Previous /sdg/ thread : >>100185213>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 news04/26/2024>ConsistentID:Portrait Generation with Multimodal Fine-Grained Identity Preservinghttps://ssugarwh.github.io/consistentid.github.io/>Writers Guild of Canada Overwhelmingly Votes to Authorize Strike Over AIhttps://www.hollywoodreporter.com/business/business-news/writers-guild-of-canada-votes-to-authorize-strike-1235881245/>Phi-3-mini in ComfyUIhttps://github.com/ZHO-ZHO-ZHO/ComfyUI-Phi-3-mini>US teacher charged with using AI to frame principal with racist audiohttps://www.theguardian.com/us-news/2024/apr/25/maryland-teacher-ai-principal>Actually Using SORAhttps://www.fxguide.com/fxfeatured/actually-using-sora/04/25/2024>ComfyUI implementation of SimDA: Simple Diffusion Adapter for Efficient Video Generationhttps://github.com/ChenHsing/SimDA>Nvidia acquires Israeli AI infrastructure co Run:aihttps://en.globes.co.il/en/article-nvidia-acquires-israeli-ai-infrastructure-co-runai-1001477343>Learning Long-form Video Prior via Generative Pre-Traininghttps://github.com/showlab/Long-form-Video-Prior>GaussianTalker: Real-Time High-Fidelity Talking Head Synthesis with Audio-Driven 3D Gaussian Splattinghttps://github.com/KU-CVLAB/gaussiantalker>The state of AI for hand-drawn animation inbetweeninghttps://yosefk.com/blog/the-state-of-ai-for-hand-drawn-animation-inbetweening.html04/24/2024>Narrative Action Evaluation with Prompt-Guided Multimodal Interactionhttps://github.com/shiyi-zh0408/NAE_CVPR2024>PixArt adds 2K Sigma model and Lora supporthttps://github.com/PixArt-alpha/PixArt-sigma#breaking-news->FlowMap: High-Quality Camera Poses, Intrinsics, and Depth via Gradient Descenthttps://cameronosmith.github.io/flowmap/04/23/2024>HiDiffusion: Unlocking Higher-Resolution Creativity and Efficiency in Pretrained Diffusion Modelshttps://github.com/megvii-research/HiDiffusion>Invoke v4.2.0a2 Adds Regional Controlhttps://github.com/invoke-ai/InvokeAI/releases/tag/v4.2.0a2
>mfw Research news04/26/2024>NeRF-XL: Scaling NeRFs with Multiple GPUshttps://research.nvidia.com/labs/toronto-ai/nerfxl/>TI2V-Zero: Zero-Shot Image Conditioning for Text-to-Video Diffusion Modelshttps://arxiv.org/abs/2404.16306>Interactive3D: Create What You Want by Interactive 3D Generationhttps://interactive-3d.github.io/>MuseumMaker: Continual Style Customization without Catastrophic Forgettinghttps://arxiv.org/abs/2404.16612>Multimodal Semantic-Aware Automatic Colorization with Diffusion Priorhttps://arxiv.org/abs/2404.16678>PhyRecon: Physically Plausible Neural Scene Reconstructionhttps://phyrecon.github.io/>TELA: Text to Layer-wise 3D Clothed Human Generationhttps://arxiv.org/abs/2404.16748>Make-it-Real: Unleashing Large Multimodal Model’s Ability for Painting 3D Objects with Realistic Materials https://sunzey.github.io/Make-it-Real/04/25/2024>DVF: Advancing Robust and Accurate Fine-Grained Image Retrieval with Retrieval Guidelineshttps://arxiv.org/abs/2404.15771>GLoD: Composing Global Contexts and Local Details in Image Generationhttps://arxiv.org/abs/2404.15447>ID-Aligner: Enhancing Identity-Preserving Text-to-Image Generation with Reward Feedback Learninghttps://idaligner.github.io/>CharacterFactory: Sampling Consistent Characters with GANs for Diffusion Modelshttps://qinghew.github.io/CharacterFactory/>MotionMaster: Training-free Camera Motion Transfer For Video Generationhttps://arxiv.org/abs/2404.15789>Sketch2Human: Deep Human Generation with Disentangled Geometry and Appearance Controlhttps://arxiv.org/abs/2404.15889>PuLID: Pure and Lightning ID Customization via Contrastive Alignmenthttps://arxiv.org/abs/2404.16022>Editable Image Elements for Controllable Synthesishttps://jitengmu.github.io/Editable_Image_Elements/04/24/2024>ID-Animator: Zero-Shot Identity-Preserving Human Video Generationhttps://id-animator.github.io/
First of /hdg/ has better posters
julien is an unemployed faggot
>>100190717No pic?
