo and normal
>>101928617I have not noticed.
>>101928637Dev is also a kind of "Turbo", compared to Pro
>>101928630Nice
>>101928607for brbniggafast4 just the nf4v2 checkpoint (put it in checkpoint) for others try different combinations here https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/1050some need t5 and ae.stf and clip.I
>>101928503thanks! that's helpful, and saved me asking about the symlink part too>fp16 t5 on cpu + fp8 or q8 on GPU interesting, haven't seen (or haven't looked for) multiple quants with image models before, but I'm familiar with it from running LLMs, good tips>you have 16GB of VRAM, not 12it's a laptop 4080, I'm pretty sure it's 12gb (dedicated, at least) unless I'm missing something (see picrel)cheers
>>101928068No, only tried this one.
Forge always OOMs with Flux. It's as if the mechanism it uses to allocate memory is fucked. I don't have this problem with Comfy. It just offloads more to CPU and gens slower.
>>101928653not my words, schnell means fast
>>101928630what in the fuck what is that?
>>101928682its sonic goes to the real world dude
>>101928665Are you using Shared? I had the same issue. Use CPU
>>101928665same, forge seems to OOM half the time and you have to restart forge for it to work
>>101928682dildonic terminal
>>101928715I tried all the combinations. I would prefer to use forge for simple stuff, but I can't.
>>101928630what do you even prompt to get that? just realizing i'm a irredonotredeem stupid prompjeet ramlet
>>101928664catbox please
>write "She is white. She is caucasian. She is European.">asian traitsREEEE
>>101928128For me it's personally worth it. I wait around 40 seconds for each image on a 4090.To copy the example of another anon, since it works so well:https://imgsli.com/Mjg4Mzc0>A drawing of Hatsune Miku with dreadlocks and light black skin skateboarding in New York at night. She is holding a smartphone on her left hand and a multicolored ball on her right hand, she has a red t-shirt with text on it that says: "MIGU". A pikachu can be seen on the top of her head. Her speech bubble says "Hard to keep me in style huh?", neons, 50's comic book styleSee how with no CFG/negative conditioning it goes completely off form on a more complex prompt? That's why CFG/Negative prompts existence is so important. I don't personally care about blur, but that's more better handled by tonemapping from what I've heard: https://old.reddit.com/r/StableDiffusion/comments/1estj69/remove_the_blur_on_photos_with_tonemap_an/
>>101928682good ole diy doohickey >>101928664flux + lora? seeing a non-buttchin is mindblowing to me lol>>101928731> a DIY wearable device strapped to a person's wrist, featuring a small screen displaying the classic Windows XP startup logo. The device appears to be a homemade or modified piece of technology, with exposed circuitry and a connected power cable. The screen emits a bright blue light, contrasting with the warm, dimly lit background. The overall aesthetic combines retro technology with a futuristic vibe. This juxtaposition of old and new tech elements creates a playful, cyberpunk-inspired image that merges nostalgia with a high-tech, dystopian future.
>>101928528>b-b-but AI doesn't have S-SOUL"Soul" is cope from bad artists/musicians meant to convince themselves that technically superior artists/musicians are actually worse than them.
>>101928750yellow fever bros, we won
>>101928750>caucASIANyou brought this upon yourself
>>101928400Well I tried turning on --lowvram and now my gen time is even lower than before. I'm happy though shit is still very slow.15s/it, AMD 6700XT.
>>101928793with Flux you can lower the resolution pretty far without getting artifacts, use that to speed up your gens
>>101928793I'm using a 5700. Try this command:>python main.py --normalvram --force-upcast-attention --use-split-cross-attention --cpu-vae
Some of you are starting to look like avatarfags. If you're thinking "Does he mean me?" then the answer is yes. Try gen something else.
>ViT-L-14-BEST-smooth-GmP-ftdoes this not work with the latest forge pull? I'm getting dict not found. Switching it to clip_l works obviously
N
>>101928840I will keep 1girling and you will never stop me
>>101928840You begin with a simple prompt and want to improve it more and more. You know how it is.
