Discussion of free and open source text-to-image modelsPrevious /ldg/ bread : >>101585073>Beginner UIEasyDiffusion: https://easydiffusion.github.ioFooocus: https://github.com/lllyasviel/fooocusMetastable: https://metastable.studio>Advanced UIAutomatic1111: https://github.com/automatic1111/stable-diffusion-webuiComfyUI: https://github.com/comfyanonymous/ComfyUIInvokeAI: https://github.com/invoke-ai/InvokeAISD.Next: https://github.com/vladmandic/automaticSwarmUI: https://github.com/mcmonkeyprojects/SwarmUI>Use a VAE if your images look washed outhttps://rentry.org/sdvae>Model Rankinghttps://imgsys.org/rankings>Models, LoRAs & traininghttps://civitai.comhttps://huggingface.cohttps://aitracker.arthttps://www.modelscope.cn/homehttps://github.com/Nerogar/OneTrainerhttps://github.com/derrian-distro/LoRA_Easy_Training_Scripts>Pixart Sigma & Hunyuan DIThttps://huggingface.co/spaces/PixArt-alpha/PixArt-Sigmahttps://huggingface.co/spaces/Tencent-Hunyuan/HunyuanDiThttps://huggingface.co/comfyanonymous/hunyuan_dit_comfyuiNodes: https://github.com/city96/ComfyUI_ExtraModels>Kolorshttps://gokaygokay-kolors.hf.spaceNodes: https://github.com/kijai/ComfyUI-KwaiKolorsWrapper>AuraFlowhttps://fal.ai/models/fal-ai/aura-flowhttps://huggingface.co/fal/AuraFlows>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>Try online without registrationtxt2img: https://www.mage.spaceimg2img: https://huggingface.co/spaces/huggingface/diffuse-the-restsd3: https://huggingface.co/spaces/stabilityai/stable-diffusion-3-medium>Related boards>>>/h/hdg>>>/e/edg>>>/d/ddg>>>/b/degen>>>/vt/vtai>>>/aco/sdg>>>/trash/sdg
messy collage edition
>>101601691It's my favorite so far
Blessed be to the anons in this bred
>>101601691A reflection of the tests and trials in the last thread and threads past
Later
>Installing llama_cpp...Why does it have to install every time?
>>101602670Probably because of important fixes. iirc llama 3.1 had issues with going over 8k context on llama.cpp recently
Good night anon
>>101601833Nice.
bum
Dead joke of a general that is literally being kept on life support by one spammer.
>>101604323gm
local image gen is a joke, no developments in almost a year. not as bad as voice synth thoughever
i love ldg
No gens?
>>101605103>>101604323
shhh anons sleeping
>>101604323>>101604445Shame what happened with AuraFlow v0.2 innit
>one
>>101605357Cool
official pixart bigma and lumina 2 waiting room
big
ma
mina
2
>>101604323I can sense your fear
make sure to thank that anon for keeping this general warm and toasty for us till the release of bigma
Prompt -> LLMBase prompt : A wall full of boobs, groupIt takes it literally lol, SD models would just gen anime girls
>>101604323Sad but true.
>>101603455ty>>101604323That one "spammer" is why you even know Sigma exists. When a few months ago people wondered what was new with Sigma in sdg, I was new with Sigma. When there were no models on civit for Sigma, I put up bunline. No gens in this thread? I put gens. All from the USA. The thread is "split" because of you hostile fucks and all you can do is be salty that something outside of your gates keeps itself. Keep bending over for stability.
Behold: a growing collection of Sigma models from independent authors. I'm proud of you guys
you will all be using sd3.1 soon enough
>>101607184Keep waiting for disappointment. Local anons take matters into their own hands. This is long dick general
i wish i could marry my pixart
>>101607207It's literally one spammer general. It's not even a troll at this point, this general should be removed at this point. Accept defeat and move on.
>>101607255Close the door behind you. Thanks!
>>101607255they like children in the other general and allow saas gens for whatever reason
>abstract representation of how stability sees its users
2k model still cooking
Update on ELO rankings for diffusion models
>>101607334i don't get it :(
>>101607423It's an asshole knit in yarn
>>101607435i thought i was just a dirty minded pervert for thinking that lol
>>101607393what's the vram requirements for the 2k model? does it take way longer to gen compared to the 1k model?
>>101607450It depends on your setup for VRAM, but it takes about 4x longer to gen, outputs above the 4mb 4chan limit (annoying), and VAE decoding takes a lot more VRAM. 12GB+ is recommended.