>>100190796Isn't it strange that he claimed he was being paid large sums of money to develop a Lora but struggled to even showcase pony and struggled to make a Lora here?
>>100190813I forgot to copy my news pics over to my laptop so I dont have access to them atm
>>100190813I've reposted his news before just to trigger the schizos. I wonder if it's really him.
>>100190759Debatable
>>100190836its not a repost. there's new news
>>100190836He said he would be somewhat busy yesterday and today so it looks like he found a gap.
Why do I keep getting this error? I'm up to date on gpu and SD>RuntimeError: CUDA error: the launch timed out and was terminatedCUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.For debugging consider passing CUDA_LAUNCH_BLOCKING=1.Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.
>>100190856good thread timing cuz I'm just having coffee, but I'll have to be on-site most of the day otherwise
>>100190876so you have a job?
>>100190876Awww mom finally forced you to get a job
Easy off-topic filter:/^ComfyUI/i/^de/i/^PW_/i
i struggle to thinkvof any anons who stand out in my mind, i am very picky.i guess my fav poster is migselv, desu. slendy over in degen was friendly.
>>100190884Judge finally granted visitation rights.
>>100190884His mom got tired of having to blow the landlord to make rent
>>100190884SAI is looking to promote me from sdg shill to CEO but they wanted me on-site to test my 1girling before they make their decision. wish me luck
>>100190912i saw your posts on /b/ a long time ago and i think it's great you're still posting here
>>100190945>disability is no longer cutting it and mom wants me to get a real jobTranslated for you
>>100190912those arent sdg posters
>>100190827>>100189941Thanks for the help
>>100190828The reason is simple: he just can't stop lying.In his eyes, it's still better to be laughed at than to be lonely all the time.
If debo had a job 90% of his critics would have nothing left to say
>>100190961I'll continue my adventure for now.
>thread schizo is a burger flipper now
>>100190984It's really fucking pathetic, he keeps getting exposed for lying and being a delusional sperg. He thinks answering basic bitch questions can cover up his faggotry when other anons do it all the time without a avatar.
>>100190905How do I use this? Is it possible to filter news spam?
>>100190992nah. they'd just say it wasn't true or find something else to harp on endlessly. they just want to be angry at me; the reasoning flows backwards from the motivation. that's why I don't bother saying what I do, cuz it doesn't matter at all. at best, its just another thing to be attacked over>>100191025the news posts always have the same headers, so you can filter that text
>>100191012nono, he really made 6 digit big cheques yearly, for like 3 weeks total before being kicked out for fraud
>>100191045I think when he was unable to post sfw pony images everyone knew he was a fucking retard.
>>100191041no one attacks you because you are unemployed, thats julienreason people hate you is because you are a shit flinging thread schizo as proven dozens of times
>>100191070Hi ranfaggot
>>100191110>its only one anon hating me!
>>100191003and won't stop samefagging, talking about himself
>>100190961>>100190999 :) Love the spooky swamp
>>100191125Ran...
>>100191124<3
>>100190832>I forgot to copy my news pics over to my laptop so I dont have access to them atmJust use https://bing.com/create like usual
>>100191041>same headersWhere do I type and what to filter it?
>>100190951>>100190961>>100190999>>100191055very nice
The Future Sound Of London - Papua New Guinea (1991)https://www.youtube.com/watch?v=wfWMv8Y1V5E
telltale signs of babies first image?
It's crazy how Julien lubimiv lies so often
>>100191282am i supposed to know who this is?
>>100191137>>100191234Thank you
>>100191261low res and/or not fixing stuff with inpainting, though a lot of people seem to just not care lol
>>100188031>>100188031>>100188031 A NEW CHALLENGER APPROACHES
>>100191366first image ever generated on it "new iphone, no camera, no charging port, charge extra"
>>100191261Something like this.
>>100191334why would anyone care for posting another generic 1girl
>>100191366>inb4 another 4 day long d*b* spergout
>>100191398after 30 maybe 60 days.
>>100191310You know
>1girl
>>100191391inpainting superfluous details > inpainting hands>>100191417nb4 "its the tech that advanced! not your ""skill"""
>>100190912You complimented my gen one time and I cherish that memory to this very day
>>100191417Day 5, or 6 here.
an [apple|cherry] in a plate, on a tableI was expecting an apple-cherry and not an apple and a big cherry. Am I doing it wrong?
