Thread #108066594 | Image & Video Expansion | Click to Play


Showing all 341 replies.
>>
Anonymous
02/05/26(Thu)13:53:03 No.108066646 Can SD make a comeback?
>>
Anonymous
02/05/26(Thu)13:53:12 No.108066647 huggingface is back up, what did they delete?
>>
Anonymous
02/05/26(Thu)13:54:20 No.108066656 >>108066594
kino fagollage.
>>
Anonymous
02/05/26(Thu)13:54:50 No.108066660 >>108066647
mrmerdachers repos
>>
Anonymous
02/05/26(Thu)13:55:27 No.108066661 >>108066656
why thank you anime2liveaction poster
>>
Anonymous
02/05/26(Thu)13:55:49 No.108066664 Fuck Comfy, fuck falseflags open source models, fuck giving money to randoms.
Also wake up, do you think the models releasing open weights actually want people to finetune them? Does BFL wants seeing people choose Chroma over their models? They probably don't give a shit, they care about getting good results with their own model and use the "open weight" as a form of marketing or publicity.
You are cumbrained stupid
>>
Anonymous
02/05/26(Thu)13:56:52 No.108066673 >>108066664
thanks for your #5342562 attempt at fudding and ragebaiting, here's a (you)
>>
Anonymous
02/05/26(Thu)13:57:16 No.108066679 How many people were turned against Comfy with this bot's FUD campaign?
>>
Anonymous
02/05/26(Thu)13:57:44 No.108066681 >>108066679
zero, this website is dead as fuck
>>
Anonymous
02/05/26(Thu)13:58:36 No.108066685 >>108066681
>still saying this retarded garbage even after the epstein files got released
lmao
>>
Anonymous
02/05/26(Thu)14:03:31 No.108066703 >rannigger's schizo nonsense still in the op
yall never learn
>>
Anonymous
02/05/26(Thu)14:08:42 No.108066725 >>108066703
You know what's genuinely sad? You spam this general to smear your ex-boyfriend that you haven't gotten over. Move on fren, plenty of fish in the sea.
>>
Anonymous
02/05/26(Thu)14:11:15 No.108066742 Tragic thing about trani's anti-comfy spam is that there are many legit criticisms of comfy but they got discredited next to his schizophrenic meltdowns.
>>
Anonymous
02/05/26(Thu)14:14:12 No.108066757 What's are the differences between chroma2_kaleidoscope.safetensors, chroma2_kaleidoscope_experimental.safetensors and chroma2_kaleidoscope_experimental_reward_model.safetensors ?
Anyone here in his trooncord that knows?
>>
Anonymous
02/05/26(Thu)14:15:33 No.108066766 >>108066725
what's sad is that you're taking sides in an avatartroon battle
>>
Anonymous
02/05/26(Thu)14:16:18 No.108066770 >>108066757
>dude literally never documents anything like at all
>never even finishes his projects
>there are people arguing that he should be given 1 million dollars in grant money
>>
Anonymous
02/05/26(Thu)14:17:56 No.108066780 >>108066770
I don't disagree with any of these.
I am just curious what precise variant of schizo training methodology he is using now.
>>
Anonymous
02/05/26(Thu)14:18:08 No.108066782 >>108066770
give him 10
>>
Anonymous
02/05/26(Thu)14:22:38 No.108066803 you would think ai would progress where it could tell me how to install sage attention
>>
Anonymous
02/05/26(Thu)14:24:15 No.108066808 >>108066766
Pretending to be a neutral third party isn't going to work when YOU have been spamming this thread into being unusable for months. But keep pretending calling out your threadshitting out is "taking sides."
>>
Anonymous
02/05/26(Thu)14:24:48 No.108066811 >>108066803
You should be progressed enough as a human being to figure it out yourself.
>>
Anonymous
02/05/26(Thu)14:25:00 No.108066812 >>108066803
imagine getting filtered by this, post hand
>>
Anonymous
02/05/26(Thu)14:25:31 No.108066817 here's your slop
>>
Anonymous
02/05/26(Thu)14:27:33 No.108066830 >>108066803
Are you literally unable to do:
activate venv
cd sageattention
pip install .
???
