Previous /sdg/ thread : >>101673399>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>SD3 info & downloadhttps://rentry.org/sdg-link#sd3>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>Animationhttps://rentry.org/AnimAnonhttps://rentry.org/AnimAnon-AnimDiffhttps://rentry.org/AnimAnon-Deforum >Index of guides and other toolshttps://rentry.org/sdg-linkhttps://rentry.org/rentrysd>View and submit GPU performance datahttps://vladmandic.github.io/sd-extension-system-info/pages/benchmark.htmlhttps://docs.getgrist.com/3mjouqRSdkBY/sdperformance>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 news08/01/2024>Stable Fast 3D: Rapid 3D Asset Generation From Single Imageshttps://stability.ai/news/introducing-stable-fast-3d>Announcing Black Forest Labshttps://blackforestlabs.ai/announcing-black-forest-labs>Flux: The Next Leap in Text-to-Image Modelshttps://blog.fal.ai/flux-the-largest-open-sourced-text2img-model-now-available-on-fal>ComfyUI: Basic Flux Schnell and Dev model implementationhttps://github.com/comfyanonymous/ComfyUI/commit/1589b5>Kolors ipadapter FaceID Plushttps://github.com/Kwai-Kolors/Kolors/tree/master/ipadapter_FaceID>The EU’s AI Act is now in forcehttps://techcrunch.com/2024/08/01/the-eus-ai-act-is-now-in-force>Video game performers picket over AI protectionshttps://apnews.com/article/sagaftra-strike-video-games-ai-f3f18ad01c5b8f4d525a836aeb531447>Paying More Attention to Image: A Training-Free Method for Alleviating Hallucination in LVLMshttps://lalbj.github.io/projects/PAI>Detecting, Explaining, and Mitigating Memorization in Diffusion Models https://github.com/YuxinWenRick/diffusion_memorization>Forgedit: Text Guided Image Editing via Learning and Forgettinghttps://github.com/witcherofresearch/Forgedit/>ControlMLLM: Training-Free Visual Prompt Learning for Multimodal LLMs https://github.com/mrwu-mac/ControlMLLM>Accelerating Image Super-Resolution Networks with Pixel-Level Classification https://github.com/3587jjh/PCSR>ComfyStereo: port of the stereoscopic script used in stable-diffusion-webui-depthmap-scripthttps://github.com/Dobidop/ComfyStereo>MeGA: Hybrid Mesh-Gaussian Head Avatar for High-Fidelity Rendering and Head Editinghttps://github.com/conallwang/MeGA07/31/2024>Bubble Prompter for Stable Diffusion WebUIhttps://github.com/captainzero93/sd-webui-bubble-prompter>Waifu Diffusion V public testshttps://huggingface.co/waifu-diffusion/wdv-tests>ComfyUI_frontend v1.2.7https://github.com/Comfy-Org/ComfyUI_frontend/releases/tag/v1.2.7
>mfw Research news08/01/2024>Tora: Trajectory-oriented Diffusion Transformer for Video Generationhttps://ali-videoai.github.io/tora_video/>Dynamic Object Queries for Transformer-based Incremental Object Detectionhttps://arxiv.org/abs/2407.21687>Expressive Whole-Body 3D Gaussian Avatarhttps://mks0601.github.io/ExAvatar/>MTA-CLIP: Language-Guided Semantic Segmentation with Mask-Text Alignmenthttps://arxiv.org/abs/2407.21654>Evaluating SAM2's Role in Camouflaged Object Detection: From SAM to SAM2https://arxiv.org/abs/2407.21596>A Simple Low-bit Quantization Framework for Video Snapshot Compressive Imaginghttps://arxiv.org/abs/2407.21517>Fine-gained Zero-shot Video Samplinghttps://densechen.github.io/zss>Generalized Tampered Scene Text Detection in the era of Generative AIhttps://arxiv.org/abs/2407.21422>Benchmarking AIGC Video Quality Assessment: A Dataset and Unified Modelhttps://arxiv.org/abs/2407.21408>SmileyNet -- Towards the Prediction of the Lottery by Reading Tea Leaves with AIhttps://arxiv.org/abs/2407.21385>AI Safety in Practice: Enhancing Adversarial Robustness in Multimodal Image Captioninghttps://arxiv.org/abs/2407.21174>Optical Computing for Deep Neural Network Acceleration: Foundations, Recent Developments, and Emerging