>>100191261burnt to shit and low-res, as seen herealso the 2nd to 50th gen are always porn, so an anon has a runway to go
>>100191555what do you get if you do [cherry|apple] instead
>>100191398>>100191417>>100191544>>100191489This is about day 20 or so
>>100191592
>>100191555Alternating prompts aren't a perfect substitute for prompt merging
a [ dog | fish ] on a bed>>100191625Ok. What's prompt merging?
>>100190131No filter! just the models texture: https://civitai.com/models/420049
>>100191602day 30, aka just download any XL model to get here.
>>100191528ah, I wonder what type of pic it was.i usually don't 1girl, normally just prompt as girl, or young mentally ill girl. Though I dont believe mentally ill means anything idk
>>100191417for me, its ipadapter
>>100191631>Ok. What's prompt merging?promptlet coping mechanism
>>100191750forgot image
>>100191631>What's prompt merging?It's how I call what you're trying to achieve, merging two prompts together at the conditioning level (at least as much as it can, there's like no conceptual overlap between fish and dog)ComfyUI can do it with the ConditioningAverage node, Automatic had an extension way back when, but it broke like half a year ago, shame really because it was really good
What's a decent XL model to use? Its been a long time since I used SD and I want to see if it can make decent skeletons nowI remember it struggling to not turn it into a mess of bones
>>100191782I did get something using [ fish AND dog ]
>>100191793>[ fish AND dog ]Where did you get that from?
>>100191398Default comfyUI workflow>>100191544Default comfyUI workflow different model>>100191602Default comfyUI workflow different model, controlnet discovered.>>100191685Default comfyUI workflow XL model, no controlnet.>>100191417trying different workflows and narrowing down how to make them your own. Stick with one model https://civitai.com/models/8030 working on prompting with help on LLM for new ideas. Documenting each day as saved workflow as you experiment with different ideas (not prompts)Day 45
>>100191820day 50 trying to figure out why your SD workflows don't work with SDXL
>>100191788Nice goth elf
>>100191788
>>100190717correct link for comfyui SimDA implementationhttps://github.com/chaojie/ComfyUI-SimDA
>>100191398>>100191417>>100191544>>100191602>>100191685>>100191820>>100191842Same prompt for all documented days.This is latest with Custom Sigmas
Comfy makes me what to gouge my eyes out. I can't find a work flow with face detailer. I found a one with control net and regional prompter, but I can't for the life of me figure out how to disable and reconnect the nodes. Anyone got a simple 1girl workflow? JFC
>>100191885really comfortable isn't it? so comfy
>>100191788A skeleton on a bed, turbovisionxlSuperFastXLBasedOnNew_tvxlV431Bakedvae.safetensors
Morning anons
>>100191871masterful, hyper detailed, award winning photo, masterpiece,(catwoman is stealing from a safe on the wall:1.2), standing,by beeple and greg rutkowski and alphonse mucha, d & d character, high heels,in front of a modern room background full of diverse resting (black cats:1.1) and flying (black bats:1.1), highly detailed portrait, artstation, smooth, sharp focus,>Good luck on your journey m8
>>100191808What do you mean? I got it from prompt guide 101.
>100191907jannie poast
Bing is quite good at making silly skeletons but it tends to block a lot of the images and if I want it to be a witch and a skeleton it rarely returns anything>>100191865She's an albino dark elf>>100191896This aint badHow do I make decent images with XL? My workflow of create a bunch, 2x upscale a good one then feed it through img2img to finalise it really doesn't seem to work as well with XLProbably because I can only really do 1.5-1.7 upscales due to having a 10gb 3080
>>100191885Haven't used it in ages but DZ-FaceDetailer seemed straightforward enouigh
a prompt i saved in august 2022 during the sd discord beta days.!dream a painted statue made of marble of Victoria Justice as the Goddess of tentacles clothed in silk looking at you intensely with seductively serious eyes while being embraced by slimy tentacles from the Squid King. ultra detailed painting at 16K resolution and epic visuals. epically surreally beautiful image. amazing effect, image looks crazily crisp as far as it's visual fidelity goes, absolutely outstanding. vivid clarity. ultra. iridescent. mind-breaking. mega-beautiful pencil shadowing. beautiful face. Ultra High Definition.-n 5 -i -s 140 -C 10.0 -H 700
>>100191922The docs on the Automatic github say nothing about using AND with brackets
>>100191973
Menstruation
>>100191973>ultra detailed painting at 16K resolution and epic visuals. epically surreally beautiful image. amazing effect, image looks crazily crisp as far as it's visual fidelity goes, absolutely outstanding. vivid clarity. ultra. iridescent. mind-breaking. mega-beautiful pencil shadowing. beautiful face. Ultra High Definition.some anons still proompt like this
>>100191913>masterful, hyper detailed, award winning photo, masterpiece,highly detailed portrait, artstation, total meme crap that just fucks up your picture
>>100192006I guess 1.0 days stick with you to the end of times.