>>101928750Skill issue. Write blonde/brunette.
>>101928840>1GIRL>I DON'T KNOW WHY
>>101928893But I want her hair to be black.
>>101928914Blonde/brunette with hair dyed black.
>>101927580Can my PC with Athlon 3000G with 8 GB of RAM now generate any good images?
>>101928946no
poster posting mode engaged>>101928902epic art style, what did you prompt for that?
>>101928946depends on your life expectancy
>>101928902I tried so hard, and got so farrrrrBut in the end, I'm gonna gen some booba
>>101928946only though saas desu
>>101928360
What's the current sampler/scheduler meta ?
>>101928993
>>101928957It's a Flux gen going into Florence2 to mask the text and then resampled at a high-denoise in 2DN-Pony. Then the text gets pasted into the image and resampled at a low noise.pic related is the original gen, kek
>>101929021how often does Florence2 mess up?
>>101929009SDE Karras 50 steps
>>101928993It's harder to get deformed hands than now, must be something you are doing. Picrel is one of the few hand issue I have had today
ComfyUI gguf Flux lora support yet?
nuketown trickshot time>>101929021smart workflow, essentially a style transfer to pony which is the illustration style I was aiming for, appreciate the info
>>101929049it was almost 100% accurate on the bounding box, which resamples around the entire sign but can create visible seamsin masking it was hardly ever accurate, so it was more about finding a balance of a good de-noise and feather to actually merge it in seamlessly. Much easier to do on just a sign, much harder to do with the background.
Does flux only work with comfyui?
kek
>>101929103Forge too.
>>101929111>happy birthday
>>101929111> checkedbday cupcake german edition
>>101929111Hmm a nice toasty beer on a cold day
>>101929112mmm I still only have A1111 setup. do you just set up the venvs in their own folders and point each different ui at your model and lora and embeddings etc? I'd rather not duplicate all those...
>>101929021>>101929088More important takeaway here really is that using Flux -> Pony to create really consistent compositions without relying on Pony's RNG booru tagging is way more reliable.
>>101929134That's a big dog.
>>101929103i'm using forge
>>101929138tiny woman
>>101929139>font
>>101929139hows gguf vs nf4 for gen speed? worth switching?
>>101929139Why do you have the faggoty teenager font on everything?
>>101929139Do you need to mess with settings or default is fine for flux?
>>101928758NTA but I should start learning from other people's prompts more, it seems I just need to put in more natural language. Probably without the juxtaposition bullshit though.
>>101929164nta but identical, if not faster on gguf
>>101929139wtf
>>101929176I just use gpt 4o to "spice up" my prompts. I do normal tagging description and it outputs me a really nice descriptive copypasta prompt since flux seems to love boomer promptingCan also take reference images and describe them for you which I use a ton now
>>101929015>>101929064Can you try with multiple characters? And close gestures like touching and such?
>>101929139Holy fucking based.
>>101929217no politics please
>>101929217hail to the king baby
>>101929213My guess is that their images were captioned by a VLM during training so it responds best to flowery VLM-style proompting
>>101929230You never play Duke Nukem?
>>101929166default's fine.i experiment with different schedulers and move distilled cfg scale around to get uglier or more stylized stuff though
>>101929215
is there a way to force t5/clip to cpu only on forge?
>>101929249no that game is political
I wish strippers near me had that body type. They are always skinny and covered in tattoos
>>101929327no, impossible, no interface can have all QoL features, the Diffusion Interface cartel demands so with the penalty of death to infringers
>>101927580i swear flux dev can only generate instagram bimbos, even if you ask for an average looking woman or heck even an ugly one you'll get that classic instagram thot with lip filler, what is wrong with that model.
>>101929368It's made to wow normies and investors
>>101929368a severe lack of candid normal photos
>>101929368try using "British" in the prompt
>>101929368You crank up the aesthetics filter which ends up removing all candid, amateur and real photography.
