Previous /sdg/ thread : >>100140236>Beginner UI local installFooocus: https://github.com/lllyasviel/fooocusEasyDiffusion: https://easydiffusion.github.io>Local installAutomatic1111: https://github.com/automatic1111/stable-diffusion-webuiComfyUI (Node-based): https://rentry.org/comfyuiAMD 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>Auto1111 forksForge: https://github.com/lllyasviel/stable-diffusion-webui-forgeAnapnoe UX: https://github.com/anapnoe/stable-diffusion-webui-uxVladmandic: https://github.com/vladmandic/automatic>Run cloud hosted instancehttps://rentry.org/sdg-link#run-cloud-hosted-instance>Try online without registrationtxt2img: https://www.mage.spaceimg2img: https://huggingface.co/spaces/huggingface/diffuse-the-restInpainting: https://huggingface.co/spaces/fffiloni/stable-diffusion-inpaintingpixart: https://huggingface.co/spaces/PixArt-alpha/PixArt-Sigma>Models, LoRAs & embeddingshttps://civitai.comhttps://huggingface.cohttps://rentry.org/embeddings>Animationhttps://rentry.org/AnimAnonhttps://rentry.org/AnimAnon-AnimDiffhttps://rentry.org/AnimAnon-Deforum >SDXL info & downloadhttps://rentry.org/sdg-link#sdxl>Index of guides and other toolshttps://codeberg.org/tekakutli/neuralnomiconhttps://rentry.org/sdg-linkhttps://rentry.org/rentrysd>View and submit GPU performance datahttps://docs.getgrist.com/3mjouqRSdkBY/sdperformancehttps://vladmandic.github.io/sd-extension-system-info/pages/benchmark.html>Share image prompt info4chan removes prompt info from images, share them with the following guide/site...https://rentry.org/hdgcbhttps://catbox.moe>Related boards>>>/h/hdg>>>/e/edg>>>/d/ddg>>>/b/degen>>>/vt/vtai>>>/aco/sdg>>>/trash/sdgOfficial: discord.gg/stablediffusion
>mfw Resource news04/23/2024>Invoke v4.2.0a2 Adds Regional Controlhttps://github.com/invoke-ai/InvokeAI/releases/tag/v4.2.0a2>AnyPattern: Towards In-context Image Copy Detection https://anypattern.github.io/>SVGEditBench: A Benchmark Dataset for Quantitative Assessment of LLM's SVG Editing Capabilities https://github.com/mti-lab/SVGEditBench>DMesh: A Differentiable Representation for General Mesheshttps://sonsang.github.io/dmesh-project/>PDM-Pure: Effective Purification in One Simple Python Scripthttps://github.com/xavihart/PDM-Pure>IDM-VTON : Improving Diffusion Models for Authentic Virtual Try-on in the Wildhttps://idm-vton.github.io/04/22/2024>MoMA: Multimodal LLM Adapter for Fast Personalized Image Generationhttps://github.com/bytedance/MoMA/tree/main>LLaMa3 Stable-diffusion prompt makerhttps://ollama.com/impactframes/llama3_ifai_sd_prompt_mkr_q4km>PhysDreamer: Physics-Based Interaction with 3D Objects via Video Generationhttps://physdreamer.github.io/>Training-Free Painterly Image Harmonization Using Diffusion Modelhttps://github.com/BlueDyee/TF-GPH>TV100: A TV Series Dataset that Pre-Trained CLIP Has Not Seenhttps://tv-100.github.io/>Hyper-SD: Trajectory Segmented Consistency Model for Efficient Image Synthesishttps://hyper-sd.github.io/04/21/2024>FlashFace Inference Code Releasedhttps://github.com/ali-vilab/FlashFace>ComfyUI MagickWand: Proper implementation of ImageMagickhttps://github.com/Fannovel16/ComfyUI-MagickWand>Moving Object Segmentation: All You Need Is SAM (and Flow)https://www.robots.ox.ac.uk/~vgg/research/flowsam/>Image Effect Scheduler Node Set for ComfyUIhttps://github.com/hannahunter88/anodes/>ComfyUI-Tripo: Generate 3D models using the Tripo APIhttps://github.com/VAST-AI-Research/ComfyUI-Tripo>Conditional Prototype Rectification Prompt Learninghttps://arxiv.org/abs/2404.0987204/20/2024>Basic Stable Diffusion API GUIhttps://github.com/ThioJoe/BasicStabilityAPI-GUI/