>>1001918851) Model, conditioning, vae etc is all self explanatory2) Use UltralyticsDetectorProvider to load your yolo adetailer models. -If the yolo model is just bbox, connect bbox_detector output to the bbox_detector input on facedetailer-If the yolo model is segmented, the also connect segm_detector output to the segm_detector input on facedetailer. If you do this, only the segmented face will be denoised and not the whole bbox.3) Connect CoreMLDetailerHookProvider onto detailer_hook, which lets you determine what size to upscale to.In theory, you can use segmentanything to do faces rather than the yolo models, but there is no real benefit.
>>100191987The brackets are there because I just didn't remove them. Same result with or without brackets.
Poney AND cat
>>100191878Albedo is good if you enjoy mediocrity
>>100192017I just paste what AI Chatbot LLM gives me.Or from cai c/pI do need to work on better bot though
>>100192099Why exactly do you think your AI LLM chatbot knows how to prompt?
>>100192131it doesn't, which is why I need a new one.
>>100192074
>>100191885>>100192031https://litter.catbox.moe/gu2lkb.pngActually, here. You can see I expended zero effort on making this look good with fiddled settings or even writing a prompt, this is just to show you how to connect nodes. You can delete the efficiency nodes and connect the facedetailer to another workflow (once you fiddle with settings) , but it'd be better to download them: they make comfyui a LOT simpler to use.If it doesn't work, make sure you have at least one face segmented yolo/adetailer model downloaded and load as I said earlier.
>>100192165nice specimen
gif of increasing koff and psykoff loras from weight 0 to 1.
>>100192209Looks like shit
>>100192209>psykoff? i have the regular koff one, is there a second?
>>100192099trusting a bot, rookie mistake
>>100192153suppies?
>>100192227ah, darn, i thought it was neat.>>100192241https://civitai.com/models/314747/psykoff
>>100192209very dreamlike
>>100192251Or dating a robothttps://youtu.be/4uE96qUlJ_4
>>100192278nice ty i cant wait to try it out. itll be the only reason i use 1.5 again since many months.
Comparing right now how different models do feet and somehow this ancient anime merge manages to destroy all competition again.
>>100192271childhood memories, waiting for mom in the car while she ran into Suppies
>>100192209nice
the overseas vendor was kinda shady, but my delivery finally arrived
>>100192408I think I misread the listing
>>100192440the top is realistic while the rest is anime-like, did you do it on purpose? or just a gen accident
>>100192556yeah, it's intentional
>>100192556He's been genning with consistence, it's clearly intentional.
>>100192651nice texture
I thought people said it would be impossible to make porn using XL models, what happened? Am I misremembering it?
>>100192718>I misremembering it?No. Now, they are pretending like they never said it. Strange times.
>>1001927181-5 schizos on /sdg/ isn't "people"
>>100192718Impossible no, just highly inconvenient for a long time. If the people want to coom, they WILL coom, the question is how many obstacles you want to put in their way and how much you're willing to compromise the rest of the user experience in service of that goal.
>>100192718I mean nude loras were appearing in a matter of days after it released. Hardcore loras and checkpoints not long after that. Hell even base SDXL could do above the waist nudity just fine which could be considered porn enough for some people.
>>100192737>1-5 schizosit's just one, and he just keeps samefagging as more than one or people paying attention to him
it took two gens for me to realize how good base xl is, back in the 0.9 days of course
>>100192309pony unfortunately mogs all other models for hands/feet. It's sad that instead of being happy with the revolutionaryâ„¢ models like JuggernautX I'm waiting for some shitter to make a good realistic pony merge. So many corporations getting mogged by an autistic pony fucker it's insane.
>>100192673it's an accidental kristjan kullamaggi tribute!
https://strawpoll.com/wAg3ArVAWy8Strawpoll still up btw
Some cool new Hyper-SD workflow stuff just dropped.