Directionshttps://arxiv.org/abs/2407.21184>Embedding Space Selection for Detecting Memorization and Fingerprinting in Generative Modelshttps://arxiv.org/abs/2407.21159>Adding Multi-modal Controls to Whole-body Human Motion Generationhttps://yxbian23.github.io/ControlMM/>Direct Unlearning Optimization for Robust and Safe Text-to-Image Modelshttps://arxiv.org/abs/2407.21035>Safeguard Text-to-Image Diffusion Models with Human Feedback Inversionhttps://arxiv.org/abs/2407.21032>SuperVINS: A visual-inertial SLAM framework integrated deep learning featureshttps://arxiv.org/abs/2407.21348
imagine if SAI wasn't so incompetent lol
oh that's my cat, neat
>>101676775I have a pretty powerful imagination but thats a hard one
so how do I use this new model, I am a brainlet who doesn't code.Is there a comfy workflow I can download or something? I have the 23gb model downloaded
>>101676825https://comfyanonymous.github.io/ComfyUI_examples/flux/
its kinda crazy that the 5090 is rumored to have only 28gb. the nextgen flagship gpu won't be able to fit current gen models>>101676825you can burn through your free gens on replicate if you wanna avoid local setuphttps://replicate.com/black-forest-labs/flux-dev
has anyone trained a lora yet for this black forest stuff? If I can't generate my waifu it is difficult
>>101676916Dang. another new good model, I'm liking Kolors+ipadapter, can't get the FaceID to work yet in Cubiq/Mateoo nodes.
>>101676825Sorry for being this annoying guy again, but if you hit Replicate ratelimits, or just wanna also try DALL-E together with Flux, I started a temp Telegram bot bot with DALL-E 3/Flux Schnell (fast)/Flux dev: @imgfun_botJust send /start and it'll show usage. I'll shut it down in a couple of days. You can get up to 4 concurrent images with any of the models.I swear i'm not glowing, i just wanna help out people since I have some extra replicate access that can be used for all the gens.
>>101676966>Kolors+ipadapterthats a cool setup. I've been wondering if people have been playing around with that combo yet. how do you feel about kolors in general? I think of all the new models, that one slipped under the radar the most
>>101676961I'm not waiting a whole day of not using my computer for a bake. if inference is this close to the edge on a 4090 training must be hellish
>>101677039Agreed, I was using SD3M until I found Kolors and it is amazing. Very good at following prompts and quality is top notch. most LoRas (sdxl) work.
>>101677187>most LoRas (sdxl) workoh really? thats very interesting. I wouldn't have assumed they were compatible
>>101677229you'll notice your cmd window throwing errors with some.
>>101677229>these all work without throwing errors.
>>101677286what is the actual output?
>>101677286https://sliders.baulab.info/weights/xl_sliders/
>>101677308What do you mean? Pretty much what you sees it does.
>>101677308smiling.pt LoRa-4 vs +4
>>101677457lol
uhh>fatal error C1128: number of sections exceeded object file format limit: compile with /bigobjappropriate thoughBIG OBJgoing to work on sequential weight loading and workspace stuff tomorrow probably. maybe some stuff to make writing IR easier tooalso, 1B
hi
old gen while I wait for the current one to finish
big data gangbig data gangbig data gang
>>101677876Why do your gens look like dalle and have the same exact resolution as dalle?
>>101677839Why do you keep generating this Addams family-ass bitch?
>>101677923because im using dalle
>>101677943my waifu
uh... replicate is suddenly requiring login, what happened?
>>101677950why not use the new model that shits on dalle?>>101677991they remembered they can sell analytics
>>101677991Obviously because they're getting a shitton of people trying the new Flux model.