>mfw Research news04/23/2024>GeoDiffuser: Geometry-Based Image Editing with Diffusion Modelshttps://ivl.cs.brown.edu/research/geodiffuser.html>TAVGBench: Benchmarking Text to Audible-Video Generationhttps://arxiv.org/abs/2404.14381>Graphic Design with Large Multimodal Modelhttps://arxiv.org/abs/2404.14368>MultiBooth: Towards Generating All Your Concepts in an Image from Texthttps://multibooth.github.io/>Infusion: Preventing Customized Text-to-Image Diffusion from Overfittinghttps://arxiv.org/abs/2404.14007>Zero-Shot Character Identification and Speaker Prediction in Comics via Iterative Multimodal Fusionhttps://arxiv.org/abs/2404.13993>RHanDS: Refining Malformed Hands for Generated Images with Decoupled Structure and Style Guidancehttps://arxiv.org/abs/2404.13984>Gorgeous: Create Your Desired Character Facial Makeup from Any Ideashttps://arxiv.org/abs/2404.13944>MaterialSeg3D: Segmenting Dense Materials from 2D Priors for 3D Assetshttps://arxiv.org/abs/2404.13923>Accelerating Image Generation with Sub-path Linear Approximation Modelhttps://arxiv.org/abs/2404.13903>Regional Style and Color Transferhttps://arxiv.org/abs/2404.13880>Enforcing Conditional Independence for Fair Representation Learning and Causal Image Generationhttps://arxiv.org/abs/2404.13798>Object-Attribute Binding in Text-to-Image Generation: Evaluation and Controlhttps://arxiv.org/abs/2404.13766>ArtNeRF: A Stylized Neural Field for 3D-Aware Cartoonized Face Synthesishttps://arxiv.org/abs/2404.13711>Concept Arithmetics for Circumventing Concept Inhibition in Diffusion Modelshttps://cs-people.bu.edu/vpetsiuk/arc/#>PoseAnimate: Zero-shot high fidelity pose controllable character animationhttps://arxiv.org/abs/2404.13680>Rethink Arbitrary Style Transfer with Transformer and Contrastive Learninghttps://arxiv.org/abs/2404.13584>LTOS: Layout-controllable Text-Object Synthesis via Adaptive Cross-attention Fusionshttps://arxiv.org/abs/2404.13579
>mfw MORE Research news>Exploring AIGC Video Quality: A Focus on Visual Harmony, Video-Text Consistency and Domain Distribution Gaphttps://arxiv.org/abs/2404.13573>LASER: Tuning-Free LLM-Driven Attention Control for Efficient Text-conditioned Image-to-Animationhttps://arxiv.org/abs/2404.13558>Motion-aware Latent Diffusion Models for Video Frame Interpolationhttps://arxiv.org/abs/2404.13534>FilterPrompt: Guiding Image Transfer in Diffusion Modelshttps://arxiv.org/abs/2404.13263>PCQA: A Strong Baseline for AIGC Quality Assessment Based on Prompt Conditionhttps://arxiv.org/abs/2404.13299>Generating Daylight-driven Architectural Design via Diffusion Modelshttps://arxiv.org/abs/2404.13353>AdvLoRA: Adversarial Low-Rank Adaptation of Vision-Language Modelshttps://arxiv.org/abs/2404.13425>Mixture of LoRA Expertshttps://arxiv.org/abs/2404.13628>GScream: Learning 3D Geometry and Feature Consistent Gaussian Splatting for Object Removalhttps://w-ted.github.io/publications/gscream/>Hyper-SD: Trajectory Segmented Consistency Model for Efficient Image Synthesishttps://arxiv.org/abs/2404.13686>Iteratively Prompting Multimodal LLMs to Reproduce Natural and AI-Generated Imageshttps://arxiv.org/abs/2404.13784>PGAHum: Prior-Guided Geometry and Appearance Learning for High-Fidelity Animatable Human Reconstructionhttps://arxiv.org/abs/2404.13862>Mechanistic Interpretability for AI Safety -- A Reviewhttps://arxiv.org/abs/2404.14082>Towards Better Adversarial Purification via Adversarial Denoising Diffusion Traininghttps://arxiv.org/abs/2404.14309>GaussianTalker: Speaker-specific Talking Head Synthesis via 3D Gaussian Splattinghttps://arxiv.org/abs/2404.14037>Plug-and-Play Algorithm Convergence Analysis From The Standpoint of Stochastic Differential Equationhttps://arxiv.org/abs/2404.13866>Detecting and Mitigating Hallucination in Large Vision Language Models via Fine-Grained AI Feedbackhttps://arxiv.org/abs/2404.14233
>>100146563
i need to take remedial math classes to understand these papers
thread challenge:post a pic of the latent space you used decoded!
