Previous /sdg/ thread : >>101582499>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 news07/26/2024>ComfyUI implements native Hunyuan-DiT support, packages model to single filehttps://comfyanonymous.github.io/ComfyUI_examples/hunyuan_dit>Fooocus v2.5.1 Updatehttps://github.com/lllyasviel/Fooocus/releases/tag/v2.5.1>HVM-1: Video models trained on ~5k hours of human-like video datahttps://github.com/eminorhan/hvm-1>Move and Act: Enhanced Object Manipulation and Background Integrity for Image Editinghttps://github.com/mobiushy/move-act>LLMImageIndexer: Use local LLM to create descriptive metadatahttps://github.com/jabberjabberjabber/LLavaImageTagger/>Generative AI for Krita: Version 1.21.0https://github.com/Acly/krita-ai-diffusion/releases/tag/v1.21.0>Rope-Live: Customized Rope for Streaming https://github.com/argenspin/Rope-Live07/25/2024>ViPer: Visual Personalization of Generative Models via Individual Preference Learninghttps://viper.epfl.ch>ComfyUI-Kolors-Translator: Translate prompts into Chinesehttps://github.com/BetaDoggo/ComfyUI-Kolors-Translator>ComfyUI-FollowYourEmojiWrapperhttps://github.com/kijai/ComfyUI-FollowYourEmojiWrapper07/24/2024>SV4D: Dynamic 3D Content Generation with Multi-Frame and Multi-View Consistencyhttps://sv4d.github.io>Open-Sora-Plan Report v1.2.0https://github.com/PKU-YuanGroup/Open-Sora-Plan/blob/main/docs/Report-v1.2.0.md>Official global launch of Kling AI's International Version 1.0https://klingai.com>INF-LLaVA: Dual-perspective Perception for High-Resolution Multimodal LLMshttps://github.com/WeihuangLin/INF-LLaVA>FoRA: Low-Rank Adaptation Model beyond Multimodal Siamese Networkhttps://github.com/zyszxhy/FoRA07/23/2024>Mammoth - Extendible (General) Continual Learning Framework for Pytorchhttps://github.com/aimagelab/mammoth>Differentiable Convex Polyhedra Optimization from Multi-view Imageshttps://github.com/kimren227/DiffConvex>Cinemo: Consistent and Controllable Animation with Motion Diffusion Modelshttps://maxin-cn.github.io/cinemo_project
>mfw Research news07/26/2024>RegionDrag: Fast Region-Based Image Editing with Diffusion Modelshttps://arxiv.org/abs/2407.18247>Imagine yourself: Tuning-Free Personalized Image Generationhttps://ai.meta.com/research/publications/imagine-yourself-tuning-free-personalized-image-generation>Sparse vs Contiguous Adversarial Pixel Perturbations in Multimodal Models: An Empirical Analysishttps://arxiv.org/abs/2407.18251>Quasar-ViT: Hardware-Oriented Quantization-Aware Architecture Search for Vision Transformershttps://arxiv.org/abs/2407.18175>GaussianSR: High Fidelity 2D Gaussian Splatting for Arbitrary-Scale Image Super-Resolutionhttps://arxiv.org/abs/2407.18046>RestoreAgent: Autonomous Image Restoration Agent via Multimodal Large Language Modelshttps://arxiv.org/abs/2407.18035>AttentionHand: Text-driven Controllable Hand Image Generation for 3D Hand Reconstructionhttps://arxiv.org/abs/2407.18034>Scaling Training Data with Lossy Image Compressionhttps://arxiv.org/abs/2407.17954>Guided Latent Slot Diffusion for Object-Centric Learninghttps://guided-sa.github.io/>Reasoning and Correcting Diffusion for HOI Generationhttps://alberthkyhky.github.io/ReCorD/>Amortized Posterior Sampling with Diffusion Prior Distillationhttps://arxiv.org/abs/2407.17907>FlexiEdit: Frequency-Aware Latent Refinement for Enhanced Non-Rigid Editinghttps://arxiv.org/abs/2407.17850>DragText: Rethinking Text Embedding in Point-based Image Editinghttps://arxiv.org/abs/2407.17843>How Lightweight Can A Vision Transformer Behttps://arxiv.org/abs/2407.17783>Quality Assured: Rethinking Annotation Strategies in Imaging AIhttps://arxiv.org/abs/2407.17596>Diffusion Models for Multi-Task Generative Modelinghttps://arxiv.org/abs/2407.17571>ReDiFine: Reusable Diffusion Finetuning for Mitigating Degradation in the Chain of Diffusionhttps://arxiv.org/abs/2407.17493>A Survey of Accessible Explainable Artificial Intelligence Researchhttps://arxiv.org/abs/2407.17484
>>101590591how was the movie
>>101590591You're alright. Don't go to Paris tomorrow.