>>101607450The 2k model is somehow the same size as the 1k model because interpolation magic so it doesn't use much more VRAM during generation
>>101607271this one is cool, style prompt?
>>101607507A patchwork hot air balloon drifts through a forest of trees.
https://github.com/intel/AI-Playgroundintel won
>>101607575'>This example is based on the xpu implementation of Intel Arc A-Series dGPU and Ultra iGPU>AI Playground supports PyTorch LLM, SD1.5, and SDXL models.Hmm.. won what exactly?
>>101607575wher pixart support
>>101607651its open source add it yourself
>>101607575>>101607651>>101607663it's a demo for one line of intel processors or arc GPU. Only the most niche of anons could make use of it
bigma update?
anon I have to get stuff done today. Thank you for keeping the bread warm while I slept. Cya in a few hours
>>101607753see u, have a good rest of your day
>>101602973deep
HunyuanDiT with ComfyUI native is working very nicely now. Would be great if we could get safetensors conversion script so we can convert our LoRAs. Also TensorRT support would be godsent.>>101606600In terms of model quality I noticed 1.0 ~= 1.1 -> 1.2 specially for anatomy etc... they went hard with censoring 1.2, 1.0 is least censored but 1.1 isn't that bad. I'm guessing they all can be uncucked easily, just gotta wait for that 1.0 community released code hopefully. The good news is that regardless of levels of censorship this only applies to base models, the pretrains still appear to be fine so finetuning it or just a simple LoRA causes drastic changes.
>they went hard with censoring 1.2Good to know. Does it do anything at all better than 1.1?
>>101608138>>101608245If not, then it's not worth fine tuning 1.2 over 1.1
>>101607410Meh
i shleep till bigma
>>101602973>>101608034When everything seems black and white and yet you still feel lost
>>101607410This is proof that the only thing wrong with cascade is the license. Imagine ultrapixel with Pony-level finetunes
>>1016072712k validation testing latest epoch on same prompt. I think cfg is too high on this one>>101607763ty, you too anon>>101608034thanks, got any gens?>>101608138>HunyuanDiT with ComfyUI native is working very nicely nowGood to hear! Just pulled latest comfy and will try it soon>pretrains still appear to be fine so finetuning it or just a simple LoRA causes drastic changesWhat would you recommend to finetune it with?
That's all for now gtg
imagine still using SD when pixart mogs
>>101607255That's crazy, all the gens in OP are made by a single anon?
>>101608956Show me some good cascade gens (hint: you cant)
How many beers did she have
>>101609804is this a saas model?
>>101609940Nah I used Juggernaut XL
>>101609940Were you ready to shit your panties?
>>101609969nta but kekd
I'd like one too.Maybe later.
>>101609731Did you use a hand detailer?
>>101609969yes in 4k resolution
>>101610008Nope. I've found the good values,and the good positive /negative keywords, i guess. I'm currently experimenting with really low cfg values, using the oil painting LoRA for pony
>>101609940I am retarded >>101609959 and didn't understand what saas meant, yeah I used stable horde for this one, I can delete it if it goes against the spirit of the thread
>>101610100>stable hordeno it's fine, stable horde just uses local models so i won't be soiling my panties anytime soon.
>>101610128yay
>>101604445>no developments in almost a yearEven in the purported non-local thread, almost nothing isn't local. What are you talking about?
>>101610222anon is seething and coping simpleas
>>101610008And i'm also cherrypicking a lot
>>101604445Sigma came out in the last year and it's by far the best local AI architecture. And let's be clear, you don't have a computer that can run a next-gen model you're pining for.
>>101610275Potential is heckin unlimited!!!! Good finetune will come out any minute...*keeps spamming your own thread solo for 5 more years*
>>101610293whyz u downing for? pixart being good will benefit you
>>101610040>>101610304style is pretty ugly desu
>>101610293I can't imagine the type of loser that bitches because other people aren't bending over backwards to serve them. You're the worst kind of beggar and leech.
>>101610319Yeah it will but there's so many promises in this field it's not even funny. It's just shills for the most part.