>>100192808and?
Prompt:>angry elf bitch
>>100192808funny how the d*b* one was a lot clearer, guess he really is the problem
Can I ask the thread for some feedback?Does anyone use the Sea Art XL model? I'm trying to train a boring negative LORA for XL models, starting with Sea Art XL and if it works I'll do AutismMix XL next. This LORA is trained on images from boorus with 0 favs, negative likes, or tags like "low quality." So putting it in the negatives should improve the quality of images.I trained 20 Epochs and I'm trying to find the best one to upload to CivitAI. Pic related is an example comparison, and there are a bunch more at https://files.catbox.moe/xayijv.pngWhich one of these looks the best to you? Does it seem like enough of an improvement to use, and would you want to see it for other XL models?
>>100192808where is the option for 'who the fuck cares stop being so goddamn invested in a 4chan thread'?
>>100192883you don't understandsome anons spend nearly every waking hour ITT
>>100192902again, who the fuck cares?
why get into a parasocial relationship with a 4chan thread when could get into a parasocial relationship with an equation running on your graphics card
>>100192924mmmmm. sexy matrix multiplications
>>100192911>again, who the fuck cares?the ones who spend a majority of the day ITT
>>100192169Thx for the suggestions. Still don't get a lot, but getting there.
>>100191885Why waste your time with an inferior product to begin with? Comfy is flawed in its design. Setting up a factory in factorio only works because what is being produced is uniform.SD images are not uniform and what needs to be done to fix an image is different each and every time.
I have meetings so can only regen these while I explain why there is no one to look after tickets.
is there anyway to connect an upscaler model with facedetailer? also CoreMLDetailerHookProvider only has low resolutions, what file should i edit to add more resolutions?
>>100193186>>100193263nice
https://youtu.be/cfKBgccCric?si=kvOjSqvU26rmBf3j
>>100193186really digging the effect. nice work!
>>100193407Don't ever reply to me nigger
>>100193407fuck off julien
>>100193508how do you get schizo so riled doing absolutely nothing? it's superhuman
>>100193527>cope phraseStop crying
>>100193527>how do you get schizo so riledhe does it to himself. I hope he gets help. poor guy
>>100193586>it's one personJulien please.... Oh right lying is what you do every day
all me btw, sorry
if i could train a lora id use images from those "BUILT 4 DILFS ZOOMERS" threads on /b/
>>100193407Thank you Fiz!
>Cumfy no longer shills SD3>no longer sperging out on public comsAmazing what actual leadership can do
playing around with a new merge
>>100190707Meta's AI generator it's actually surprisingly underrated. Give me prompts to test it out with please
>>100194004
Any Azula?
>>100194088>>100193944>>100193923>>100193907>>100193843>>100193699>>100193186<3 Lovely
>>100194004a handsome muscular man lounging on the beach :3
anyone have a workflow for tiled diffusion/MultiDiffusion for comfy? Been looking through the menus for like 10 minutes I wish there was a search function.
me irl
>>100194213You're wasting your time with comfy my friend. If you can't see through the matrix it's not for you.
>>100192869I'm not sure it is actually working, but I suppose 05 is best in this. Is there more indication that it actually did learn?Also test with some boys and girls present.
No ADetailer
>>100194124Only a very basic one. Maybe you should do some better variants?
>>100194381ADetailer prompt: "(narrowed pale pink eyes:1.4),( black wingtip eyeliner:1.3), blushing,( sculpted black make-up eyebrows:1.1),"
>>100194213double left click on the graph
>>100192810looks nice
>>100194213try this from the exampleshttps://files.catbox.moe/c0xlfk.png>>100192869shouldn't you be testing in the negs?
>>100194385Very nice, anon on /b/ did some lewd Azula and X-23 figures, I'm adding this to that collection
>>100192718I think on sdg many people actually thought it'd be quite easily possible.
bros seriously i have a problem, some of my png outputs get corrupted. it worked yesterday and now i cant even open it
>>100194486I think you could quite easily do some lewd Azula figures if you can run the pony/sdxl figure models.
>>100194530That's nothing to go on. Have a look at the pngtext exif kind of data of these images and maybe some older ones, see if you can't spot the problem?
>>100194663cool can you add like zombies to it or somthing?
>>100194539I'm done with school in 2 weeks, maybe it's time to dive in
>>100194589i just get file format error, cant check the metadata or sanythingholy shit this is bad bros, this whole folder will not open
>>100192808this only neatly shows there is like 30 people here
>>100194708Permissions?