REAL THREAD CHALLENGE:Don't be a retarded avatarfag.
>Seething
>>100146698most are probably fake research anyway
>>100146719>>100146679i dont use detailers or anything on most of my images desu
Hi guys, I haven't been messing around with SD since at least August of last year. Has there been a major breakthrough? What model should I use for realistic image gen with auto1111. Thanks. <3
>>100146698Even a mere glance is more than what most are able to accomplish
what do you want your ai girlfriend to look like ?
>>100146844model or lora?
>>100146854Was XL out then? If not that was the big thing.
>>100146879realspice_v20
>>100146736I don't see any avatarfags ITT
>>100146868the img is great but the irises are uncanny
>>100146886Last model I used was epiCRealism
>>100146892how 2 into detail like that?
>>100146938budget mint pussy lmao
>>100146938Git gud
>>100146938try detail lora
>>100146945>>100146949Imagine trying to copy mint pussy of all posters
>>100146974Fr, copy someone who actually posts good stuff
nogens seething lmao
Anyone keeping track of more stuff that increases image quality/reduces inference time like PAG, Align Your Steps, Hyper SD etc?
>>100147097autoCFG reduces inference time but in my opinion can have a negative impact on quality
>>100147193>>100147232Nice
2many 1gorl
pixart-sigma is definitely weird with people, does it excel anywhere to justify the resource burden?
>>100147329I don't think she would eat egg unless commander Ikari told her to.
>>100147358Can you change the hair/eye color
>>100147278>the route i took when i escaped from his shack
>>100147097I am writing my own API interface for this reason. I can't keep track of the tools, options, models, loras. I need code and there isn't a UI that does this. I am hoping to get to the point of unit tests for everything, but it is a long way off.
>>100147374You're probably right
>>100146610>>PDM-Pure: Effective Purification in One Simple Python Script>https://github.com/xavihart/PDM-Pureartcels keep losing
>>100147542>I am writing my own API interface for this reason. I can't keep track of the tools, options, models, loras.I don't understand what an API does for keeping tabs on tools
>>100147673office cat lady
>>100147677cute momiji
PAG lets you get away with some high denoise during high res fix
>>100147599>deepfloydcan local gen run this? I thought requirements were stupid high>>100147619The UIs push the info from the images into data. Image storage works really well when generate close to the same thing every time. If you don't want to concentrate on one image for 2 weeks at time I feel you need to have more information up front. I can't be going back and forth from the UI to test using x/y/z. I have been moving between A1111 and Forge because they have different updates and tools. The gap in the APIs made a large step as of today (or maybe a few days ago, I don't update like I should). By standardizing how I call the API I can start running tests better and more logically. I have talked about this before, but I have a (kinda) front end that will pull from civit and generate a image card for the lora and change its behavior on how I classify the lora. Works 90% of the time now. With that in place I have been running some more complete tests with sigma (experiment of the week) in hopes to get some sane settings. I was tired of change 8 things at once, getting confused and giving up.
>>100147683yeah>>100147677very cute!>>100147783>can local gen run this? I thought requirements were stupid highhigh VRAM requirement yeah so it didn't do too well
>>100147730Interesting - what is PAG?
I hate coomers and indecent women so fucking much
>>100147826https://github.com/pamparamm/sd-perturbed-attention
>>100147819Thanks!And your cat lady being constantly pissed off makes her even cuter.
>>100147838chud detected
>>100147838I'm sorry you feel that way
>>100147841
>>100147882is this a new lora?
>a masterpiece impasto artwork depicting a beautiful swan floating in a serene pond full of plastic refuse and garbage, the swan looks very sad and spreads its filth-stained wings forlornlypixart-sigma, for non-humans, hm maybe okay
>>100147839does this work on a1111 or just forge/comfy?