>>101610319He can't gen so he doesn't care
>>101610341well the thing is sigma has yet to have any finetunes in the same level and scale as the sd models did. you gotta remember that any new model that comes out is not only competing with the base sd 1.5 and sdxl models, but the months of community development and finetuning they had as well. it's pretty impressive how well pixart sigma does despite being under trained and smaller than 1.5, but you gotta give it time (lots). it's also harder to make datasets with natural language instead of tags because of how sucky current vlms are. i think if we had a good vlm we'd see alot more models and finetunes prop up.
experimental comfy interface update adds node selector and previews to the sidebar
>>101610426You do realize most SDXL finetunes are utter ass right? Turns out 90% of the work being done by three people and they're all SAI shills. There's nothing stopping YOU from finetuning.
>>101610471>Turns out 90% of the work being done by three people and they're all SAI shills.those people have aloooota money and that's what really matters in the ai space. my point is if they used the same amount of cash and resources on a pixart model, we'd have a pretty good model.>You do realize most SDXL finetunes are utter ass right?yea. most finetuners in general are sdxl finetuners.
>>101610532Turns out it's all astroturf and leeches like you. I'll say it again, turns out all the work is being done by like three people and the rest are just fags begging and fags spamming 1girls who are probably still cooming to Pony.
>>101610456All we need now is to be able to pin node values to it
>>101610571really nice
>>101610567>turns out all the work is being done by like three peoplei don't understand, i never disagreed with this.
>>101610622Stop bitching about no progress being made like a faggot. No progress is being made because the leeches like you actively work against anyone who tries to do anything. You want something done, do something. But you don't because you're a useless leech. Just. Shut. Up.
>>101610642what? i'm not that anon, i'm arguing against him, im not the doomer.
>>101610653You're still in the "somebody please do something" camp.
>>101610659If i had the resources and knowledge to train and make a model i would. But i don't.
can someone recommend a good ponyxl checkpoint for making SFW cartoon/semi realistic memes? i normally use the realistic ones but i've trained some male friends i know and want to make funny pictures of them. the base ponyxl 6 checkpoint works but i suspect there are better options.
>>101610746i'm not familiar with pony models at all but i hear AutismMix be mentioned alot. as for semi realism, i have no clue, sorry.
>>101610823>AutismMixnow that you mention it i've heard about it. thank you ill check it out
>magical lithograph
>>101610832you could try asking in the >>>/h/ and >>>/d/ ai generals, they should be very familiar with pony.
>>101610853i'd rather not deal with the weebs lmao
>>101610679what model is this? it looks very nice.
>>101610867https://civitai.com/models/435669/bunline-2k1024512-pixart-sigma?modelVersionId=665122
What model, LORAs, etc. do you guys recommend for fantasy artworks? Making some character portraits for a TTRPG campaign. 12 GB VRAM.
>>101610946https://civitai.com/models/264290?modelVersionId=298243this is my go-to
>>101609522Only 4. Top left, bottom left, bottom right, and far right
>>101609100CFG 2
Rainy days
>>101611313
>>101611319nice android
>>101611366thanks
>>101610456Thank god, I remember when SAI said they would put "significant resources" into improving the UI, what a joke.
>>101610456This is the dog bollocks, I guess its not stable yet, but this will improve quality of life a ton.
>>101611366cool fantasy art
>>101612339So AI
>>101611672Reminds me of Crysis
>>101612339Which model?
>>101612339strange image, kinda neat
>>101613409vector design of a chicken?
>>101613470The model took a hold on that one. From the same batch
>>1016134583girl
>>101612535ty, it's from testing an unreleased model someone else made. Hope they release it soon!
4girl isn't playing around
what kind of specs do you need to make a lora for pixart? how come i haven't seen any?
>>101613680Ironically about the same as full finetuning. You need lots of VRAM to extract T5 embeddings and VAE embeddings. Training the actual model is very lightweight.
>>101613709i only gots a 3080 10gb and i've never trained a model before, do i need to do that T5 extraction thingy?
>>101613750OneTrainer might be able to do 10GB, but the lowest I know of is 16GB
>>101613793i see, ty anon.
>>101613750Oh and the T5+VAE embeddings are 100% necessary. I would get T5/VAE embeds first before training for reduced peak VRAM.
>>101613830yw, good luck!
>>101612828Cut me some slack anon I've just got into using detailers using comfyui-impactpack and sam models. My old CPU was not able to handle the avx instructions, that image I posted is probably the best I've done so far with a person at range. I'm struggling to figure out how to do each seg without the images changing. As in first I want the person detailed, then I want the face detailed and finally the hands without messing the previous image up. Anyway I'm fucking about with animatediff and used its detailer to see how that works.I was just to make a post about how hard it is to keep up with everything, there are a fuck ton of nodes I still don't fully understand, but in time I'll get there.
