Previous /sdg/ thread : >>101761689>Beginner UI local installEasyDiffusion: https://easydiffusion.github.ioFooocus: https://github.com/lllyasviel/fooocusMetastable: https://metastable.studio>Local installAutomatic1111: https://github.com/automatic1111/stable-diffusion-webuiComfyUI: https://github.com/comfyanonymous/ComfyUISD.Next: https://github.com/vladmandic/automaticAMD GPU: https://rentry.org/sdg-link#amd-gpuIntel GPU: https://rentry.org/sdg-link#intel-gpu>Use a VAE if your images look washed outhttps://rentry.org/sdvae>Run cloud hosted instancehttps://rentry.org/sdg-link#run-cloud-hosted-instance>Try online without registrationsd3: https://huggingface.co/spaces/stabilityai/stable-diffusion-3-mediumtxt2img: https://www.mage.spaceimg2img: https://huggingface.co/spaces/huggingface/diffuse-the-rest>Models, LoRAs & upscalinghttps://civitai.comhttps://huggingface.cohttps://aitracker.arthttps://openmodeldb.info>Black Forest Labs: Fluxhttps://huggingface.co/black-forest-labs/FLUX.1-schnellhttps://huggingface.co/spaces/black-forest-labs/FLUX.1-schnellhttps://comfyanonymous.github.io/ComfyUI_examples/flux>Index of guides and other toolshttps://rentry.org/sdg-linkhttps://rentry.org/rentrysd>View and submit GPU performance datahttps://vladmandic.github.io/sd-extension-system-info/pages/benchmark.htmlhttps://docs.getgrist.com/3mjouqRSdkBY/sdperformance>Share image prompt info4chan removes prompt info from images, share them with the following guide/site...https://rentry.org/hdgcbhttps://catbox.moe>Discord6wUwtcJsr2>Related boards>>>/h/hdg>>>/e/edg>>>/d/ddg>>>/b/degen>>>/vt/vtai>>>/aco/sdg>>>/trash/sdg
Thank you for baking, Debo! :D
>mfw Resource news08/06/2024>ComfyUI August Blog: Flux Support, New Frontend, For Loops, and more!https://blog.comfy.org/august-2024-flux-support-new-frontend-for-loops-and-more/>Lumina-mGPT: Illuminate Flexible Photorealistic Text-to-Image Generation with Multimodal Generative Pretraininghttps://github.com/Alpha-VLLM/Lumina-mGPT>VidGen-1M: A Large-Scale Dataset for Text-to-video Generationhttps://sais-fuxi.github.io/projects/vidgen-1m>MeshAnything V2: Artist-Created Mesh Generation with Adjacent Mesh Tokenizationhttps://buaacyw.github.io/meshanything-v2/>Mini-Monkey: Alleviate the Sawtooth Effect by Multi-Scale Adaptive Croppinghttps://github.com/Yuliang-Liu/Monkey>SynopGround: A Large-Scale Dataset for Multi-Paragraph Video Grounding from TV Dramas and Synopseshttps://synopground.github.io>MiniCPM-V-2_6: GPT-4V Level MLLM for Single Image, Multi Image and Video on Your Phonehttps://huggingface.co/openbmb/MiniCPM-V-2_608/05/2024>Dimensional Latent Perlin for ComfyUIhttps://github.com/NeuralSamurAI/ComfyUI-Dimensional-Latent-Perlin>Arc2Face ComfyUI Node Libraryhttps://github.com/caleboleary/ComfyUI-Arc2Face>OpenAI co-founder leaves for Anthropichttps://techcrunch.com/2024/08/05/openai-co-founder-leaves-for-anthropic>CogVideoX-2b: 清影 Video Modelhttps://huggingface.co/THUDM/CogVideoX-2b>CogVideoX ComfyUI Wrapperhttps://github.com/kijai/ComfyUI-CogVideoXWrapper>Flux-Magic: LLM-Powered Image Generation with Flexible Optionshttps://github.com/ahgsql/flux-magic>VAR-CLIP: T2I Generator with Visual Auto-Regressive Modelinghttps://github.com/daixiangzi/VAR-CLIP>Groq lands $640M to challenge Nvidiahttps://techcrunch.com/2024/08/05/ai-chip-startup-groq-lands-640m-to-challenge-nvidia>Flux Dev & Schnell Support Added to Pallaidiumhttps://github.com/tin2tin/Pallaidium#change-log>AIマンガ「LearningToon」無料公開:社会人向け縦読み学習https://prtimes.jp/main/html/rd/p/000000011.000138218.html
>close thread>open thread
>avatarfag chatroom