>>100194736i have permissions. i am going to run chkdsk
>>100194756Otherwise just delete the folder if you can
Good evening bros. How do I get the alleged 60% speedup by using WebUI Forge with SDXL? Normally I use ComfyUI and it takes me about 2 minutes to go through the 20 steps on a 1024x1024 image, with a bit extra at the beginning to start the gen and at the end to spit out the image. Maybe 2.5 mins in total?I tried Forge and it still takes 2 minutes to actually do the steps. The delay at the start and end of gens is way smaller so it's actually closer to 2 minutes and maybe 10 seconds. That's still a time save, but nowhere near as dramatic as the advertised 60-75%.Specs:>i5 10400f (not that it should matter?)>16GB DDR4 2400>6GB GTX 1660 Super>models and web ui stored on my spare 240gb sata ssdDid I miss any settings? Adding arguments actually makes it crash or take ~15s per step instead of the usual ~6s
attempt at azula... don't really care for the character, don't like avatar's style anywho.
>>100194800Cool
>>100194699this is great, can i see the workflow? what loras, etc
>>100194808velvet tenties? interesting idea..sorry I'll try to avoid more of my horrible style for today
>>100194819It was a mystery anon on /b/ who did it unprompted, you could ask around there, also there is figure gens here, you could ask these anons
M hair
>>100194699>maybe it's time to dive init is! you do take some breaks, yes? you can get an azula already trained in the pony pvc model never mind elsewhere as lora, no need for a lengthy training effort right now>>100194708create a disk dump on another storage and run the windows filesystem checkif you have no other storage it is probably a good time to get some for your backup now
>>100190707Did I do something wrong with how I train my LoRA?While the base model managed to learn the style well, the innate knowledge the base model has regarding the prompts like "train" or "a steampunk city" which my dataset lacks, is lost when the LoRA is applied.The LoRA's training data:https://pastebin.com/W2UdqWVeLeft image:https://files.catbox.moe/59bv3x.pngRight image:https://files.catbox.moe/jf2onb.png
>>100194921I'd like to believe that little square on the left is a box of jack-off tissues
>>100194788I found the timesave. The timesave is that I can now generate batches of 2 without running out of VRAM. Meaning I can make 2 images in ~3.5 minutes instead of in ~4.2 minutes. I shall now try a batch of 4 and see if my VRAM can still handle it.
>>100194846The style was meant for the lingerie but bled over to the tentacles.
>>100194819>>100194921pony can gen figures natively but there's also figure loras for pony on civit.
>>100194957It's been school, hockey, pc gaming, and gym for the last 2 weeks
>>100194965Lol, most likely
>>100192924I dunno, does the 4chan thread or the LLM have greater context memory at this point?
>>100195023Nice!
>>100195067Good eyes :)
>>100195023Cute feet
How do I make my gens less blurry?>DPM++ 2M>20 steps>cfg 7
>>100195126do you start at a lower resolution and then upscale, or use a detailer? that will help if you don't already
>>100195171I start at 768x1280 and then Highres-Fix with 4x-UltraSharp 0.3 denoiser to 1.5x
>>100195171
>>100195184try highres fix with latent at .6
>>100195224>latent at .60.6 denoise? Won't that gen double head and stuff? It alters a lot
>>100195246sis.... have you tried it?
>>100195184look up multidiffusion (and then use mixture of diffusors)
New Thread>>100195136>>100195136>>100195136
>>100195263It became even worse
>>100195328>>100195202I'm a comfyfag but set your hires steps to 20, schedule is fine I guess but set it to karras, denoise with 0.25 to 0.65 to see what it doesyour hires fixing, but 0 times, so no hires, if I'm reading it right
>>1001953490 hires steps means it does the same amount of hires steps as normal steps.
>>100195349ok thx I try
>>100195264mogged
>>100195359ah I seewhelp do what >>100195264says I can't into A1111 anymore
>>100195328did a latent upscale (at 1x? lol?) instead of multidiff with this one. should kill the thread too.
>>100195375No what you said here >>100195224also works (and kind of blew my mind). Going to experiment with it later.
>>100195376well that looks a lot better doesn't it? maybe a bit burned maybe set the cfg to 3-5 see what that does?
>>100195392wasn't 'me' but the other anonI'll send your regards to him
about time i actually tried using some of the features of these nodes ive downloaded, fast groups muter is based