>>
Anonymous
02/05/26(Thu)14:32:31 No.108066860 >>108066803
chatgpt can, at least, you have to tell it what you want, and spam it with any errors you are getting
>>
Anonymous
02/05/26(Thu)14:32:59 No.108066865 >>108066830
>desktop
yikes
>>108066803
hop you arerunning portable and not desktop
>>
Anonymous
02/05/26(Thu)14:33:23 No.108066866 >>
Anonymous
02/05/26(Thu)14:35:04 No.108066878 >>108066866
neat
>>
Anonymous
02/05/26(Thu)14:35:39 No.108066884 >posting on /g/
>not understanding virtual environments
y'all goofy
>>
Anonymous
02/05/26(Thu)14:36:56 No.108066894 Blessed thread of frenship
>>
Anonymous
02/05/26(Thu)14:37:03 No.108066896 >>108066884
It’s shocking how esoteric that knowledge seems to be to regular people.
>>
Anonymous
02/05/26(Thu)14:37:24 No.108066899 does the GOAT bigstationW still visit this forsaken place?
i'm once again asking for Anima NAG implementation.
>>
Anonymous
02/05/26(Thu)14:38:46 No.108066906 >>108066899
Use a better model that doesn't need nag.
>>
Anonymous
02/05/26(Thu)14:39:38 No.108066910 >>108066906
it doesn't. i just like snake oils
>>
Anonymous
02/05/26(Thu)14:39:55 No.108066912 >>108066817
Too many anon think professional looking photography is slop...
But they're amazed by shitty cheap blurry smartphone with nasty like jpeg compression and shitty toes and fingers
>>
Anonymous
02/05/26(Thu)14:40:52 No.108066920 >>108066896
i know ai will take a lot of jobs but not for anyone who has the patience to read one manual, one time
>>108066906
>>108066899
simply learn how to prompt
>>
Anonymous
02/05/26(Thu)14:41:48 No.108066925 >>108066912
dude its soul!!! lol
I remember when qwen img got published, I just applied a grain filter on top of the gens and everyone here was clapping like a retard
removing detail tricks your brain.
>>
Anonymous
02/05/26(Thu)14:42:41 No.108066931 >>
Anonymous
02/05/26(Thu)15:01:34 No.108067027 so i read a bunch of bullshit on r*ddit and did some quick comparisons. it does look like if you use adafactor to train a Z-Image lora and apply it to Z-Image Turbo you get better effect under 2 strength. that said the second link is still using adamw8bit and it works plenty. (nsfw).
adafactor: https://files.catbox.moe/jkjtrw.jpg
adamw8bit: https://files.catbox.moe/zjepyc.jpg
>>
Anonymous
02/05/26(Thu)15:08:55 No.108067064 >>108067027
Z who? wtf is z-image turbo did you just make that up?
>>
Anonymous
02/05/26(Thu)15:11:52 No.108067086 lil pipsqueak
>>
Anonymous
02/05/26(Thu)15:12:25 No.108067090 chadly taking the barmaid
>>
Anonymous
02/05/26(Thu)15:12:58 No.108067095 >>
Anonymous
02/05/26(Thu)15:13:31 No.108067099 >>
Anonymous
02/05/26(Thu)15:14:03 No.108067103 idol and his superfan
>>
Anonymous
02/05/26(Thu)15:25:13 No.108067166 >>108066148
thank
>>
Anonymous
02/05/26(Thu)15:26:50 No.108067174 >>108067064
yeah
>>108067027
basically i'm still training zit loras
>>
Anonymous
02/05/26(Thu)15:29:18 No.108067185 >>
Anonymous
02/05/26(Thu)15:41:50 No.108067246 yall make fun of me for posting 1girls but then use several of my 1girls in the collage
fuck you
>>
Anonymous
02/05/26(Thu)15:43:53 No.108067257 >>108067246
HE comes for your 1girls
>>
Anonymous
02/05/26(Thu)15:44:45 No.108067267 >>108067086
>>108067090
>>108067095
>>108067099
>>108067103
im starting to suspect a foid actually posts here
>>
Anonymous
02/05/26(Thu)15:45:44 No.108067272 >>108067267
anon not a single person with a vagina has ever touched comfyui
>>
Anonymous
02/05/26(Thu)15:46:26 No.108067277 >>108067272
a tranny wouldn't make gens like that
>>
Anonymous
02/05/26(Thu)15:46:56 No.108067280 >>108067277
that anon is just a guy my guy
>>
Anonymous
02/05/26(Thu)15:47:00 No.108067282 >>108067272