>mfw Research news08/06/2024>LaMamba-Diff: Linear-Time High-Fidelity Diffusion Models Based on Local Attention and Mambahttps://arxiv.org/abs/2408.02615>Fairness and Bias Mitigation: A Surveyhttps://arxiv.org/abs/2408.02464>ProCreate, Dont Reproduce! Propulsive Energy Diffusion for Creative Generationhttps://arxiv.org/abs/2408.02226>Dense Feature Interaction Network for Inpainting Localizationhttps://arxiv.org/abs/2408.02191>Rethinking Affect Analysis: Protocol for Ensuring Fairness and Consistencyhttps://arxiv.org/abs/2408.02164>PanoFree: Tuning-Free Holistic Multi-view Generation with Cross-view Self-Guidancehttps://panofree.github.io>LDFaceNet: Latent Diffusion-based Network for High-Fidelity Deepfakeshttps://arxiv.org/abs/2408.02078>Step Saver: Predicting Minimum Denoising Steps for Diffusion Modelshttps://arxiv.org/abs/2408.02054>Faster Diffusion Action Segmentationhttps://arxiv.org/abs/2408.02024>Self-Introspective Decoding: Alleviating Hallucinations for Large VLMshttps://arxiv.org/abs/2408.02032>Dataset Scale and Societal Consistency Mediate Facial Impression Bias in Vision-Language AIhttps://arxiv.org/abs/2408.01959>ST-SACLF: Style Transfer Informed Self-Attention Classifier for Bias-Aware Painting Classificationhttps://arxiv.org/abs/2408.01827>MiniCPM-V: A GPT-4V Level MLLM on Your Phonehttps://arxiv.org/abs/2408.01800>Landmark-guided Diffusion Model for High-fidelity and Temporally Coherent Talking Head Generationhttps://arxiv.org/abs/2408.01732>A Novel Evaluation Framework for Image2Text Generationhttps://arxiv.org/abs/2408.01723>Music2P: Multi-Modal AI-Driven Tool for Simplifying Album Cover Designhttps://arxiv.org/abs/2408.01651>Using a CNN Model to Assess Visual Artwork's Creativityhttps://arxiv.org/abs/2408.01481>Cascading Refinement Video Denoising with Uncertainty Adaptivityhttps://arxiv.org/abs/2408.02284>Label Augmentation for Neural Networks Robustnesshttps://arxiv.org/abs/2408.01977
>>101770391>>101770400>>101770342have you ever considered not going to the thread or just like, ignoring the things you don't like instead of being mad?
Why isn't debo in the discord?
>mfw Resource news08/07/2024>ControlNet Canny FLUX.1-devhttps://huggingface.co/XLabs-AI/flux-controlnet-canny>Diffusion1B: Dataset of 1.2 billion images generated by diffusion modelshttps://huggingface.co/datasets/bigdata-pw/Diffusion1B>Diffusers v0.30.0: New Pipelines, New Methods, and New Refactorshttps://github.com/huggingface/diffusers/releases/tag/v0.30.0>Diffree: Text-Guided Shape Free Object Inpainting with Diffusion Modelhttps://github.com/OpenGVLab/Diffree>LLaVA-OneVision Easy Visual Task Transferhttps://llava-vl.github.io/blog/2024-08-05-llava-onevision/>IP Adapter Instruct: Resolving Ambiguity in Image-based Conditioning using Instruct Promptshttps://unity-research.github.io/IP-Adapter-Instruct.github.io/>SiCo: A Size-Controllable Virtual Try-On Approach for Informed Decision-Makinghttps://github.com/SherryXTChen/SiCo>Photorealism LoRA for FLUX.1-devhttps://huggingface.co/XLabs-AI/flux-RealismLora08/06/2024>ComfyUI August Blog: Flux Support, New Frontend, For Loops, and more!https://blog.comfy.org/august-2024-flux-support-new-frontend-for-loops-and-more/>Lumina-mGPT: Illuminate Flexible Photorealistic Text-to-Image Generation with Multimodal Generative Pretraininghttps://github.com/Alpha-VLLM/Lumina-mGPT>VidGen-1M: A Large-Scale Dataset for T2V Generationhttps://sais-fuxi.github.io/projects/vidgen-1m>MeshAnything V2: Artist-Created Mesh Generation with Adjacent Mesh Tokenizationhttps://buaacyw.github.io/meshanything-v2/>Mini-Monkey: Alleviate the Sawtooth Effect by Multi-Scale Adaptive Croppinghttps://github.com/Yuliang-Liu/Monkey>SynopGround: Large-Scale Dataset for Multi-Paragraph Video Grounding from TV Dramas and Synopseshttps://synopground.github.io>MiniCPM-V-2_6: GPT-4V Level MLLM for Single Image, Multi Image and Video on Your Phonehttps://huggingface.co/openbmb/MiniCPM-V-2_608/05/2024>Dimensional Latent Perlin for ComfyUIhttps://github.com/NeuralSamurAI/ComfyUI-Dimensional-Latent-Perlin
Morning anons