>>100147916I think there's a version for all of them, I'm not using A1111 atm tho just forge
>>100147897it is actually! I baked it this morning. thanks for noticing senpai
>>100147923ok thnx
>>100147839Is there a short version of what it does - does it just make hires fix better?
>>100147942No, it's not hires fix. It's a -on your end relatively mild- variant to make sampling do a slightly more expected thing. Arguably an improvement on current sampling but not even super clearly so.
>>100147932Can she still :3 a lot?
>>100147965So, just turn on, leave on default settings and get better images? No matter which sampler I have selected?I'll try it out.
>>100148015people keep saying its "free" but its really easy to make it go haywire. its very sensitive
>>100148015> leave on default settings and get better imagesAt 1.5-2, this seems fairly feasible. But the effect also isn't that extreme.You'll see if it works for you. It's not like you're suddenly using cascade or pixart in terms of being able to fit in 1-2 more concepts, it generally just adheres a bit better to already feasible prompts.
Is it faster to batch gen or do each gen one at a time?
>>100147780Thanks, inspiring.
>>100148066Batch, to the limit of what fits into your VRAM.One at a time may be more practical depending on what you do and/or depending on VRAM utilization (you wouldn't be the only person to do something else too while larger amounts of images gen)
>>100147910>a masterpiece impasto artwork depicting a beautiful swan floating in a serene pond full of plastic refuse and garbage, the swan looks very sad and spreads its filth-stained wings forlornlyThis is from ella+base SD1.5 in comparison
>>100148001there were so many dangerously lewd outputs my penis is crying a little. that model you are using looks fun
>>100148120I really like these breasts
>>100148120Is this also based off the artwork of a pro trans artist?
>>100148120>dangerously lewd outputs my penis is crying a littleIt's "worse" on my end - many lewder outputs are that I can't post here. Plug in the text gen from this and maybe bypass the top left character node with your own character and you approximately do what I do:https://civitai.com/models/400395/wanking-setup-or-comfyui-workflowModel is the figure model by Wangka (former Miaoka) that I had mentioned a few times, absolutely an excellent model:https://civitai.com/models/400329/pvc-style-modelmovable-figure-model-pony
Afternoon anons
>>100148262afternoon
>>100148290fak u carl
>>100148120Basically, you could auto-lewd your smug cat with the textgen part of that setup.
I just downloaded SDUI and some LoRas. Do I need WD as well?
how do i make the character fit in the environment more, like, she looks pasted in...
>>100148152I do too>>100148220no. sorry to devastate you>>100148259>>100148320I do have a similar setup with wildcards and my nodes to do batches of random situations and art styles. it's good to share though so thanks!
>>100148309what?
>>100148407Perhaps your wilcard format is compatible, a bunch of these are a pretty good idea for pony derivatives.
>>100148038>>100148064I feel like the backgrounds are a lot more repetitive with it on. But all in all, it's not worse... but different. Interesting.
>>100148478>her last selfie
>>100148490do a grid, same seed and a range of 1 to 6 with PAG slider.
>>100148510I probably should. Do I need a scheduler plugin for that, or something? I currently use the default Forge settings for the most part. The thing about the repetitive backgrounds is the most obvious change I'm noticing.
>>100148559>I probably should. Do I need a scheduler plugin for that, or something? I currently use the default Forge settings for the most part.I'm an auto + comfy user, not tried Forge yet so you'll have to ask this guy again >>100147839 :)Having said that, doing 6 images manually with different pag values shouldn't take too long unless you're on Nvidia 9xx/1xxx card, or worse still... and AMD card.
>>100148407I'm not disappointed because you're finally acting normal
>>100148666>finally acting normalLol
>>100148713Small steps anon
Love distorts clear thinking
>>100148655succulent kitty milk. need it
>>100148789the beauty of SD3
>>100148821>>100148789This can't be comfy, gens looks objectively worse than a month ago
>>100148821>>100148820>beauty
Greetings, gentlemen
>>100148853Hi
>>100148821when are you back? I miss you>>100148853hello emma anon!
>>100148853hello, nice to see you. whatcha cookin up today?
>>100148908I came back yesterday.