>he forgot about the trans community again
>>
Anonymous
02/05/26(Thu)15:47:14 No.108067286 >>108067267
a wom*n mogging all the photorealistic sloppers would be unfortunate
>>
Anonymous
02/05/26(Thu)15:48:01 No.108067290 >>
Anonymous
02/05/26(Thu)15:48:53 No.108067297 gm ai sisters
>>
Anonymous
02/05/26(Thu)15:51:01 No.108067314 I'm 100% sure that a big part of Anima being good is that it's only captioned in one language, as opposed to two or even three like a lot of the other recent anime models on post-SDXL architectures
>>
Anonymous
02/05/26(Thu)15:51:49 No.108067319 >>108067314
>being this xenophobic
>>
Anonymous
02/05/26(Thu)15:52:20 No.108067322 >>108067282
What is the transcommunity and how do I sign up?
>>
Anonymous
02/05/26(Thu)15:54:40 No.108067343 >>108067322
>go to tumblr.com
>make an account
>>
Anonymous
02/05/26(Thu)15:56:20 No.108067356 man why the fuck does the comfy acestep workflow not let you just put in minutes. im running out of chatgpt uses making a table of minute and seconds to seconds conversions
>>
Anonymous
02/05/26(Thu)15:57:18 No.108067365 >>108067314
multi lang captions gotta be the SOTA of mental retardation
literally could just use chatgpt or whatever to convert their shitty ass language to english instead of confusing the model
>>
Anonymous
02/05/26(Thu)15:58:11 No.108067372 >>
Anonymous
02/05/26(Thu)15:58:41 No.108067378 >>108067356
ask chatgpt to make you a node to do that??
>>
Anonymous
02/05/26(Thu)15:59:43 No.108067385 >>108067319
I'm not being xenophobic, it doesn't matter what the languages are, it matters how many there are. Captioning like that is basically guaranteed to degrade results when prompting in probably all of them compared to what the results would be like with a model captioned in just one.
Additionally I'm not sure it's even useful to literally have multi-lingual captions like that to begin with. I've experimented with prompting my Qwen Image Loras, which are only captioned in English with NLP captions, using machine-translated-to-Chinese prompts before, and in all cases they produce nearly the exact same result as the same prompt in English on the same seed, presumably because Qwen the TE already knows all the languages and basically generates the same encoded data for the image model either way.
>>
Anonymous
02/05/26(Thu)16:00:18 No.108067389 >>108067322
>>
Anonymous
02/05/26(Thu)16:00:29 No.108067392 >>
Anonymous
02/05/26(Thu)16:04:04 No.108067419 Fk9b can make correct anatomy :v
used as an edit model with correct anatomy input...
>>
Anonymous
02/05/26(Thu)16:04:11 No.108067421 >>108067267
>woman
uhmmm im a big guy tbqh, why do u think im a girl, nerd? u wanna fuck me? lol
>>
Anonymous
02/05/26(Thu)16:05:54 No.108067431 >>108067421
I kek'ed x)
>>
Anonymous
02/05/26(Thu)16:07:34 No.108067450 >>108067356
Are you the "Microwave Time" poster?
>>
Anonymous
02/05/26(Thu)16:08:38 No.108067455 >>108067421
you make very feminine gens for a big guy
>>
Anonymous
02/05/26(Thu)16:09:05 No.108067459 >>108067322
>>
Anonymous
02/05/26(Thu)16:12:48 No.108067489 >>
Anonymous
02/05/26(Thu)16:14:27 No.108067507 >>108065361
>>108065361
workflow pls
>>
Anonymous
02/05/26(Thu)16:16:43 No.108067529 >>108067314
>>108067365
qwen encoder is already trained on chinese too so that's wrong
>>
Anonymous
02/05/26(Thu)16:17:44 No.108067535 >>
Anonymous
02/05/26(Thu)16:18:33 No.108067542 >>108067450
>>
Anonymous
02/05/26(Thu)16:18:48 No.108067543 >>108067535
reminds me when i used illustrious with the wrong settings
>>
Anonymous
02/05/26(Thu)16:21:23 No.108067556 >>108066594
>>Maintain Thread Quality
>https://rentry.org/debo
>https://rentry.org/animanon
why are you obsessed with putting schizobabble in the op?