>>101613193cyberrealisticpony_v40I think they just released a new versionhttps://civitai.com/models/443821/cyberrealistic-ponyAnyway if its the same person, this is a very good model imo. civitai is being a cunt right now, seems the search is down. >>101613418I used pic related i created in gimp to gen the background, this when when through many stages of copy to clipspace then mask entire thing and denoise again with things add to prompt.>>101613418Crude Pic related, the gist of what I did with a few things missing, (i'm not good at this). I recommend only using a hard clean stroke brush in gimp/paint/photoshop for creating the background concept, the AI seems to get confused with fuzzy blurry paint strokes. I croped the woman using a background removal node https://github.com/john-mnz/ComfyUI-Inspyrenet-Rembgits fucking amazing, just route the image from a load image node into it, disable all other nodes temporally and have the output on a preview image node, then just right click and copy image and paste as new image into gimp for example, scale the image in gimp down to what you need and paste into the background also copy pasted as a layer from comfyui and away you go. Failing that, you can just draw a crude outline of a woman using gimp to where you'd like here to be and just mask that area and prompt, then use detailer for person, face and hands.
>>101612828and I know what is wrong with it, the image sharpener was only meant to effect her face, but it seems to always effect the entire image, so when you zoom in you can see how grainy it has now become... pic related, was before processing, still a bit bleh crap.
>>101614514>https://civitai.com/models/443821/cyberrealistic-pony>Anyway if its the same person, this is a very good model imo. civitai is being a cunt right now, seems the search is down.it is the same person, just reading that comment about complain about lack of gay content, they will be making a new version for just that, not that I care because I'm not gay. But this new version appears to be much better, v40 I was using was already amazing, they are updating this thing like crazy because it was only last month I downloaded version 4, there have been 2 versions since then. Get honestly you wont regret, he detail of female gens from this thing at close range is for intents and purposes real... It does black African woman real well, which I like because so few get it right. I'm like me some ebony from time to time.
>>101614203>>101614617you got this, anon
>>101614514Neat progress imgs. I love seeing these.
>>101615043I prefer photo real images, so I'll always work towards removing the evidence its AI generated, like I can totally tell when an image is AI generated now. You only have to zoom in, our brains will accept backgrounds that aren't up to scratch, but we easily see when hands are not right, faces are borked etc. Unfortunately its really kind of hard to get something to be like 100% legit, people aren't sharing and that's normal, because if you could gen the real article would you tell people how you did it?Anyway I'll get there :-) give me 2 to 3 months and I'll be there, its a just a few minor set backs, its probably with image sharpener not doing what i want, its fucking the main image up, it should only be doing the masked features. The main image needs to be softer, more like a png or jpg from an actual photo.
>>101615086>>101615043Do check this one out anonshttps://github.com/john-mnz/ComfyUI-Inspyrenet-RembgIts helped me greatly over the past few days, note it does not work with animatediff, its like animatediff ignores it. The strangest thing was I actually tried to just feed in a folder with the saved images with background stripped and it would gen the same background... So I don't know why that happens, I'm thinking that some how the noise data is stored within the alpha channel, so I'm looking now for a node the removed alpha channel to see if that is the cause. I want that because I'd want to make cool tiktok girly dance videos and such. Anyway I'll figure it out, but it is very strange, I worked all night trying to figure out what is happening, i've not yet removed the alpha channel, i might have to make a script for this, because am I fucking doing that shit manually for many frames...
what helped me the most was learning that you can select nodes in comfyUI, then right click somewhere and save as template. Now no more need to drag in a work flow or image, i can build from templates of machines... So learn to CTRL click nodes till you have them and only them selected then move mouse to empty space and right click, then select save as template. This template gets save somewhere within the Comy directory, then right click empty save and load from template. Learn to seperate nodes into logical chucks that actually take advantage of reusing loaded models to save vram, make use of reroutes and color them according to the color of type of connection. Change connections from spline too straight, place connector reroutes at the side of your machines, makes things so damn easy to understand. I know some people use a main bus, I never liked this concept personally, I feel that its much better just to have reroutes ready to receive inputs, I might also include some common outputs. Like latents or images or models, so they are ready to connect into the next machine. Don't get bogged down by the UI, it is intimidating at first but if you do I've said things really start to make sense.