>>101678002>>101678010So I didn't get my IP flagged or something? Thank goodness. This happened exactly when I started to try to generate some spicy pics
>>101678079the good thing of this model is that it's local, you can run it on fp8 and the quality is still great, only needs 12-13gb of vram though
>>101678079You can't gen NSFW pics from their website now too, they enabled the safety checker on the model, so you can only gen spicy pics with the API.
>>101678002i have more important things to do than wait minutes for each image, like developing ait to make the new model run faster, also big data gang
>>101678187comfy and I make random shit all the time while we dev, git gud kek
>>101678202i am too lol. don't be a dick
>>101678187>>101678224just use flux on the api instead of dalle then? dalle sucks, flux is literally better
>>101678234why are you so upset about what model im using
>>101678267Check what this general is about again, it's not about dalle, it's about local models and SD/Flux in particular. There's a de3 thread on /g/, go there.
>>101678224I'm not, just odd that you can just do the same thing locally now or use the api like anon said. dalle is dead dude and openai can rot. we will probably get leaks faster when it dies
>>101678278not my problem>>101678280idk man you seem pretty bothered about it. whats the big deal. its all the same at the end of the day
>>101678323>not my problemIt is your problem though, as you must follow the rules of the general and the board you're on.
>>101678280>dalle is deadreal clown hours kek
>>101678328He's right though, flux is literally cheaper and is not extremely prompt filter cucked like dalle (even uncensored endpoints have it btw)
>>101678325is this coming from an off-topic avatar blogposter?
>>101678346No, I posted plenty of on-topic Flux gens in the thread.
>>101678346If you mean >>101678280, that's not me, that's some other dude. I don't agree with him about OpenAI dying - GPT-4o can do image gen but they just didn't release the feature, and they can be cooking dalle4 for all we know.
>>101678344it has very limited copyright and artist knowledge in comparison to dalle
>>101678325get a grip desu>>101678344dalle is free on bing. i really don't care
>>101678323>idk man you seem pretty bothered about it. whats the big dealthis is a local general>>101678328https://www.asiafinancial.com/openais-8-5-billion-bills-spark-bankruptcy-speculation
>>101677864this is peak ballet performance>>101677876>big data gang>big data gang>big data ganghttps://suno.com/song/68597cfa-38c0-4fb7-a52e-7458bc67d9b8>>101677991too much CSAM. the story is always the same>>101678234post gens if you wanna be uptight about other people's gens. you're barking at other people for what they post without having anything to post of your own
>>101678379>post gens if you wanna be uptight about other people's gensI posted multiple in this thread
>>101678379nicei made this one earlier, i really like ithttps://suno.com/song/26817194-cf0f-4d03-82a9-c81c863f3a1a
>>101678435thats great. I love when suno hits that chillstep/ambient vibe. it does that dreamy atmosphere really well. I put together something similar recently, maybe you'd dig ithttps://suno.com/song/e1bba736-c831-4c69-b772-cb0ce54b1870while you're out there mining data, can you mine me some extra suno coins? I'm so poor...
anyone have success running Flux Dev on a 24 GB card?mine keeps running out of memory and crashing. on linux + Comfy. Tried a bunch of different stuffShould I try Schnell instead?
>>101678627can you make her younger?
>>101678633>Should I try Schnell instead?Anon, it's the same size, it just gens faster (and worse quality)
>>101678633>You can set the weight_dtype in the “Load Diffusion Model” node to fp8 which will lower the memory usage by half but might reduce quality a tiny bit.
>>101678649Yeah, did that. I get a memory allocation error. It's trying to allocate vram for the whole unet file, so the full 23.8 GB. I think it's a bug.
>>101678656make her even younger
>>101678656can you forcefully rape and impregnate her?
Happy national gf day!
>>101678731can you make her a loli
there, pushed flux>>101678585ive noticed the lyrics are better when you describe a song instead of generating in custom modewill be cool to try with my lyrics llm. waiting until i have a BIG amount of lyrics scraped. i know it will upset some posters if i use the gpt2 based training code from the last one, not my problem tho and ngl its funny how mad they get
>>101678367Maybe this can be added back in
>>101678375I'm sure Microsoft will just let them go like that.