>>
Anonymous
02/05/26(Thu)16:22:04 No.108067563 >>
Anonymous
02/05/26(Thu)16:23:17 No.108067568 when are we finally going to admit that generating from scratch has reached its limit? it looked like slop in 2021 and it still looks like slop today
edit based models are the future and we should put all our effort into optimizing them
>>
Anonymous
02/05/26(Thu)16:24:05 No.108067580 >>108067556
good morning ani
did you have any dreams?
i had a dream that i was making out with a hapa with braces.
shit was fucking cash, i was playing with her tiddies and pussc.
and then her parents came in and cock blocked the whole thing .
ye.
>>
Anonymous
02/05/26(Thu)16:24:35 No.108067587 >>108067568
I don't care about real people.
>>
Anonymous
02/05/26(Thu)16:25:10 No.108067592 >>108067580
>good morning ani
meds
>>
Anonymous
02/05/26(Thu)16:26:16 No.108067600 >>108067568
yup, the real shit is DragGAN https://github.com/XingangPan/DragGAN
>>
Anonymous
02/05/26(Thu)16:30:56 No.108067641 >>
Anonymous
02/05/26(Thu)16:31:45 No.108067645 >>108067592
i already too em
>>
Anonymous
02/05/26(Thu)16:36:24 No.108067687 >>
Anonymous
02/05/26(Thu)16:41:54 No.108067723 unless anima magically improves its anatomy and background coherency by at least 2x, it's looking like a really disappointing "upgrade".
stylistically it's great but practically it's unusable when 9/10 gens have some kind of irredeemable body horror and 9/10 of the remaining ones have some other huge flaws that ruin them
at this point the hobby might as well be dead to me until someone manages to make a non slopped full booru finetune on klein or zib, with the same level of creativity as noob or anima but not as incompetent. i have little faith in anime being fixable without slopping it up, due to its size
>>
Anonymous
02/05/26(Thu)16:42:41 No.108067730 >>108067723
xister I think its just a early stage proof of concept, it was only trained on 512 too, keep calm
>>
Anonymous
02/05/26(Thu)16:44:59 No.108067744 any news on flux2 character loras yet or is training still shit?
>>
Anonymous
02/05/26(Thu)16:45:29 No.108067746 >>108067723
That sound like one hell of a skill issue lol. I have genned hundreds of image and never got any body horror worst I got is bad hands.
>>
Anonymous
02/05/26(Thu)16:45:59 No.108067749 >>108067723
put this on your blog
>>
Anonymous
02/05/26(Thu)16:47:13 No.108067763 >>108067723
At some point a model can only do so much for a braindead user, so yeah probably should quit while you are ahead
>>
Anonymous
02/05/26(Thu)16:47:39 No.108067766 >>108067730
it's a 2b with .6b text encoder, i'd say they did an impressive job considering that, but it's cope to think it will get more than marginally better. training beyond 512 will only improve detail but not fix the other flaws
>>
Anonymous
02/05/26(Thu)16:48:03 No.108067769 >>
Anonymous
02/05/26(Thu)16:48:46 No.108067774 >>108067723
learn2draw and fix the flaws yourself
>>
Anonymous
02/05/26(Thu)16:49:03 No.108067777 >>
Anonymous
02/05/26(Thu)16:49:53 No.108067782 >>108067746
must be easy when all you gen is 1girl, looking at viewer
>>108067774
there are too many, i might as well just use sdxl at that point
>>
Anonymous
02/05/26(Thu)16:50:14 No.108067788 >>108066594
HIROSHIMOOT IS A GREEDY GOOK.