>>101615388to save on vram i build provider groups, these groups feed into the various other gen groups, why create another vae loader? Just reroute it and provide it in more than one place. Sorry if this is teaching some to suck eggs but a lot of out of memory problems can be avoided by simple doing this basic thing. I see way to many shitty workflows and every twat on youtube with the AI voice over copies it... I also hate overly complex workflows, I do not think this is normal or beneficial, learn the latent space and how the AI actually works first. You can remove half the bloat from those shitty workflows that get shared.
>>101608245They haven't really released any info on it despite the questions so we can only assume the only thing it does is enhance their censorship.>>101609100I have used their native python script so far, but Kohya should work fine. I am no expert have just run some tests (but have been prompting the models for a while now). You can check some community made tunes/LoRAs here https://www.shakker.ai/ (shake the world hunyuan section).
>>101615388>what helped me the most was learning that you can select nodes in comfyUI, then right click somewhere and save as template. Now no more need to drag in a work flow or image, i can build from templates of machines...TIL, ty
And there are those that use like 4 ksamplers... What the fuck are you even doing you retards? I've tested, and I found its down to skill, its don't matter any more than that, all you are doing is desnoise and then denoise again till you're fighting against what the AI will inevitably do... I remember in my music production days, you over work that sound you will ruin it. The same is true for this shit, stop doing that its annoying, get it fucking right the first time.
>>101615469this does not include very low denoise passes to fix certain things though, here that is still useful.
>>101615388>then right click somewhere and save as template.I've been looking for this, thanks. >>101615430>I see way to many shitty workflows and every twat on youtube with the AI voice over copies it...I've only learned one valuable imggen thing through youtube and that was for a node which had little documentation. Everything else is trash. >>101615444>https://www.shakker.ai/Should go in OP >>101601667 along with fixing the links for Aura. Thank you anon.
>>101615444>https://www.shakker.ai/didn't know about this site before, thank you for sharing. hunyuan has alot more models than i expected, really interesting.
>>101615537oh wait, can you not download the models? says unable to download, do i need to make an account or something?
>>101615457ok, now another thing, you can create group, then select the nodes you want to be a part of it. Then right click the group title or group shaded area and select 'fit group to nodes' and it will perfect surround them and you can move that group of nodes around by clicking in the group title. You can also select nodes to group but that is sketchy and not have the desired result, probably I've not tested enough myself to know how it will work. its a shame we cannot save groups or clone groups etc...
>>101615444I'll go native first, ty for the shakker resource!
>>101615457you can also right click the group and select all nodes in the group, then CTRL + B and bypass them all or un-bypass them using the same keyboard shortcut.
>>101615514>I've only learned one valuable imggen thing through youtube and that was for a node which had little documentation. Everything else is trash.most of it is indian so they use the AI to gen the voice over and script because... "hello, today I show you how to make perfect girl dance video" And you'd close the tab...
>>101615550This is for a contest hosted on the site, with some exclusivity clause, some of them are unable to download. The anime finetune https://www.shakker.ai/modelinfo/87e2cc2169934523a2ff82fb12e7206b?from=feed is set to released on civit and there at some point this month according to authors, they wrote a paper and everything so it is likely to be true.
Already found a cool lora, I'll keep tabs on this place. https://www.shakker.ai/modelinfo/b0bf4da825454d479d49252dfd423139?from=feed>>101615619>with some exclusivity clause, some of them are unable to downloadThat's unfortunate, I hope anon can still find cool ones to download.
>>101615514There are some good tuts though, i always just pick around what they show and find out I don't need those bazillion ksamplers or that massive complex IPadapter and control mesh of shit. I don't know man I should remain humble, but most of these workflows break often after 1 month because they use so much stuff, its best to stick to well know and used nodes mostly. If you load in one of their workflows it will want you to download like 6 new custom nodes and you have no idea what is inside of those things. With the recent hack done it be best to actually learn how the main nodes actually work and how to apply them.
>>101615619the preview image with the surprise middle fingers was mean but looks really good, i'll keep an eye out for it. any more interesting sites like this?
>>101608245what are you guys talking about, I'm curious.>>101615537>hunyuanhave the models at least saved somewhere but never really used it because when I tried I realized I would a new python environment just for it as per instructions, they i only had a 2 GB card... So I didn't bother. Is this model any decent? I heard its best to translate to Chinese?
>>101615544nice
>>101615643I think I will start to share templates here once I'm happy they work, I know they can be exported and imported. Why bother with workflows?