>>101678323Sorry man, nothing against your waifus but once you have been politely pointed told you are offtopic you should oblige. Creating Dalle and creating SD-based generations are a different set of skills, which is why we have a dall-e general down the aisle as well.
>>101678899Friendlier windows times...
modelism is worse than racismgoo nihgt
>>101678887>they just buy black forest instead
>>101678877>making loras of a 24gb modelthere is enough vramlet settings cope as is with XL/pony loras. can't imagine what kind of coping there will be for a model this big.
>>101678888>told you are offtopic you should obligethe avatars never do, and diary entries are as off-topic as it gets
>>101678976People whined at any model size anyway, even back from nai leak days.
I don't even know how this happenedhttps://litter.catbox.moe/nx8jpg.png
>>101678679I got it to work by running with --gpu-only.It was overloading my system ram. I should buy more LOL
>>101678787>ive noticed the lyrics are better when you describe a song instead of generating in custom modeI think the custom lyrics are just a gpt dip. I notice a ton of the same lyrical tropes between the two>neon nights>silent realms>digital skies>unraveling talesetc etcbut maybe suno asks for a more specific structure so there's a better rhythmic meter to their auto-lyrics which makes it fit to music better>>101678862>>101678878>>101678899these are trippy. perfect UI + content if I were stroking out. wait, I'm not stroking out, am I?>>101678912says the guy who will never try new modelsgn>>101678913not even a crazy idea. why out-compete your competitors when you can just buy them out and shutter them instead. its usually cheaper too
>>101678998I think the avatarfags and the waifu fag who literally like an ape just replied "my waifu" to a question from me are also AIDS. But it doesn't help the quality if more noise is added. If you post dalle here, you will never engage with a post asking about a ComfyUI workflow or what some good Webui settings are. You see how Dalle fags posting here shit it up, then.So kindly, just piss off.
12T models/ about half of them720m + 274m + 4m + 7m = 1005M!!i sleep>>101678888i don't care. stay mad>>101679055yeah the custom lyrics follow silly patterns>>101679072i've contributed more than you ever will desu
>>101679072that anon is talking about datasets, meanwhile blogposters are wasting space or spamming/flooding the thread.keep moving goalposts, it only makes it looks worse for you.
>>101679072why are there so many shitlord nogens trying to gatekeep today? no one wants you faux-policing what people do and don't do here. post your gens, post your workflows, post your insights. if you feel compelled to tell other people how to behave or what to post, just don't instead
so far, i feel lux needs negatives, i think that would migitate the quality inconsistency
>>101679181>says this>posts a shit gen
>>101679181>policing>>101678379>post gens if you wanna be uptight about other people's gensLook inward debotard
This model feels like it was trained with sdxl juggernaut, dalle and mj images
>>101679184someone was saying the dev model can accept negative prompts. I haven't tried it myself>>101679186I'm running low on flux gens and you weren't worth a good one>>101679201I'm not policing peoples gens or what they want to contribute to this space. I'm policing shitlord nogens who offer nothing yet feel justified in barking at posters. if you don't wanna put up your own content, stop shitting on others. and even if you wanna put up your own content, there's plenty of space available for other people to do what they want too.
>>101679256who is ipvau
is that gook \ 1.5 AI face baked in or are you just chaining models together
>>101679280>gookPlease don't be racist.
I have a 3080 and a 4080, can I assign part of the generation to one, then the rest to the other?
>>101679286we've had the same ai face for the past 2 years, I'm tired of seeing it desu
wen will replicate let us generate pics of up 2 mp :((
nb4 mass reply
>>101679324thats because of inbred models, most finetuners train on gens since gens dont have/need licenses
>>101679366where nipel?