>>101770466>mfw Research news08/07/2024>MDT-A2G: Exploring Masked Diffusion Transformers for Co-Speech Gesture Generationhttps://arxiv.org/abs/2408.03312>DopQ-ViT: Towards Distribution-Friendly and Outlier-Aware Post-Training Quantization for Vision Transformershttps://arxiv.org/abs/2408.03291>ReSyncer: Rewiring Style-based Generator for Unified Audio-Visually Synced Facial Performerhttps://arxiv.org/abs/2408.03284>Contrastive Learning for Image Complexity Representationhttps://arxiv.org/abs/2408.03230>An Object is Worth 64x64 Pixels: Generating 3D Object via Image Diffusionhttps://omages.github.io/>Multitask and Multimodal Neural Tuning for Large Modelshttps://arxiv.org/abs/2408.03001>DreamLCM: Towards High-Quality Text-to-3D Generation via Latent Consistency Modelhttps://arxiv.org/abs/2408.02993>Diffusion Model Meets Non-Exemplar Class-Incremental Learning and Beyondhttps://arxiv.org/abs/2408.02983>Diffusion Models as Data Mining Toolshttps://diff-mining.github.io/>MMIU: Multimodal Multi-image Understanding for Evaluating Large Vision-Language Modelshttps://arxiv.org/abs/2408.02718>RCDM: Enabling Robustness for Conditional Diffusion Modelhttps://arxiv.org/abs/2408.0271008/06/2024>LaMamba-Diff: Linear-Time High-Fidelity Diffusion Models Based on Local Attention and Mambahttps://arxiv.org/abs/2408.02615>Fairness and Bias Mitigation: A Surveyhttps://arxiv.org/abs/2408.02464>ProCreate, Dont Reproduce! Propulsive Energy Diffusion for Creative Generationhttps://arxiv.org/abs/2408.02226>Dense Feature Interaction Network for Inpainting Localizationhttps://arxiv.org/abs/2408.02191>Rethinking Affect Analysis: Protocol for Ensuring Fairness and Consistencyhttps://arxiv.org/abs/2408.02164>PanoFree: Tuning-Free Holistic Multi-view Generation with Cross-view Self-Guidancehttps://panofree.github.io>LDFaceNet: Latent Diffusion-based Network for High-Fidelity Deepfakeshttps://arxiv.org/abs/2408.02078
>>101770480gm
>first 10 posts are research paper spam>all threads everyday
https://suno.com/song/267e7d2f-5cfe-4f9c-b424-a596384cf867lyricsgpt desu
>>101770444Do you ever wonder why so many anons find Debo to be a nuisance?
>>101770536the resource news is kinda crazy today though. canny for flux? big data gang? unity's ipai? photorealism flux lora? pretty interesting stuff happeningyou can be mad about the research work though. the middle of summer isn't the best time for completed research
>>101770679too busy fluxing to care about research
>>101770679Just put it in a pastebin like anons we have been telling you for months you sperg
>>101770694what a steal. I'd pay twice as much. I'm excited for the collectors edition figurine too
>>101770763ESL?
>>101770763He'd rather piss anon off
>>101770782Not by first language, I think I got my point out?
>>101770770why do you always add the doxxcord to OP?
>>101770830imagine thinking discord doxxes you
>self-doxxes before the discord was even in OP
>>101770446his feefees got hurt when someone called him out on his bullshit so he ragequit.
>>101770632i really don't, this site is full of people i find annoying but i just ignore them because then i am not mad and it's nicerdebo def brings the thread average quality up tho so no hate there
>>101770887guaranteedreplies.jpg
>debo def brings the thread average quality up thoany examples?
the consistent worst gen poster is comfy, he will post some cool new tech and use it to incompetently make the same blurry fried fennacs
>>101770904he posts gens and answers questions, half the people here just nogen complain or post "hot" takes so it's pretty easy to be above average
>>101770444>>101770887>>101770923How new are you?
>>101770916100% agree
>>101770932been hanging out here since the early 1.5 days
>>101770916the consisten worst gen poster is that horse/centaur poster who flood the threads with melting garbage
>>101770978How often are you in the thread?