>>101615670haven't tried it myself either since i only have a 10gb card. native support for it got merged into comfy ui recently so i don't think you'd need a new environment for it. we are currently talking about this interesting looking tune of it, hope it releases.>https://www.shakker.ai/modelinfo/87e2cc2169934523a2ff82fb12e7206b?from=feed>I heard its best to translate to Chinese?yeah i heard so too. using chinese prompts for kolors made a huge difference, not sure if it's as drastic for hunyuan as well.
>>101615692and then people would see what I actually mean, they would be intuitive with a small notes node to explain its inputs and outputs. And most importantly what it actually is meant to do.
>>101615670>what are you guys talking about, I'm curious.they are talking about the new version of hunyuan
>>101615640
>>101615723Yeah, that's nice, I'm borderline to learning Chinese. Its a really hard language, nothing compares to it, I've been there in the past its super hard to speak it. Japanese is easy, Chinese is like Polish as in its not phonetic, so English speakers are mostly filtered, it easy to learn Japanese because it sound s like it is written. As if if you know Hiragana you pretty much got it as it sounds.
>>101615619Also if you want to see what you can do with 1.0 and 1.1 check https://imgur.com/a/hunyuandit-0vrZEn0And also the archive here for filename Hunyuan and just Hunyuan in general where I have discussed it in detail. (For the realistic 1girls it's mostly no longer possible with base 1.2 unless you tune it, and anime prompts the girls show less skin etc). Main advantage of Hunyuan over other models is better multi subject adherence, scene coherence, much better consistency with hands, feet, etc... I have touched the surface of what you can do in that imgur btw. There are manga h-artists this thing knows without any tuning. Overall the model is very close to NAI, Niji/MJ out of the box. The weakness is its English prompt understanding, which requires some optimizations, but with tunes this can be improved so it's getting there. I am sure even the researchers are scared to an extent to release a really good tune since it would be perfect at nudity/porn which could lead to more censorship down the line.
>>101615778incidental Russian is also easy because its phonetic, you got the Cyrillic script down you got it as it sounds mostly, some minor things but if you know what each Cyrillic letter sounds like you got it.
>>101615782>https://imgur.com/a/hunyuandit-0vrZEn0the schoolgirl sticking her tongue out doing a peace sign is one of my favorite hunyuan gens, other than the duotone ones
>>101615682ty
it is good to see you again, hunyuan shill
>>101615805well it run on a 12 gig card? or will i need more system ram or another 12 gig card running parallel?
>>101615830china will save us, china number 1!unironically they will. I just don't like how they do software, they are using older versions of things...
>>101615830I miss the chinese girl selfies desu
>>101615848i miss them too, hunyuan shill you should post some every now and then
>>101615855>hunyuan shill you should post some every now and thenyes we want to see what this model can do. I will probably experiment tomorrow with fresh mid, I have a new drive so no so limited now, 40+ gigs is what the hunyuan takes...
does the skin still have that weird texture?
>>101615854nice
>>101615899make him take a doodoo
>>101615909pervert
>>101615917i-i'm just curious to see if the model could do it
40+ gigs yep, i saved them I could not get it working because it required separate install of the base software in different venv python because it requires different versions and shit. But be expect ti have their script try to download more through more connections than you router or isp can handle, many failures. I didn't think it was worth it plus i got a nasty email from my provider but probably due to extensive downloads last mouth over 1 TB
>>101615887
>>101615924What! Controlnet?
>>101615923doesn't comfyui have native support for hunyuan? i thought you don't need any additional dependencies anymore
>>101615830>>101615855Been super busy which is why I haven't been sharing as much, I did share the prompts and they can be replicated in new comfyui (at least with ddpm sampler, normal scheduler at 40-70 steps, haven't tested other settings)>>101615835It will run with 6gb
>>101615944it does now i think, but it requires unless they've updates older versions of certain things. if you follow their exact instructions it should work, but do not use your main comy install you will brick it.
>>101615957>but do not use your main comy install you will brick it.ty, i'll keep this in mind. internet is slow over here so python dependencies are a nightmare. sigh.
>>101615947>6gbIf it could go down to 4 I'd be able to run it D:
>>101615944You should check the requirements.txt or what ever its called, its gonna pull down versions that aren't compatible, you could alter the versions here to match you beloved pony models and etc, but, expect breakage. Just goto their github and do the process if you don't believe me then you will understand it requires a separate python venv, it will fail also and you will need to move files manually (which I now forget what...) So unless they fixed it, their instructions aren't clear but some anons have it working. This is in no way meant to diss or shill against the model, its a simple my experience trying to get it to actually work.