>>101929368you can gen uglies i just dont save them, unironically dont use "woman" or "girl" keywords and just use a oddball name like martha or bertha
I have achieved what essentially amounts to lewd flux via inpainting. The possibilities are endless
>>101929350for me strippers are always overly thicc and fat with kinda blocky bodies and plastic surgery, we could use some more skinny ones
i like my 10 second rough draft gens, 22-30 second "good quality" gens, but i realize that lower quality model means i typically burn a lot of seeds trying to get what i wantfor those of you that tried out flux NF4, how quick and dirty can you go for rough drafts as a composition tool?
>>101928753disconnected the negative prompt and connected the ones that went through it to something called emptyConditionZero out.it skimmed off 15s from 65s to 50sput on nf4 which is supposed to be (niggafast) but not much difference desuwhat are the settings that add up to double the almost double the it/s? the sampler type (deis), the scheduler (ddim_uniform)?
https://x.com/Thom_Wolf/status/1824397053239103961Yeah no shit. Amazing what a competent model can do for the community. >>101929368skill issue
>>101929404ok saar
comfy, what the fuck is this? the old ui was fine
>comfy ui>fineno
Anyone got a gguf workflow for flux? my shit is coming out like picrel and I can't figure out what the problem is
>>101929215You're not doing anything wrong anon. It generally goes like this.Frequency of good hand generations on average goes down as:>style goes from 3D styles to rougher and rougher 2D styles>complexity of desired hand interactions and poses goes up>the desired hand pose and camera angle taken of the that hand pose is seen less during training (on the internet)For instance, one of the more rare camera angles of hand poses seen across the internet is the viewer's hand outreached holding a character's hand. This very often results in mangled hands.
>>101929450literally no issues with the previous ui once i got over the learning curvebut this looks like nu-slop
the prompt text box in ComfyUI suddenly decided to stop spell checking.I'm ESL, I need it!
>>101929291This looks real
>>101929414Gross on the plastic surgery. Would love a slightly chubby stripper though.
>>101929417Instead of connecting an emptyZeroOut to the AdaptiveGuider, replace the adaptive guider with a BasicGuider that connects only the positive.
>>101929467A UI is shit if it has a substantial learning curve.
>>101929464Thanks for the nice explanation
>>101929502Learning curve wasn't substantial though, I stopped being confused after a day or two of learning. Same as Automatic1111/Forge. Anyway you're not even talking about the UI at that point, you're talking about the node-based workflows.But yeah, if you can't get to grips with Comfy in a day or two then you probably just got filtered.
>>101929368
>>101929498wow super bazooka fast speed30s
anyone mind explaining to me exactly what the forge settings "Swap Method" (Queue vs Async) and "Swap Location" (CPU vs Shared) do? I have doodoo feces computer and am trying to optimize running with gguf quant and separated text encoder but that's about as much as I know. Also is it normal that forge expects me to select clip AND t5 (and a vae) before it lets me generate? I thought the whole point was you can choose between clip and t5 but maybe I misunderstood?
>>101929529I stopped being confused after about five minutes of using A1111/Forge.
>>101929546Yeah it's gonna be pretty much a 2x speed increase, since the negative prompting is generating a "reverse image" to compare to the positive image at every step - so it is just double the time.
>>101929560I actually stopped being confused after about 30 seconds of using Forge. I had already made my first pull request my minute 5. At an hour, I had forked the project. The week after I had an API running serving 10,000 clients per day. At the end of that month, I managed to sell the website for 1 million to a venture capitalist.You're just at the beginning of your learning curve, unfortunately.
>>101929582This is true, I am this guy's girlfriend's husband and saw him working on the website while cleaning his cum off of my wife (with my mouth of course).
>>101929616
>>101928829I tried different gens with all and some of these options one by one. With all of them on I got 18s/it over 15. I tried different combos but in the end just--lowvramand nothing else is the fastest.
--lowvram
>>101929139At first i make fun of you for your strange font choice, But i see you're also based.
>>101929021Can you automate all of that with comfy?