>>101770916for someone who works on this shit full time he sure is bad at making anime fox girls
>>101770444is this flux btw?>>101770887this guy is heckin based
>>101771011>blurry overfried shit
https://suno.com/song/73dfeb9b-d1c2-4654-9b97-606eacd3a383really impressed by lyricsgpt desu, ive only changed a couple wordswhy tf are suno's lyrics so shithttps://huggingface.co/bigdata-pw/lyrics-gpthttps://github.com/bigdata-pw/LyricsGPT
>>101771114Probably based off chatgpt which is kind of the same
>>101771254yeah i think they use openai's api, it just produces lame af rhyming couplets instead of actual lyricsmad how 125m params trained from scratch on only 400k tracks beats itnext version will be trained on more tracks and support starting from a short description desuim gonna look into what model to base audio generation off desuhttps://suno.com/song/c5fb3e77-1fbc-4e1b-bef9-bed384a6aed2
Will Ran ever come back?
could I humbly request an anthropomorphization of the White Whale, Moby Dick, into a cute anime girl?
suno down >:(
im genning things and sometimes an image takes 30 seconds to a minute but other times it takes 3 minutes or more. same prompt length. i dont know whats going on. is my graphics card dying or something
>>101771607go for udio, it's better
>>101771551Hi ran
>>101771769>julien moment
Good afternoon, anons! I hope everyone is doing well :]
me on the right desunice productive day considering i only had 3 hours sleeplikes on my model and datasets would make it better though>>101771795but i already paid for suno coinshttps://suno.com/song/d7ca010e-6451-4048-a916-6d05601ae10eill go for local soon:tm:
>purple schizo deployed
I thought the knife looked nice.
>>101771946hi pw. rare sighitng>>101771978>processing | 9.4/461.0swaiting patiently for my new lyrics. I don't know what any of these settings do
>>101771993You don't enjoy reading her tangentially on-topic conversations?
>>101772081>her
>>101772017It does look nice! I like the outfit too :]>>101772031Good afternoon, big data anon!>>101772041Heya Debo! I decided to wake up early today hahaI got some inspiration to work on my game today! I dunno if I ever mentioned it, but my old game is gone :[It was on my external hard drive and it fell like a week ago LOLOddly enough tho that kinda kickstarted my motivation!>>101772046Heya Quokkanon!
>>101772117Ask Ani
>>101772017how and what
>>101771978I broke it>>101772126>It was on my external hard drive and it fell like a week ago LOLoof, that sucks. hope you didn't lose too much important stuff. I always save projects to the cloud (usually github) cuz I'm paranoid of losing work>I got some inspiration to work on my game today!there are some new 3d gen options available if you're interested
>Rainygirl trolling again
>>101772041best to run it locally desu. hf's cpus suck. the demo on github is set up for cuda but you can just remove .cuda() to run it on cpu. should only use about 1gb ram or vramon my 3060 i get like 50-60+ token/sthe defaults should be good, might change the default repetition penalty to 1.3 though, seems to help>>101772126good afternoon! it's 9pm here lol>>101772230yeah thats probably the repetition penalty, try messing with that and the temperature up or down a bitit's not actually trained for multiple artists either, just thought i'd put that in there as an experimentand because it's trained from scratch it doesn't have any general knowledgev2 will be better. soon:tm:
>>101772230It's funny cause I was making room on my pc and thought, "Oh, it's on my external, so it's ok!" then a couple days after that it broke LOLI almost got rid of my 3D folder too, i'm SO glad I didn't!I honestly didn't lose too much! It was just a small scene you can walk around with NPC interaction hahaha! I'm also gonna try using Unreal again, and it was in Unity beforeOhh! Yeah! There's this new one called Rodin that works really well!! Might use that for small background objects and stuff! :]>>101772315Oh wow hahaha! Good evening :]
>>101772218Check the file name.
>>101772126Hello PW
>
>>101770995not that often lol, it comes and goes, there will be weeks where i post every day and months where i don't visit once>>101771011thanks debo, i think your gens are bad a lot of the time but i appreciate that you try different things and share stuff even if it is garbage (and sometimes your gens are top tier)>is this flux btw?no, it's magnum32b prompting SDXL turbo -> some custom image processing i wrote, i'm not really sure what i want to do with it but i'm thinking of trying to make like a AI powered paranormal investigation twine game? i'd like to use flux but it's too slow, i toyed with the idea of making it like, a correspondence game to give me time to gen because flux is a *LOT* better at complex prompts but big LLM is more important than big T2I to make something compellingi probably will try to do schnell + 7-9b llm at some point tho (picrel, flux wouldn't have made more than one cabin)
>>101772703>not that often lol, it comes and goes, there will be weeks where i post every day and months where i don't visit once Ah, so then it's understandable that you don't find debo to be a nuisance because you take frequent breaks. I wish debo would do the same but alas, where else would he go? He has nothing else going on in his life.
>>101772017lol the enormous pocket knife>>101772324>>101772230would like to try a 3d gen thing, any good recommendations? most of these things can only do a single object right, like it would fail at "mountainous terrain" or "a city block"
>>101772779Rodin worked really well for me! I haven't tried anything big like that yet, just character stuff and a burger hahahaIt might work tho! It just takes an image and turns it 3DIt'll give you a preview before it renders it too!