MEDS
>>101679366grim, habsburg mode colapse soon
>>101679235>someone was saying the dev model can accept negative prompts. I haven't tried it myselfboth dev and shnell can't accept negative promps because to use neg you need to have a cfg > 1, and when you do that it just hurts the output, that's a shame
>>101679475you can use this node and play with the values, and yeah there are no negatives
>no negative promptDoA
>>101679582>>101679594you can use the regular ksampler to do it just keep in mind what >>101679475 is saying
>>101679582>>101679594if Schaduled CFGGuider doesn't hurt the model at decent CFG, then we'll be able to use negatives
>>101679614for example in this image, i was wondering why it was blurry and it was because the cfg made it closer to the negative prompt than the positive prompt, after testing it, imho for now its like using a double positive(?) which is the same as using t5 and clip l i think
at least the reactor users will have a blast
>>101679679alien feet
>>101679679>perfect hands>horrible feet:(
>>101679669show me your nodes anon, maybe you did something wrong
>>101679669Negative prompt works, just be careful to not fry your model with a too big cfg, I used this metadata >>101679222
>>101679794>just be careful to not fry your model with a too big cfg, I used this metadataI've heard of another CFG that is less prone to fry outputs, CFG++ no? can it be used on ComfyUI?
>>101679794>>101679869can you use pag or dynamic thresholding with it?
>>101679794thanks bro, but it didnt work for me, I believe comfy, there are no negatives...for now, if you use >>10167958 and play with the ranges you can see which prompt gets picked up or even see if they blend, but i think is just the same as cliptextencodeflux
>>101679912what cfg did you use? it must be over 1 to work, for me it worked at cfg = 1.5
>>101679937i just changed the prompt in the workflow that >>101679794 postedthis is a dumb example but you can see how it kinda works, from_cfg = 0 means it ignored the upper text prompt and to_cfg = 0.5 it picked up the text prompt from below
>>101680006both same value
>>101680006>>101680023why such a low cfg? I'm not sure if a cfg < 1 makes the negative prompt work
>>101680023upper text prompt value > lower text prompt valueso yeah, it works like a double positive for me and yeah there is no negative because of maybe a negative cfg value doesnt work? i dont know lol ahaha
>>101680072maybe when cfg < 1, the negative prompt acts like a positive prompt, try with a cfg > 1
>>101680037>why such a low cfg?The basic update rule for a diffusion process can be represented as:xt+1=xt+f(xt,t)Δt+g(xt,t)ΔWtxt+1=xt+f(xt,t)Δt+g(xt,t)ΔWtWith classifier-free guidance, the update incorporates a scaling factor ss (CFG scale):xt+1=xt+s(f(xt,t)+(1−s)⋅funcond(xt,t))Δt+g(xt,t)ΔWtxt+1=xt+s(f(xt,t)+(1−s)⋅funcond(xt,t))Δt+g(xt,t)ΔWtSo naturally the differential with respect to the CFG scale ss, we can compute:dxt+1ds=(f(xt,t)−funcond(xt,t))Δtdsdxt+1=(f(xt,t)−funcond(xt,t))ΔtFrom there, the cfg scale should make more sense.
>>101680037>>101680088heres what i want to illustrate:from_cfg = "positive" prompt valueto_cfg = "negative" prompt valuebut in reality there is no negative, only two positive prompts that blend together depending of the values of the from_cfg and to_cfg ,
>>101680088that's how it works on every model
Purp
>>101680229so cfg < 1 really makes the negative prompt act like a positive prompt?
>>101680250It flips the cond to the opposite effect. positive becomes negative etc.
>>101680278see, you got your answer, stop using cfg < 1 it messes up the negative prompt effect >>101680205
It's Schizo Time. Also 50 credits is never enough to finish songs in a day ; ;https://suno.com/song/b263ee92-d8c6-4674-aec3-b698af03d0e9
>>101680298nah, there is no negative prompt period, if there is post some proof then
IMGGEN WILL NEVER BE THE SAME AGAINNNNNAYYEYYEEEEEEEEEEEEEEEEEEEEEEEEEE
>>101680317>50 credits is never enougI feel that>>101680343imggen is totally different until we run out of replicate credits, then its back to the same old
Does anyone know of a segmentation (seg) control net model that works well with sdxl?
>>101680321>if there is post some proof thenwhat's this then? >>101679794
>>101680317How do you keep turning shitposts into club bangers?