>>101590550That's a shame...
Why does the American government keep tall white alien waifus from us? Do they want to have em all for themselves?
goo night
>>101590882Night night.
>>101590882Goodnight.
>>101590893I like the film grain, good aesthetic
>>101590593I liked it more than i thought i would, it was pre fun.>>101590626Kek
>>101590965Thanks, it's definitely the look I'm going for with this latest batch (phone camera/low quality). Adds a bit of extra realism I think.A lot of the upscalers don't like it though, and it's just pure luck that they don't ruin the whole image 90% of the time. Still trying to find consistency when it comes to that.
>>101590480>Hunyuan-DiTCan it run on 6GB vram yet?
>>101586231>https://files.catbox.moe/7662s9.mp4Nice, I didn't know it worked with video. I used SVD and it's not as good as the cloud models, but I made it work with liveportrait.How much VRAM do you think these video models need? (runway, kling, etc.)
Who posts the news when news guy is sleeping?
>>101591487No one actually reads news beyond pointing out absurd headlines like>Rope-Live: Customized Rope for Streamingwe people do repost the new in their absence it's just a thread of news posts
217m from ~95 dayshalf way to Diffusion-1B with the other setsmight do some more on ait, either carry on debugging why stable-cascade-prior works and stable-cascade doesnt or just test another modelits a weird bug where a tensor gets removed from the graph or overwritten or something, tracked it to up_blocks_repeat_mappers being >1 compared to -prior, decoder works with it set to 1, -prior doesnt work either if i change the config to 2no wonder meta abandoned the project desu
>Model recognizes every cartoon character I can think of>Except SpongebobSo annoying
>>101591840Try>Mr. Sponge Bob, sponge with big cartoon eyes, half human half sponge,
>>101591865>>101591840>Mr. Sponge Bob, sponge with big cartoon eyes, half human half sponge,Works for me, same exact prompt...
>>101591865>>101591890Took me a bit to get it. I had a couple already queued to finish.Let's see what I get.She's cute at least.
>So, about your application...
Last one. This is CinematicRedmond model btw.
This roasted chicken isn't roasted at all! lololo
i'm bored of portrait alreadythere's very little you can do with it
Need some tips on training LoRAs.I'm doing a concept LoRA based on a niche fetish (if you really must know, google r34 xxx urethral insertion). My number of training images is large and I'm very pedantic about the quality of the dataset.1. Should I keep everything on 1012x1012 or is the aspect ratio whatever?2. Captioning. Should I focus on the concepts only or should I go fullblown autism like the booru taggers?
>>1015925901. Aspect ratio should be rectified and anally inspected to be exact same.2. Better go autism.
>>101591840>SpongeBob Squarepants
so is this now forge level performance or what
>>101592888if they didnt fix the bugs in it then its slower than 1.9.4
>>101593216>tfw no bizarre alien gf who lives in dark basement and only eats rodents
i miss schizo anon
>>101591840
>>101593482Are you that data hoarder?
>>101590306how do I sandbox SD on Windows so a rouge github tranny doesn't pwn my machine?
>>101593518I think you should not bother with genning images at all if this is the first question what comes to your mind.
>>101593512yh>>101593518>red scare
>>101593538I've been genning for months, and ever training my own loras. I just think it isn't smart to run so much ever-changing code without a sandbox.
>>101593518airgap your inference machine
>>101593597Well, you could use a virtual machine with gpu passthrough but I'm sure it will cause complications. Or maybe not. You could test it.
>>101590306anyone know anything about goomaxing?