>>101772041>Look in thread>see this abomination Jesus Christ >>101772081Only a "Her" when onlineThis is a cope thread for a few mentally ill outcast
>>101772946
very schizophrenic
>>101772946Looks like her fella knocked a tooth out her mouth
>Julien
>>101773026cause and effect from sucking too much BBC
i miss koffi hope he is doing well
Can I train a SDXL Lora using Prodigy with a 3090?
>>101772779>any good recommendations?SAI has this new 3d thing out. dunno if its any good but its the newest thing on the radarhttps://stability.ai/news/introducing-stable-fast-3d>like it would fail at "mountainous terrain" or "a city block"I've seen some projects attempt to do environment gens like that but I haven't really paid particularly attention to it. you might be looking to fuck around with NeRFs and gaussian splatting at that point>>101772915any chance you could show some of the rodin gens? I'm curious what you've been getting out of it>>101773105I was wondering yesterday cuz we haven't seen him in a few days..
is there a character limit for prompts?
>>101773279not a character limit but a token limit. it will depend on your workflow and your model
>taking any kind of advice from deboNewfags will learn one day
>>101773373do you really have nothing else in life?
>>101773386He doesn't and giving him attention is the wrong move anon
>>101773373do you realize why you basically have your own containment general?
>>101773564You didn't gatekeep hard enough and now he has this space. It's perfectly fine just go to the other general like everyone else is doing
>>101773239Here's a PW chibi :]
>>101773728How does it feel to be laughed at every day?
>>101773239and here's a burger haha
Check out my no gen
I miss schizo anonHim being not here really makes the general worse
>>101773750Sampler/scheduler? How many steps?
>>101773754They are right here>>101773728>>101773239
>>101773239>>101773728>>101773744Buy an ad
>>101773728cute! are you able to add bones or whatever to it from there? maybe there's value in prompting for T-poses?>>101773750damn, thats nice. you forgot to add arbitrary trolling to your post tho xD
>>101773779You forgot to add effort to your gens :)
>take a huge art repository (very different things within)>Divide it between (stuff I like and stuff I don't)>Train a lora on itWould it be gibberish or could it zero in on the general aesthetic thing?
>still no dedicated chip for sd>fuck cloud-only4k 8k ultra sad
>>101773791Why are they always missing teeth BTW?
>be d*b*>*schizo spasms*>end of story
don't ask me if this line is about you
This general needs to die already
>>101773799I've seen a lot of loras on civit with that theory. "built this lora off a bunch of pictures I like". seems like it works if you know what you're doing
>>101773940should I start hanging out in ldg?
>>101773944Try harder
>>101773744>>101773728is there a thing like this that's fully local?
>>101773959Stay in your containment general please
>>101773959A rentry about you will be added to OP.
>>101773962you can run triposr in comfy, maybe newer text-to-3d models too
>>101773959God no, stay here thread schizo
>>101773981Too many to count in this one
Fucking finally...https://files.catbox.moe/hsi2zn.pngNow that we can increase the CFG value on flux, see here:https://reddit.com/r/StableDiffusion/comments/1ekgiw6/heres_a_hack_to_make_flux_better_at_prompt/It's time to look at the sweet spot between the Guidance scale and the CFG, and for that we'll use that prompt:>A drawing of Hatsune Miku with dreadlocks and a black skin showing her fists on the streetThe goal here to check whether the photo shows Miku having a black skin + dreadlock, if neither is present, the photo is eliminated.Here's my observations:- CFG = 1 won't change Miku's skin or add any dreadlocks. This is one of the reasons why it's important to have a higher CFG if you want Flux to perform as well as possible in terms of prompt understanding.- The sweet spot seems to be at CFG = 5, that's the minimum CFG where we got the most success.- Having a low guidance doesn't work well with a high CFG, the images gets broken.- High guidance seems to make the model worse at prompt understanding, for example at Guidance = 5.1, there hasn't been a single image that was successful.
>>101773779Thanks :DYeah i'd imagine putting it into blender or something would be easy enough to add bones to! Yeah! Doing a T-pose would be a good idea! I was just testing it hahaOmg that looks amazing! I love the hair and outfit!>>101773962Not that I know of, but Debo responded with something! That's something I kinda wanna look into as well!I don't really wanna pay for rodin, but also wouldn't really mind if I found it useful enough!
>niggers tongue my anus :]
>>101774036who are you voting for in november?
>>101774018CFG <1 makes it take 2x as slow tho right?def gonna have to keep using flux for this project, ugh, sdxl would have never got this prompt right
>>101774214>CFG <1 makes it take 2x as slow tho right?CFG <1 and CFG > 1 makes it 2x as slow yeah
3day vacay is over.ready for technology discussion.
>>101774407You make the same garbage every day
still no idea wtf i'm doing, gonna go for a walk and think about it
>>101774431incorrect
>>101774448Meds now!