>>101680565why aren't her eyes red
>>101680498I dunno lol, I guess having 5 accounts to fuck with things helpshttps://suno.com/song/4a714800-2096-429e-93fa-e7470ede058c
abloo bloo bloo
>>101680608have you tried out udio1.5 yet? you can do 2min+ gens with it nowhave you tried flux either? its the new hotnesshttps://suno.com/song/2eb1d12b-2241-4f0c-9c01-46b6d85eda2e
>>101680861it's a shame it knows so few characters/artists, i blame these llms being used to tag everything as generically as possible
>>101680883dalle3 knows a shitton despite being trained on 95% synthetic prompts generated with an older worse version of gpt4 with vision
>>101680907>95%source?
it's amazing how good it is with text.SD3 could never
>>101680883Is it possible to add the artists/characters in a finetunes?
>>101681006unlikely
>>101680987https://cdn.openai.com/papers/dall-e-3.pdf>To test our synthetic captions at scale, we train DALL-E 3, a new state of the art text to image generator. To train this model, we use a mixture of 95% synthetic captions and 5% ground truth captions. The model itself is a scaled-up version of the model we used in the above ablations, with several other improvements.
>>101681018Im not reading all that
>>101681030TLDR: any hopes of being able to make loras on this thing rely entirely on jewvidia allowing us to have more vram in this next gen.
>>101681047>in this next gen.the 50 series caps out at 28gb. they couldn't change that if they wanted to at this point
Good evening, anons! I hope everyone is doing well :]
is this company planning on releasing an 8B/6B model? they could wipe stability from the map forever
>>101681083the fuck are you getting on so late for purple bitch?
>>101681092LOL just got home so I figured why not hahaha!
>>101681083>>101681119I want to do unforgivable things to your purple witch
>>101681165LMAO.Uhhh I dunno how to respond to that, anon
>>101681119get in on the current hype train, new checkpoint with day 1 support from comfyhttps://huggingface.co/spaces/black-forest-labs/FLUX.1-schnellhttps://comfyanonymous.github.io/ComfyUI_examples/flux
>>101681193oops, checkpoint page also useful too...https://huggingface.co/black-forest-labs/FLUX.1-schnell
>>101681018>>101681047I see the screenshot as "it's possible but requires a shitton more of VRAM so it'll be costly and we need funding", not "impossible".
>>101681208when was "impossible" mentioned?
>>101681193Ohhh fun! I love new things!Thanks, anon! :DI'll dl it now!
>>101681063my dick is out>>101681083hi pw. there's a new hotness model out that has captivated everyone. highly recommend playing with it
>>101681260I just heard! Uhm, do I just DL the flux1-schnell.sft file and put it in my checkpoints folder?
>>101681302comfy has the usual instructions for where to put stuff here...https://comfyanonymous.github.io/ComfyUI_examples/flux/
>>101681302comfy posted a writeup on ithttps://comfyanonymous.github.io/ComfyUI_examples/flux/
>>101681323>>101681326Thank you both so much!! I'm gonna get it now! :D
>>101680321this is the wolf running prompt I posted las thread with red jersey in negneg works fine in dev
I like the way this came out!! I'm gonna try a more simple proompt next!
why can't we quantize flux like the lmg chads do to run their hundred billion+ parameter bullshit
>>101681654because images have way more infomation density than text
ola
>>101681745Hello, anon :]
amazing you can actually gen 1920x1088 on flux with a 4090 if you set both model and text encoder to fp8>>101681757gyo morning!
Oh nice! I just reset my pc cause it was going kinda slow hahaIt's been on for hella days
Stylization in flux is subpar, but you can use img2img to add style. It's not perfect e.g it can easily genderbend characters. Left Style with SDXL and right the original from Flux
>>101681797>>101681832Oops forgot to reply hahaha!
>>101681837ya true, I think its wonderful tech, but a subpar data set.. I wan't finetunes
I'm gonna cum
>>101681860yea I agree that guitar is fucking sexy
How do I stop masturbating /sdg/
>>101681912stop looking at sexy men/women/ponies (whatever you are into)
>>101681912Condition yourself to not want to masturbate.Try this: after every gooning session, start hitting yourself in the balls as hard as you can. Soon, you will develop a psychological reaction and you start associating masturbation with pain. Or you will develop a taste for cock and balls torture.