How do I get a Pony model to generate a vampire with fangs and stuff?>as a vampire, vampire (fangs:1.4), bloody mouth, night time
>>100148894Hi there! What are you using for those colors?>>100148908Hello! So are you cooking a new LoRA every other week?>>100148916Yo. I'm trying to make cool and slightly evil witches today. I'm also looking for interesting words to spice up the background.
>>100148637So it makes things more repetitive and will sometimes deep-fry the image but also, I guess, more consistent...?
>>100148929I'm glad you had a safe trip. you should tell me about what happened later!>>100149015>So are you cooking a new LoRA every other week?just when I have the itch I guess
>>100149063Have you shared them somewhere? This style looks very cute. I've also missed anything you did in the past half year or so.
>>100148981Ok it was face restoration that removed them fangs, but how do I get a proper face without codeformer altering it?
Life could be dream
>>100146945>>100146974>>100147023literally fucking what are you talking/crying about? actually don't even tell me..It's literally just some random shit from an old dalle prompt, along with some extra stuff:score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, kiki delivery service, gothic fashion, 1990 anime screenshot, film grain, lo-fi, vibrant colors, petite girl standing on the rainy neon lit streets of hong kong, from behind, looking at viewer, looking back, sideboob, one arm covering breasts, lowleg panties, long twintails, neon green hair, <lora:BlueTheBone:0.3> <lora:BanashiPony:0.4>>>100149063>>100148929you should actually die
>>100149063ohhhh you mean you don't actually have a job in the anime industry anymore that would require your professional loraswooow, what a shame huh
>>100149153You do know blue the bone is built in right?
>>100149093cool image
>>100149153relax
>>100149190You raise a good question, I think Ani got let go when they realized he had no idea how to actually make loras which has been proven by this general
>>100149196yes, and?
I'm so based.
>>100149233naisu
Is using PAG better when you don't have a background? Because it really makes those worse.
>>100149252what's a PAG
>>100149252Setting it to U-Net Block to input imo does the best job
>>100149153rude
>>100149093>Have you shared them somewhere?not really but I should just get them uploaded somewhere. civit is kind of cancer though so maybe huggingface, mega or pixeldrain. I kind of want to include the dataset as well
>>100149265looks cool, what checkpoint is it?
>>100149327theif right here
>>100149205Thanks. Cool image too!>>100149291Then I'll wait patiently.
>>100149105Anyone? How do I...?
How do I connect my silly tavern to image gen? I got the extras tab, but it's not accepting my kobold cpp link. It works when image genning at just koboldlocalhost
>>100146798
>>100149355have you tried face restoration and then after inpaint the teeth?
>>100149395uhh thanks, I guess
>>100149401GFPGAN doesn't destory the teeth too much but the mouth still looks awful. If I inpaint the teeth they again look like this mess
>>100149430>her face when you toss her prostitute ass out in the rain after you give her a black eye for trying to steal your wallet and making fun of your teeny weenybad customer service, bad servicer treatment
>>100149493LOL!!!!!
>>100149402>nu-dog generalinb4 Sesshoumaru
>>100149441worked for me kinda, takes some luck with seed though
>>100149527Did you use GFPGAN or Codeformer?
>>100149527>beheads your vampire
>>100149553i didnt, i just inpainted it, using differential diffusion node, high steps and cfg, think you can also use ipadapter with some image of vampire teeth and would work more often
>>100149493uhh, definitely
>>100149603>high steps and cfgMaybe I try that, thanks for hint
>>100149626how i set up my workflow
>>100147164nice, I like the lightning and the style, is this sd 1.5?
>>100149657I am using A1111 but I think I can translate that workflow somewhat
>>100149728cool, did you use regional prompting or something similar?
>>100147973kinda disturbing
>>100149752nah. this prompt+model just likes to do multiple characters and somewhat dynamic scenes. its pretty hit or miss (picrel)
>>100149787cool composition tho, those hands could be fixed with inpaint or facedetailer
>>100146938people jumping like sharks lol
>>100147092>>100146923Asuka best girl but I can appreciate this
>>100149888GOOD NIGHT
>>100149938Gn anon
>>100149327Autism mix
>>100149938helloI'm sure you'll improve after watching a video
>>100149732 >>100149274 >>100149127cute
Next>>100149954>>100149954>>100149954
>>100150065Thanks!
>>100150059ok, four talon faget
>>100149742https://files.catbox.moe/vkoenj.png
>>100149902thanks