>>101929464Compared to the other models the results are quite precise and amazing if we ignore the small details
>>101929705cool
>>101929487thanks, I try
Which versions of flux do Loras work on? Only fp16? Fp8?
has anyone made a good script for kohya to queue up lora training?
>>101929731thx
>>101929824is this someone known?
You people are fucking weirdhttps://civitai.com/models/652143/bombshell-biceps-flex-flux?modelVersionId=729576
I rember sd models being trained on a specific resolutions. Does flux has preferences like that?
>>101928228where do i put the width and height?>sorry if its a retarded question this is all shot in the dark new stuff
>>101929700Yup. Here's a catbox if anyone wants to try it. Bit unoptimized for all the strides with made with Flux since it's an older workflow of mine. (Although I swapped it out with GGUF for this run)https://files.catbox.moe/mz76t2.png
>>101929847go buy an ad
>always the bulgewhat did flux mean by this?
>>101929842wow prompt? lora?
>>101929889If I was going to be shilling anything it would be this onehttps://civitai.com/models/652149/professional-nude-photography-flux?modelVersionId=729582
>>101929905flux likes a puffy pussy.
>>101929827just ask chatgpt or Claude to write one
>>101929845nope, just some random generated person
>>101929905menstrual pads adverts, that's how Flux learned about panties
>>101929905training data? more like tranny data
>>101929905Probably scrubbed images of women in underwear from the dataset to keep us safe
what's the EasyDiffusion/Fooocus but for generating videos?and if it doesn't exist, what service do you recommend? paid or not
>>101929885noice, thanks
don't forget that Jesus died for our sins
>>101929908> an image with heavy film grain thats slightly out of focus of a soldier standing in an empty snowy plane in russia. There is blood splattered all around the solider. The image feels cold, dark, anxious and dystopian.added the extra noise in post since I had to convert to jpg to get the size under 4mbmain things are steps 15 and guidance 1.0https://files.catbox.moe/6xdewe.png
GUYS PLS ANSWER QUICK HOW DO I STOP PROMPT IN COMFY??? QUICK GUISE PLS
Just one nice cold beer
>>101930006you have to wait now
>>101930006view queue to right > clear queueif its midway through a node, its basically soft locked. Just CTRL+C spam the cmd prompt until it force exits and just relaunch
>>101930039thanks man mom was about to freak
>>101930006>>101930055dont worry, the feds cant see what you're genning
>>101930073plot twist the fed is his mom
maybe two beers.
evil evil image dont look at it
>>101930108it's for the best that i forgot the image
>>101930008>>101930086epic face styling
>>101930149do you add noise in post?
>>101929905>>101929564the negative prompt tends to add 15-20 secs to a 30s gen. should I only use it if i'm not getting what i want or does it makes images better in a general sense? i dont know
>>101930176Personally I think it does, but that's just anecdotal.
WOW SEXY OL (OFFICE LADY)GEM!I'LL HAVE THREE BEERSAND SEXY TIME???
>>101930162yeah i do, hard to get it after upscaling, but initial gens seem to be okay with it >>101930002
going away now mom is freaking out fucking stupid bitch thanks for helping me
>>101930209DOI
>>101930185this is solid, prompt? color grading looks great
>>101930232Teebs?
>>101930087The chin is like a watermark.How do you prompt for such varied faces? I always get generic asian or generic spic.
>>101930252who is he? thanks again have to go
Wait, I guess it's because you're using dev and not schnell. God fucking damnit.
ultimate sd upscale alternative:https://x.com/Machinedelusion/status/1824138204834140431https://github.com/filliptm/ComfyUI_Fill-Nodes
>>101930293don't download this node it creates mustard gas
>>101930293>those gen speedsI'm shaking right now.
Hi anonsIs there any way to convince flux to generate large breasts? I wanted to upgrade the model I use to generate booba from epicrealism_naturalSin to any of the flux varieties, but they all seem to actively fight against generating anything but the flattest of women.The faces and the backgrounds are nice, but the model is worthless to me because I like big booba for my cooming needs.