>>
Anonymous
02/05/26(Thu)16:51:20 No.108067795 >>108067766
>training beyond 512 will only improve detail but not fix the other flaws
yes but even the 512 training isn't finished, I have no clue how much compute they have and how much this kind of model requires so god knows what we will get out of this, even if the anatomy/capabilities are only on par or slightly better than sdxl I'll still take it for the better rendering of styles/textures
>>
Anonymous
02/05/26(Thu)16:51:35 No.108067797 can anustudio support Z Image or edit models yet
>>
Anonymous
02/05/26(Thu)16:54:39 No.108067819 >>108067769
youre a boring autist
>>
Anonymous
02/05/26(Thu)16:55:34 No.108067825 What's the best way to install Git if I'm only going to be using it for Stable Diffusion?
>>
Anonymous
02/05/26(Thu)16:56:15 No.108067832 >>108067825
bless your heart
>>
Anonymous
02/05/26(Thu)16:57:48 No.108067843 sample pics look much better compared to ZIT when training with ZIM, I guess that's a good sign
>>
Anonymous
02/05/26(Thu)16:59:36 No.108067855 >>108067843
same for me, but the end results were dogshit and dont even work on turbo
i gabr up on zim training
>>
Anonymous
02/05/26(Thu)17:00:25 No.108067858 >>108067855
s k i l l i s s u e
>>
Anonymous
02/05/26(Thu)17:02:12 No.108067872 >>
Anonymous
02/05/26(Thu)17:02:52 No.108067879 Does Triton work with WebUI Forge Neo?
>>
Anonymous
02/05/26(Thu)17:04:59 No.108067889 >>
Anonymous
02/05/26(Thu)17:06:05 No.108067897 >>108067858
feel free to cope if you want, but its not for me
the distill lora is dogshit
>>
Anonymous
02/05/26(Thu)17:08:53 No.108067919 >>108067897
incoherent response
it's time for your meds anon
>>
Anonymous
02/05/26(Thu)17:09:11 No.108067923 >>108067795
listen i do hope it improves significantly but i have my doubts.
all i'm saying is im seeing some very fucked up poses and proportions on fairly simple prompts, including out of place floating objects and really bad clipping, and absolutely mangled hands and feet if i try to go for any kind of interesting style. if it gets significantly better at backgrounds it would at least become usable with inpaints, i guess.
>>
Anonymous
02/05/26(Thu)17:10:45 No.108067933 >>108067723
Just wait for the Anima API, it’s going to be amazing. That’s why Comfy sponsored it with 1million dollars ;)
>>
Anonymous
02/05/26(Thu)17:14:07 No.108067958 >>
Anonymous
02/05/26(Thu)17:16:16 No.108067968 >1m USS for a body horror 2B slop model.
Yeah, so now we know the minimum amount of money needed to make an anime model. Anyone trying to invest less than that is doomed to snake oil and failure
>>
Anonymous
02/05/26(Thu)17:17:19 No.108067976 >>108067968
thanks for fudding again, we appreciate your contributions to the thread
>>
Anonymous
02/05/26(Thu)17:20:55 No.108067994 >>108067968
anon, it's called money laundering.
lmao
>>
Anonymous
02/05/26(Thu)17:20:56 No.108067995 >1mUSS for a body horror 2B slop model.
So now we know the bare minimum amount of money needed to make an 3/10 anime model.
Anyone trying to invest less than that is doomed to snake oil and failure.
>>
Anonymous
02/05/26(Thu)17:21:57 No.108068001 >$1M for a body horror 2B slop model.
That’s the floor for a 3/10 anime model.
Anyone spending less is selling snake oil and heading for failure.
>>
Anonymous
02/05/26(Thu)17:23:29 No.108068008 unc drowning in Culture rn
>>
Anonymous
02/05/26(Thu)17:25:33 No.108068016 haven't forgotten yall niggas chompin at the bit to download every chroma epoch
>>
Anonymous
02/05/26(Thu)17:26:03 No.108068019 fix your bot bro
>>
Anonymous
02/05/26(Thu)17:26:41 No.108068028 >>108068019
fixed
>>
Anonymous
02/05/26(Thu)17:28:12 No.108068036 >>108067976
If we can't criticize Comfy because he's the mod in /ldg/ cord, mark this as a pro ComfyOrg general in the OP so we can take our opinions elsewhere.