>>101773386
>>101774407nice to see youname your three favorite technologies
How long should it take for me to train a SDXL Lora on my local machine with a 3090 with 8gb VRAM? The last few attempts I've done have shown ETAs of over 12 hours, but I'm not sure if that's due to the training itself or me not configuring the settings correctly.
>>101774486a.i., video, audio
>>101774510Ugh, it's probably your settings, I trained only one (successfully) and it took me 2-3 hours on 3060 12GB.
>>101773959So you admit that you're a unwanted poster. Go figure you wasted two years of your life being a waste of space.
ran...
canny working in flux means you can turn literally any image into a flux prompt, since it's basically a 1:1 lineart controlnet
>>101770274Holy crap. I hadn’t looked in a while and now vast/runpod prices have totally cratered. The hype cycle is ending finally.
>>101774724I kind of like the idea of that and was thinking about it earlier myself. But ComfyUI gives me aids so RIP.Doing a Flux prompt then cn or img2imging it with superior pony/1.5 would be nice.
HOw do you make a split text (used as a prompt) go through the full process one-by-one in a queue instead of being bundled together in ComfyUI? Is it even possible? Instead of making every one go through the first sampler together, then the next sampler together, just have one go through the full thing, save and then do the next one? It tends to crash at the end of the process if you put through ~50 images or more and my current workflow pretty much means you're gonna be putting through hundreds of prompts in a single queue for the desired result. it's driving me crazy over here
>>101773959You already tried spamming the news there and they wouldn't have it
I'm hesitant between "Flux is kinda trash" and "Flux is the greatest open source advancement we've had this year"
Enjoying Flux.
I wish I didn't have to be an autistic rocket scientist to figure out how to add controlnet to comfyui
>>101774546If it's not too much trouble, could you post your settings? No matter which guides I follow or what settings I tweak, I keep either running into memory errors or 14 hour ETAs
>>101775616Sorry, it's been around a year ago, I don't have them anymore.
insane how sd 1.5 is still the obvious #1 best choice for Real Art
It took me long enough, but I got my character to work in UE5 finally! HahahaI'll see you in like 4 hours, anons! Gonna go out for a bit :]
fluxguidance node acts like a "realism" slider1.5 is "realism"3.5 default5-6 leans more towards animethis is 5.0 but also said "pixar style".
https://youtu.be/hpLwIB101qQ?si=G9LabzqnDwnkVDJaflux= dalle3.5?
flux anime looks like SHITworthless model
>>101776068Flux everything looks like shit, the only thing it has going for it is prompt comprehension. I'd prefer looks good but doesn't listen, personally. Flux looks like it was aesthetically tuned by a team of autistic jeets who based it off top Civit gens desu
>>101774510>ETAs of over 12 hoursIn the beginning, the presets people were sharing took that long and that's about right. People have settings now that are a lot faster, but you'll have to try to find them. I only attempted to make like 2 loras and gave up. I use the civitai trainer. https://github.com/bmaltais/kohya_ss/tree/master/test/confighttps://github.com/bmaltais/kohya_ss/tree/master/presets/lora
>>101774510use the default settings in kohya, even with as little as 10 images I managed to make a simple 2B lora just to test if instance prompts would work. dont do something crazy like 100000 epochs.
sdxl + mspaint
>>101776869
>>101775542Some of the gens I've seen are starting to make me interested in downloading it but it seems inconsistent going by the HF demo I tried. And the elves are not very cute, the girls have butt chins too. >>101776230>I'd prefer looks good but doesn't listen, personallyWe've already got those though. Hopefully someone fixes Flux's styles.
https://civitai.com/models/628763/androflux?modelVersionId=702968
>>101777355old
>>101777355pee pee
>>101777362Well, I didn't know if anyone had seen the flux dick lora yet, sorry
>>101777084Yeah anime is pretty bad with it, at least with the default setup
>>101777416No, it's quite good with anime, just proompt it properly and use Dev, not Schnell.
tick tock comfy....
trying to get a girl climbing into a mecha but so far I've only gotten a robot being inappropriate with a child
>>101777472Yeah flux gives me really young looking 1girls a lot. I had to stop using "girl." I tried woman, but then they were old. I finally used 18 year old and that seemed to work somewhat
>>101777560yeah idk how the (((media))) haven't noticed yet, flux can gen very realistic [REDACTED] nude photos, and they don't look mutated like sd 1.5 or something, and this is just the base model
flux doesn't know what a mecha / mech-suit is>>101777560I'm noticing. trying to figure out how to get a very slender girl without her appearing 12
>>101774616>Can someone share an example workflow with high cfg and the cosine thing? I have no idea what it is.Take this anon: https://files.catbox.moe/nxz32g.png
>>101777416Mostly it's the elf ears, they are not cute at all.