>>101930348Have you tried things like "large, natural breasts", "ample bossom" and "large chest"?
>>101930344that's sped up video
>>101930382Yeah, I know. You don't understand my pain.
>>101930348>>101930375For me?It's (Voluptuous Body:1.5)>>101930293i barely understand how any of this works and looking at this flew right over my head
>>101930375I have tried a variety of prompts. It gives them pregnant bellies, but nothing above a c cup
>>101930407nice
>>101930404Flux understands weights?>>101930407kek
>>101930348if you lower the guidance it seems more receptive about certain words... to the detriment of everything else
>>101930421I'm partial to 2 (on schnell, even 1). What do you use?
>>101930348Don't specify gender, either say character or non-binary. I'm being unironic - female just skews the results massively.>The image features a high-quality, original depiction of a non-binary person with tanned skin. They have long, pure white hair. They have gigantic breasts that are two times the size of her head, 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".
>>101930451YEEESH
>Here's your hot asian gamer wife
>>101930454They are just throwing up gang signs
>>101930454yeah fella it was a quick gen to demonstrate the prompt, spare me your scorn
>>101930393we've all been there man
Any cute JSON files that automatically creates helpful nodes on flux?I'm trying to recreate this dude's workflow, but i have a low attention span and really bad eyesight.
>>101930492This is just a carbon copy of Comfy's starter workflow for Flux. Go drag the image in.https://comfyanonymous.github.io/ComfyUI_examples/flux/
>>101930492that looks like a default workflowshould be available here:https://comfyanonymous.github.io/ComfyUI_examples/flux/just drag the images into comfyui
Can I get a QRD on tips/tricks with flux? Or is it essentially the same ol' stuff with SD3/XL and nothing really new to learn?
>>101930518>>101930524thanks!
>>101930492i fugging hate every single "ai" youtuber out thereevery single fucking one of them
>>101930516Same prompt but using "woman" and "she" everywhere.
>>101930531Le epic reddit threads to go through, since we don't have a flux rentry:https://reddit.com/user/Total-Resort-3120/submitted/
>>101930531Verbose/descriptive writing like you're trying to describe an image to a blind man
>>101930547I'll take a gander, thanks bud
>>101930433i like to gen 4 or 5 pictures with the same prompt at different guidance levels, just to see the difference.If i need a precise but plain picture i crank it up, if i need a certain artstyle i lower it to 2I'm using dev with fp16>>101930209
>>101930540he might be a bit obnoxious, but i think he does a good job of demystifying the subject.
>>101930558So boomer/autistic prompting is back on the menu?
>>101930006>>101930055>>101930232
>>101930571>a bitunderstatementand they all hae obnoxious accents (no offense to accented people out there)i'd rather watch subtitles>>101930547i always thought that was debo's username
is anyone using flux loras with 16gb vram? Almost every lora I try practically bricks my shit
>>101930209And just one beer! haha
>>101930576yeah, thanks to the T5 encoder
I'm going to have to buy a 24 GB card and pay it in a year at this rate...
>>101930420>Flux understands weights?far as I know it will only apply the weight to the CLIP prompt
>>101930597hah
There's no way to use T5 on SD, is it?
>>101930634https://github.com/TencentQQGYLab/ELLA
>>101930634SD3 uses T5
>>101930451>>101930545Same gen but with "a woman" and she pronouns. Made a noticeable difference on my workflow.https://imgsli.com/Mjg4NDA1
>>101930587Same.
>>101930404>>101930421>very nice>>101930451Thanks for the suggestions and help anons. I would post a comparison for my results, but >image limit reached.REEEEEEEEEEE
>>101930680>woman looks more masculine than "non-binary"wtf
so you guys come here often?
Come and get it people...>>101930742>>101930742>>101930742
>>101930712Oh, and voluptuous is part of my default prompt.I think I'm going to give up on trying to make flux work until there are some good loras for it
>>101929783Very nice
>>101930586>i always thought that was debo's usernameOh please, don't put me in the same line as that retard ;_;