>>
Anonymous
02/05/26(Thu)17:30:49 No.108068062 >>108068036
you don't need to ask permision to leave, just leave
>>
Anonymous
02/05/26(Thu)17:35:08 No.108068089 >>108068062
Dismissive response that ignores my point about transparency and disclosure of bias, how cute.
>>
Anonymous
02/05/26(Thu)17:37:39 No.108068108 >>108068089
and here comes the racism
>>
Anonymous
02/05/26(Thu)17:41:35 No.108068132 So if I have to be an /ldg/ bro, I have to be:
Pro Comfy
Pro Lodestone?
What else? I’m a newfag and I’m making my /ldg/ bro BINGO.
>>
Anonymous
02/05/26(Thu)17:42:02 No.108068138 >>108068001
>>108067995
anything else you wanna vent?
>>
Anonymous
02/05/26(Thu)17:42:04 No.108068139 >>108067819
STFU
>>108067769
neat
>>
Anonymous
02/05/26(Thu)17:42:23 No.108068143 ```The exact same real photographic blue haired East Asian woman from photographic image 1 is now standing in the same right hand extended pose as the green haired girl from anime image 2 and wearing the same clothes as the green haired girl from anime image 2 against the exact same background from anime image 2.```
>>
Anonymous
02/05/26(Thu)17:43:49 No.108068155 >>108068089
Hey hey hey antisemitism isn’t welcome here!
>>
Anonymous
02/05/26(Thu)17:44:20 No.108068159 >>108067723
If you get that much body horror on Anima somehow your prompts have to be unimaginably retarded and you're probably using equally retarded sampling settings
>>
Anonymous
02/05/26(Thu)17:44:45 No.108068163 >>108068132
gay
>>
Anonymous
02/05/26(Thu)17:45:27 No.108068171 >>108068143
Now do it reverse and marvel as Klein slops the animu girl.
>>
Anonymous
02/05/26(Thu)17:46:28 No.108068180 >>108068159
How about your model is shit and you take the criticism like a good dev for once?
>>
Anonymous
02/05/26(Thu)17:46:38 No.108068182 >still no wan news
>still no good ipadapters for flux/chroma
sigh..
>>
Anonymous
02/05/26(Thu)17:47:29 No.108068191 >>108068180
Hey! >>108068155
>>
Anonymous
02/05/26(Thu)17:48:07 No.108068196 >>108068182
>ipadapter
>when edit models exist
LOLLLLLLLLL
>>
Anonymous
02/05/26(Thu)17:50:30 No.108068209 >>108068182
>ipadapter
3 years ago called
>>
Anonymous
02/05/26(Thu)17:50:55 No.108068213 lmfao all the shit emojis
>>
Anonymous
02/05/26(Thu)17:51:47 No.108068221 >>108068213
Nigga really needs to kick all the indians out
>>
Anonymous
02/05/26(Thu)17:51:57 No.108068223 >>
Anonymous
02/05/26(Thu)17:52:09 No.108068227 >cant go 10 posts without mentioning discord or posting discord shit
why dont you fucking stay in discord with your stupid petty drama like holy fucking shit no one CARES
>>
Anonymous
02/05/26(Thu)17:52:26 No.108068230 I want to laugh a bit, does anyone have, or can redirect me to, the links from last year when Flux Kontext came out and everyone shilled it like crazy, while some were saying, "This is the future" or things like that?
>>
Anonymous
02/05/26(Thu)17:52:45 No.108068234 >>108068223
which anime
>>
Anonymous
02/05/26(Thu)17:52:50 No.108068236 >>108068227
stay mad
>>
Anonymous
02/05/26(Thu)17:53:27 No.108068240 >>108068227
discord general
>>
Anonymous
02/05/26(Thu)17:54:23 No.108068249 >>108068143
>>
Anonymous
02/05/26(Thu)17:54:25 No.108068250 >>108068213
why are peolpe mad, lmao?