>>101777592neatoif you're the anon that did the comparison grid, ty for your gpu cycles
>>101777472>>101777560>>101777578>>101777585It's a product strategy decision to force commercial users to use flux.pro with negative prompts. If you embed flux.dev or flux.schnell into your user-facing tools, someone will figure out how to generate CSAM with it.Welcome to the actual state of corporate ethics
>>101777715Anon, but flux pro doesn't have a negative prompt even on their API, are you fucking retarded?
>>101777715>someone will figure out how to generate CSAM with it.>what is NSFW checkingBoth FAL and Replicate already have them enabled by default
>>101777712thanks anon, it was a pleasure o/
>>101777735are any of these dynamic thresholding params worth tinkering with or is this basically *the* settings?
>>101777715>It's a product strategy decision to force commercial users to use flux.pro with negative prompts.we can already use negative prompts on dev now:https://reddit.com/r/StableDiffusion/comments/1emow5p/finding_the_sweet_spot_between_guidance_and_cfg/https://new.reddit.com/r/StableDiffusion/comments/1ekgiw6/heres_a_hack_to_make_flux_better_at_prompt/?sort=confidence
>>101777753- for the "Cosine Up" on both "mimic_mode" and "cfg_mode" it's the optimal one, I found it with this XY plot:https://files.catbox.moe/b4hdh0.png- scaling_startpoint = ZERO instead of MEAN is mendatory or else your picture will look like complete shit- variablility_measure = STD give much smoother pictures than AD, that one shouldn't be touched either- mimic_scale at 1 is supposed to stay like this, it's mimicking CFG = 1Imo, maybe "separate_feature_channels" and "interpolate_phi" could be touched, if you find something better with those 2, just tell everyone about it.I also tried changing "sched_val" but it did nothing kek
>>101777560Can confirm, girl is definitely literal. "Lady" works fine.
>>101777084This person is right>>101777425Schnell worked awfully for me, but I started using DEV and it works perfectly fine. Also switching from the basic prompt to CLIPTextEncodeFlux helps out with styling.
It's amazing how blurry this one is, it truly adhered to the "grainy low quality" part of the prompt.
>52s/tumm maybe I dont care about negatives that much
>>101777456Yeah I can't wait for him to deal the final blow to A1111 then abandon forge again once he gets tired of dealing with what's left of the userbase.
>>101778042trani
>black skin
>rent free seethe posting for two years
>>101778042what happened there
>update comfy>its slowerwhy
>>101777818>>101777753ok I think "interpolate_phi" is working better a 0.7, less saturation than on 0.8
>>101778239I can't fuck with negative prompting cuz it nukes my gen times :(
So one thing I dont see is anyone talking about "safety measures" ot censorship in flux. I remember that being a big deal with the SD releases. Whats the deal there? I dont do NSFW gens its just odd to me nobody brought it up.
>>101778204This is a 4 step model rendering at native 1920x1088 is what happened.
is there any way to do txt to video for ponyxl? all the guides i see even from a few months ago use 1.5.
>>101778274Love this
>>101778281Comfyanonymous is so toxic it's wonder why he's still employed.
>>101778042voldy doesnt give a shit about you dude, Illya on the other hand.... lol tick tock comfy
I guess I finally got a girl in a mecha, but at what cost>>101778280BFL hasn't been sued yet. it took a while for the lawsuits to start piling up on SAI too
>>101778280"safety measures" was a head cannon story that Emad created on his head, currently there are NO LAWS against AI generated content
>>101778342Its so weird how nothing you said answered the question at all
>>101778368First time?
>>101778349That's not what I asked. Is the model censored or "guided" to avoid porn/racism etc like was done in SD?
>>101778368>>101778368>"why doesn't anyone talk about safety measures?""because safety measures were a response to getting sued over 'unsafe' models">"wtf why didn't you answer the question"are you trolling or retarded? I don't get it
>>101778382>>101777879
>>101778420Its funny how you dont understand hyperbole
>>101778434Was going to reply the same keke
>>101778442Lol
>>101778420>"because safety measures were a response to getting sued over 'unsafe' models"that's like suing a pencil company because some mangaka drew a loli
>Miku dancing on a pile of vegetables in a cyberpunk city with a big hologram of miku in the sky
>>101778458but then the pencil maker started putting all its resources into making 'safe' pencils in response
>>101778382No, is not, obviously all models try to mitigate "harmful" content (i.e. CSAM), but not the way Emad enforced it, "safety measures" was something that Emad created on his head
>>101778477We dont need you to elaborate on homilies debo, we got it. You taking it further is just stupid
>>101778270can you try those new settings quickly and tell me if there's an improvement over the old one?