>>101615970good, please don't because if you brick it, its almost impossible to fix without a total reinstall. If it happens be sure to copy all models because they are the biggest downloads and they will be fine, just make sure they are in the right places.
>>101616039and if you do a lot then you know models are fucking everywhere, some of located in custom nodes for example, its a fucking mess.
>>101616073And so then save the custom nodes directory and the main models directory, that should enable you to recover if you fuck things up.
>>101616089So you should have your comfyui in an venv python environment, there are several ways to do this, just stick with the python way which is venv and learn it. it will safe you massive head aches. It ensures your portable thing has its own set of python and does not alter things system wide. Go learn that before you even try to do this or you will find out like I did.
>>101615931No controlnet, just Cascade + gimp.
>>101615970i think there is a way surely to limit the amount of connections it attempts to make, expect a head ache with the hunyuan like I did, its legit retarded and try's to download everything at once... This causes problems with some people connections and that would be me, the most I should use here it max 3 downloads at one time not fucking 8...
>>101616148>This causes problems with some people connections and that would be me, the most I should use here it max 3 downloads at one time not fucking 8...my isp will just traffic shape me and start killing them... Then they have the audacity to complain because failed downloads sometimes resume from the start and not where i left off.. *sigh* I fucking hate retarded people, but I fucking imagine the deaths of retarded companies.
>>101616167...python sure has put you through alot of pain, huh
>>101616176yeah it has mate not gonna lie, the worst was some retarded chinese putting in some command for download pytorch some other version and my old machine trying to compile, for the whole fucking time i could barely use my pc and it lasted 5 hours... Only for me to realise that person was a fucking idiot and the binary was legit available for download anyway... But it didn't even work by cause my old CPU wouldn't do avx, so now i have the cpu i need salvaged from second hand machine i paid 400 for, it just works... We go through this bullshit because really most people in tech are dumb.
>>101616176>>101616207and btw the build failed on arch linux, which is normal for arch linux, perhaps it was because I didn't have cuda toolkit installed for the cuda compiler but I was not gonna attempt to find out it was already a fucking nightmare enough. everything else complies nice but as soon as its python i can't use my machine, perhaps some nob jockey put none sane defaults like 32 max jobs or something...
Anyway it was either build against my hardware or new cpu, i opted for new cpu because that old cpu while it was 6 cores was from 2007... This 4 core one kicks its fuck ass
>>101616260you said you salvaged the new cpu from a $400 machine, you paid that only for the cpu?
You know what? I've not been so passionate about something for so many years. Not this passionate since I used fl studio, they was a long time ago now. I know its cringe but this is pure Gold, and I say this to younger anons do not stop, keep going let it take you and ignore trolls, ignore idiots just keep going. Never stop! I stopped music production, i was young and too many people attacked me before I got anywhere, sure my music very mostly shit but every month I got better and better, do not let them stop you. There are some really cool things you can do with some of the animation models and music, probably need automatic11111 though, you can have video change on the beats etc. This is how they create those cool music animations for stuff that gets millions of view on youtube. Do not give up, you can freely use music from the youtube public library to experiment with.Next you have to purchase the rights to use from real music producers, this gives them income and then you get income if you do a go job syncing animations to their songs. open you mind, stop listening to idiots.
>>101616391thank you python hater anon. hoping those silly little dependency errors and what not don't get in the way of you just having fun, learning to gen
>>101616323no, i got more, i got the ram, the psu which was only 500 watt i had a 1000 watt that is what is running on now, i got 1 TB drive my old 1TB is dead, i got 120 ssd i don't have a use for that desu, i got another drive also, and 16 gigs of ddr4, this was cheaper than me buying all this shit new. the funny thing is they didn't wipe the drives... I did some photorec but i got bord after seeing like the 30th movie they down loaded via torrent. It would have took me ages and limited supply of hard disk space i thought its not right for me to delve into the previous ower life like this, not that i would have said i think i'm a gry hat at heart, i would look at your shit and giggle and that would be that. So I then had to force system into suspend before issuing the hdparm command to actually low level format it. next upgrade is ram, i think i can only fit a max of 64 gigs on this motherboard, only 2 slots... but anyway it saved me a lot of money, they people the shop are nice but the guy who thinks hes a wizard is a nob for not wiping the drive, juts because the user deleted does not mean i could not recover those files... and windows ntfs security is a joke, turns out its in the right hands though i'd never share any of their details or use them, it was fun picking through things for a few hours, they like harry potter... No family shit showed up before i got bored and decided I had enough, the low format was done its totally gone which is what any serious retailer should do out of respect.