Give me a video inpainting model so I can impregnate women
>wonder why 6 minutes per image feels so much worse than 2 minutes >look at my thrice as large folder of candidate images that just need blowing upSDXL is a waste of time for anime and normie usage in general. You need to be an actual graphic designer for it to be worthwhile
this is reasonably fast with the schnell model
>>101682009ya I am getting 15s gens of this quality in full hd with schnell
>>101682047around 50s for this, 20 steps, full model. still ok I'd say.>>101681912you don't. keep going until your dick falls off. lession learned, boy. mwhaha>>101681926morning!
>>101682082>Mr. Einstein we dont have an apointment!
For those trying to make styles work on flux-dev, try to use the karras sheduler, somehow it just make it work>>101681711>>101682084
>>101682196nice, thanksthis is sgm uniform, 20 steps
>>101682212for karras you need way more steps or else it won't work, 40-50 seems to do the trick
It does duos really well!!
>>101682214ahh I see above 40 steps it seems to produce something coherent. 77 secs for 40 steps / 83s for 50 on my machine, "yikes"here, 50 steps euler karras
same prompt, 20 steps sgm uniform, thats a remarkable change in.. everything. hm
>>101682285>>101682249yeah it really makes things really different compared to the other schedulers
i don't need it...
>>101682334ya text recognition is the bomb with FLUX
>>101682346It really is!!! I'm really impressed with it!LOL that's amazing!
>>101682378good day. made the sign I posted earlier extra for you lol !this is already pretty sloppy, oops
can someone post me a basic flux comfy workflow catbox please? i'm downloading the model and updating comfy rn
>>101682467https://comfyanonymous.github.io/ComfyUI_examples/flux/
>>101682480oh nice durr, thankshow much vram to realistically run all this shit? does comfy support splitting across gpus? the text encoder is huge right?
>>101682249whats your gpu?
>>1016825453090 @ 80% power.
>>101682415I want to upgrade.But I'm poor and the GTX1060 is the best gpu I have, both in my laptop and my main rig (which has 2).
>>101682597yeah you mentioned it. ugh I got a 2080 collecting dust here, I'd give it to you.
>>101682662lol my gf would hate this so much, she's an artist who draws like 40% birds, like she's better than this for sure i guess we still care about humans playing chess so it'll probably work itself outare you using the full size t5? the fuck gpu do you have?
>>101682727oops one of the lines in my comment got deleted, i meant to say she's better than this for sure but like it's only a matter of time, what a weird time to be alive
>>101682727>guess we still care about humans playing chess so it'll probably work itself outmarkets don't work that way sadlybetter comparison would be ikea mass-produced furniture vs manually crafted
>>101682763i mean same shit, like she's above the cutoff and good enough at marketing that it doesn't matter, in your analogy ikea does not impact her ability to sell $5000 epoxy river tableslike people think "oh nooooo AI art is going to collapse the art market" as if the art market is only based on supply and demand as a function of the quality of art, i mean *some of it is* but i think there's a lot that's based on like, money laundering, status seeking and effective marketing and is kinda just vibes based and that stuff is somewhat insulated from AI artflux seems good, but this shit took 2 minutes T__T
>>101682812AI fucks up the entry and mid level, the best specialists are more likely to benefit from all this or remain unaffected with shit like gacha splash arts, stock pics, porn arts etc etc I don't think most consumers give a fuck about "vibes", they want their product quick and passable, and that affects significant amount of artfags who drew slop for a living. >flux seems good, but this shit took 2 minutes T__Tit's magic we even got this at all, pretty much dalle level comprehension (with some caveats but still)
i'm not even mad, i like that there's diversity, easy to prompt out and makes me feel like the model is creative? 2 minute gen times are killing me tho
>>101682592>>101682592>>101682592
>>101682662>got a 2080 collecting dust here, I'd give it to you.If only...It's ok though, kinda like working with limitations. I end up being more efficient.
>>101681222This is a remark not a rebuttal.
>>101682949CUTE!!!