oh shit, closest yet>>101778502sure, but it'll be a few minutes
>>101778540go for AD instead of STD instead, I think I found the sweet spot
>>101778473
>>101778554already started one with STD with those settings. gen speed is still way slower. but another part of the problem is I updated comfy to pull in the new nodes and now comfy is 3x slower too for some reason afterwards. idk whats going on
>>101778540>>101778502yeah, MEAN + AD gives the soft lightning compared to the rest
>>101778618I saw someone else complaining about the new update of comfyUi, looks scary to me I won't pull for now kek
>>101778631I think it works but its too painfully slow for me to experiment with more>>101778683seems like somethings fucky :(
>>101778713>seems like somethings fucky :(can't you revert back to the preview commit? ask bing chat on how to do that
>>101778734I just hit the "Update All" button so idk which commit I moved from. there's a dozen or so commits over the last two days I dunno which ones I need and which I don't. I'd have to try one at a time and I don't have the energy for that
>>101778780desu it's not that hard, just go for the commit just before the one you currently have
Good evening, anons! I hope everyone is doing well :]
>>101778304He's not
>>101778810>git reset>start comfy>start gen>load models>wait two iterations>"nope, still slow">start over until the correct commit is foundthats like 5m per attempt. I don't have the energy for that rn. I'm sure comfy will fix it, right?>>101778817hello. are you doing a second pass with a different model or is this all flux?
>>101778893>>start over until the correct commit is foundyou don't need to jump one commit from one commit, try like 5 commits before and maybe you'll get the good one quicker
is there a way to use inpainting with flux yet?
>>101778893It's all flux hahaHeya Debo :]
>>101778915Someone said they did earlier. Check /ldg/
sd mech paradise
>>101778945thats such a rich style. how are you prompting for it?>>101778957>mech paradiseI wish. I just get normal ass robots
Where do I go to use flux? The cloudflare one doesn't seem to work now.
>>101779004there are huggingface spaces you can use for a good bit before they cut you off. I think there's a few options if you're OK with making an account too
>>101778966Thanks so much! :DIt's just my normal proompt with whatever I think of at the time LOL>[PW proompt], excited facial expression, determined eyes, cute smile, air of motivation, sitting at computer, working on unreal engine, making a video game, fist pumphahahaha I don't think proompts need to make sense, as long as they work, right?
>>101779029Will there be another cloudflare one any time soon?
>>101779030>>[PW proompt]that must be where all the extra charm is coming from>>101779038idk, webapp anon hasn't been posting any updates. I think he's limited by the api keys he can get his hands on and might not be able to get any more
>>101779004Hugginface demo is your optionhttps://huggingface.co/spaces/black-forest-labs/FLUX.1-devBut I'm sure the high step count gives you less gens.>>101779038I hope so, it was the best way to use FLUX
>>101779059LOOOL no way hahaha! I'm sure if you give the rest a go it'll be pretty much the same :]
>>101779144I'm too busy trying to get a girl to climb into a mecha cockpit. difficulty: impossible
That's insane how well this is workinghttps://files.catbox.moe/kqaf0y.png
>>101779167with higher CFG, flux would respond to your prompt better, but yeah I know you have some commit issues at the moment
>>101779068
Flux ankha :)
>>101779204rawr
anime girl paints anime girl:
>>101779167That sounds fun! I'll give it a try too hahaha
>>101779231nice
>>101779231neat, I gave it a different artstyle prompt and it worked:A blonde anime girl drawing a painting. The painting is of a blonde anime girl. The painting has a manga artstyle and is in black and white. The setting is an art studio.
>highres>flickr>_blolthose are 1024 longest edge
>>101779231I love this!! :DSo fun hahaha
https://files.catbox.moe/u19cqn.png
>>101779059Do you know how to get oil paintings? I’ve tried using this prompt that someone else used, but the images still look off.oil painting art style, renaissance painting, oil paint, portrait of a dog
>>101779261>the painting is of super marioworks
>>101779231>anime girl paints anime girlThat's just a self portrait, anon
stability ai painting:
>>101779276I've never really done much in the way of oil paintings but I tried this>oil painting of a portrait of a dog in the style of renaissance canvas art, the oil impressionist technique takes an abstract form to paint the dog in a creative and artistic expressionfor whatever reason "portrait of a dog" seems to overpower the rest of the prompt
A higher CFG helps Flux to make celebrities look more alike, nice.https://files.catbox.moe/wthai4.png
A blonde anime girl drawing a painting. The painting is of a man crying while holding a picture of a girl lying on grass, he is saying "why SD3" in a speech bubble. The painting has a manga artstyle and is in black and white. The setting is an art studio.
>>101779356thats the whole point of dynamicthresholding, being able to use a higher cfg without burning the image, how much of a newfag are you?
>>101779364>>101779364>>101779364
:egg:
>>101778618>>101778683Yeah same, just did git checkout 2a02546e2085487d34920e5b5c9b367918531f32 - this commit werks just fine