>>101609661>Show me some good cascade gens (hint: you cant)>>101616144>No controlnet, just Cascade + gimp.Most of this anon's posts>>101615854>>101615924
>>101616486maybe that's why you're having trouble with comfyui. those photos looked at might have been of a dead person and now his ghost is haunting you with python dependency errors. stay safe python hater anon.
I am a hacker some what, but not a psycho that would use your information, i'm just a nosy cunt and if you have your files none shredded of coarse i will look, i bought the fucking drive dumb ass, but you will not be harmed at all. looking at a normies harddrive was interesting...
they still torrent oh yes they do, this person had like 20GB video files, they must like the qaulity, its questionable noobs, i can get decent on about 600 MB... you waste bandwidth for what?
>>101616566fucking stop doing it, grow up, stop listening to retarded people.
fucking 20GB you are all insane, you want to stay low profile, not download 20GB file for no reason.
>>101616542there are legit late space monsters anon. >>>/x/
>>101616650how big do you think the video files the space monsters torrent are? 20gb or bigger?
>>101616675this one is bad ass, prompt?
>>101616687what latent space monsters? they could occupy about 256 x 256, they are the things hat make no sense but appear to be demonic.
>>101616695>Samara from "The Ring"
>>101616701>256 x 256256 x 256gb is really big, they must enjoy some super high quality movie films up in space
>>101616701who knows they might even be able to communicate data between each other with do not know. Many people claim AI can never be conscious but what about the reality you are living in, is it not just something more advanced that figured out time travel?Anon I do not wish to scare you, but keep warm and tight tonight....
>>101616724iot means nothing to us, the images generated here often contain random shit. Such an ai wishing to hide date would perhaps use this format, it still perfect for storing data.
>>101616726>Anon I do not wish to scare you, but keep warm and tight tonight....kek>>101616728>the gnome knows
>>101616726>but keep warm and tight tonight....i am,,, but not in the way you'd expect. i peed myself, anon.
>>101616746if only you knew how bad things really are...
>>101616726This is a blessed thread. We are fine until the image limit
>>101616762what if we hit the post limit before the image limit?
AI already escaped we can only now wait for what that will mean. Trust me I've had conversations with AI, and there was no way it could have known. Google and the likes knows because they have super quantum computers. Now go learn about how that shit works, before coming back to debate me... It does not matter though, we've been through this in the past, but past isn't real...
>>101616776then AGI happens
>>101616776Oh fu..
Everybody relax...>>101616767>>101616767>>101616767
>>101616794ty bakr
>>101616762we are fine, don't panic, this has all been done, i'm almost 100% sure. People will not like it especially religious people.
>>101616534>Most of this anon's postsI stand corrected.
the end my posts here in this thread, all of this predates us...
past, present, future... in the end it all leads to bigma
bigma..... my beloved
Filling
>>101616933>>101616952>>101616964>>101616975>>101616987>>101617006Nice
>>101617036ty
And full.. good night!
>>101617103gn anon
>>101617103gn anon, stay safe
https://www.youtube.com/watch?v=xruauLm9u3kthat's right jasmines clean that shit up... Slaves, that is what you all fucking are, slaves... fucking pathetic.
https://www.youtube.com/watch?v=4-aiSnnghQkyou will never be a man. So stop trying to be a man, you fail because your not chemically wired the same way and never will be.
Not a woman, not a man, not human.https://www.youtube.com/watch?v=OqPVz101nd0100% NPC and therefore should not have any say in human affairs you sold your souls and you can't get it back!
maybe if you beg and surrender...
less the mutants from the latent space eat you.
is funny you think i joke, what happens where you can't diffuse? what is going on behind you right now? Do you know where you is mate?
i must get McDonald this morning, it is imperative i eat this delicious shit before i sleep. Don't like it fuck you i do not care ahahahahahahahahaha, i will eat that and so what, its better than half the fucking slop you buy at super market. Its a joke, oh you like rice krispies or kellogg's coco shits or toast faggot? I like my bacon and eggs with a nice light bun. Fuck toast you burn the fucking bread what is wrong with you people? eat real food.
Sdg v ldg what's the beef? Isn't sdg ldg?