Mght as well post second one.
>>101593657yeah, because I have so many machines to spare...>>101593668I was considering it, but it's an atomic option. don't want to deal with two graphics cards etc. I guess I could run it as some restricted user, but it's Windows so I wouldn't expect this helping much.
>>101594025You don't need two cards because you already (in most cases) have an integrated gpu on your cpu.I'm not familiar with virtual machines and gpu passthrough anyway but I know using integrated one is possible and you won't need two cards. Besides, you could have something very cheap as a second one anyway...I'm sure an ordinary firewall configuration will help to isolate any unwanted traffic, for Windows I recommend SimpleWall by Henry++.
>>101594095How did you define this painterly style? I just use>oil painting by Brian LeBlanc for example but I find it hard to get specific styles like this. They all look like 'something but not quite'. It should be doable with almost any SDXL checkpoint.
mornin
>>101594192good morning, i hope you are feeling well
>>101594143Sometimes i use realistic , or oil painting or realistic anime, but it's quirkyt's more like the combination of euler, simple, and the right amount of denoise and cfg parameters that triggers it
>>101594192GmornWas listening to https://youtu.be/tpKCqp9CALQ?feature=sharedThought about your muse in this situation
>>101593319>bizarre alien gf who lives in dark basement and only eats rodents
>>101594062thanks anon, some good advice there. not all of it applies to my situation (for example, I don't have integrated GPU in my 5950x), but thanks anyway. I guess I was hoping Windows would have some better sandboxing mechanism by now, without resorting to virtualization. for now I will try using a restricted user. I guess it's marginally better than just raw dogging it as an admin user, kek.
>>101594192>want to gen a fox peeking his head up from the snow>get a 1girlthanks, ponyanyway, goo morning!
>>101594406I think you would be more vulnerable on Linux (as Python is integrated way deeper) than on Windows especially if you're using standalone venv things. I mean you are exaggerating this.
indeed, it is caturday, expecting lots of feisty felines aujourdhui
>cats are egotistical, look at me I'm a cat. lizard hybrid, look at its eyes.
>>101594639Very nice style combination, amazing.
>>101594660Kolors model with ipadapter, Art by (Theodoros Ralli:1.05) and (Michael GarmashJean Giraud)posted workflow for base gen and upscale last thread.
Art by (Theodoros Ralli:1.05) and (Michael GarmashJean Giraud)
Anyone that paranoid has something to hide
>>101594673My gpu is so slow I haven't bothered with ipadapter, lmao.Try adding >(art by Paul Lehr:x.x)He has very colorful paintings. Don't have a new example of this so I won't post any images.
>>101594702>>(art by Paul Lehr:x.x)thank you, mines a 12GB 3060,, need a 16GB now. I have multiple workflows to finish one image kek...
Not as strong as I wanted it to be but whatever.
>>101594841Very nice.>>101594702>(art by Paul Lehr:1.2)
>>101594889lmao well... I can swear it works on some things
>>101594923Super saturated, took away a bit of realism, but I also have other artists in prompt so most likely didn't transfer well, moved on.lol
i think trani is shit (as a human)
Any groundbreaking realism models for SD3?
>>101595087no one has been training for sd3 because of the shit license
>>101595309I'm wondering what has happened with these Chink models? I tried the latest Comfy portable with Hunyuan DiT but it couldn't recognize the checkpoint (all from the comfy example page). I wonder maybe it was because the portable wasn't updated.I mean aside from the initial propaganda I do think these might have potential when adequately trained.
Final.
>Conned by a Bangledeshi and cucked by a JeetWhat's next for Stability AI?
howdy
FInally,after so much struggle, i've made the scene i had in mind.Multiple characters are a pain in the ass,also making them fight
here it is
>>101595364appreciated
headache
>>101595511
>>101595364neat
>>101594978based schizophrenic thread guardian of frenship
Testing. I don't remember if this was done with albedoXl or vanillaSdxl. Next one is the same prompt with Redmond but upscaled with sd1.5 and also in different resolution.
>>101596038The aspect ratio makes it more squeezy.
>>101596052Original is better because it was upscaled with the same sdxl model.What was the point? I don't know there is no point.