>>
Anonymous
02/05/26(Thu)17:55:11 No.108068256 >>108068223
>>108068249
disgusting klein slop
>>
Anonymous
02/05/26(Thu)17:56:15 No.108068269 >>108068250
>ai-generated marketing plug for jeet patreon scams
Why wouldn't you be?
>>
Anonymous
02/05/26(Thu)17:56:29 No.108068271 bro is my sword big enough?
>>
Anonymous
02/05/26(Thu)17:57:37 No.108068281 >>108068171
What are you expecting her to look like exacty?
>>
Anonymous
02/05/26(Thu)17:59:25 No.108068292 >>108068271
sensing gay energy here
>>
Anonymous
02/05/26(Thu)17:59:32 No.108068294 >>108067600
I don't get why everyone forgot about this. Having actual sliders to adjust an image is pretty nifty instead of praying to the RNG gods.
>>
Anonymous
02/05/26(Thu)18:03:35 No.108068323 >>108068294
Well start porting it to comfy.
>>
Anonymous
02/05/26(Thu)18:04:07 No.108068330 >>108068234
Yu-Gi-Oh, though idk if she appears in the anime.
>>108068256
Wow, rude.
>>
Anonymous
02/05/26(Thu)18:05:37 No.108068345 >go to >>>/biz/
>grab something
>drag and drop into klein "make it 3d"
>profit
>>
Anonymous
02/05/26(Thu)18:07:15 No.108068360 >>108068345
tasty
>>
Anonymous
02/05/26(Thu)18:11:56 No.108068396 >>108068345
>even /biz/ has post IDs
Maybe I should invest in crypto
>>
Anonymous
02/05/26(Thu)18:13:58 No.108068413 >>
Anonymous
02/05/26(Thu)18:21:31 No.108068482 >anima lora werks
lovely, now its just a matter of waiting for the final model
>>
Anonymous
02/05/26(Thu)18:22:03 No.108068488 >>108068171
I think you're acccking on this one my friend.
```The exact same closed eyes green haired anime girl from anime image 1 is now in the exact same kneeling pose as the blue haired East Asian woman from photographic image 2 and wearing the exact same tank top and pants and high heels as the blue haired East Asian woman from photographic image 2 against the exact same studio background from photographic image 2. The cob of corn is now completely gone.```
TLDR Klein can do basically anything with the right details specified and the right wording.
>>
Anonymous
02/05/26(Thu)18:23:17 No.108068502 >>
Anonymous
02/05/26(Thu)18:24:21 No.108068510 Subgraphs and show widgets in Comfy are neat. Why should I use bloated Swarm or Forge Neo if Comfy already wonned with this simple feature?
>>
Anonymous
02/05/26(Thu)18:25:48 No.108068521 >>108068482
Do you have a moment to talk about ComfyOrg culture?
>>
Anonymous
02/05/26(Thu)18:26:15 No.108068524 >>
Anonymous
02/05/26(Thu)18:26:37 No.108068525 >>108068482
What trainer supports it? I know the author said he wasn't going to add it to his own until later
>>
Anonymous
02/05/26(Thu)18:27:25 No.108068534 animu
>>
Anonymous
02/05/26(Thu)18:28:01 No.108068538 >>
Anonymous
02/05/26(Thu)18:28:05 No.108068541 >>108068525
I used this https://github.com/bluvoll/diffusion-pipe on WSL, someone also made a pr to kohya's sd-scripts
>>
Anonymous
02/05/26(Thu)18:28:28 No.108068545 >>108068488
You’re clearly not into anime, because that image absolutely massacres the character’s style. Anyone with a minimally trained eye can tell.
>>
Anonymous
02/05/26(Thu)18:29:04 No.108068552 >>
Anonymous
02/05/26(Thu)18:29:15 No.108068556 Anyone who claims Klein replaces traditional IPAdaters is unable discern slopped models from non slopped models btw.
>>
Anonymous
02/05/26(Thu)18:29:29 No.108068559 >>108068482
it's not your lora, tdrussell owns it
>>
Anonymous
02/05/26(Thu)18:31:13 No.108068571 >>108068482
If you know a bit about human muscle anatomy, it’s obvious the model messed up the traps, delts, lats, and quads and added muscles that don’t exist.
>>
Anonymous
02/05/26(Thu)18:32:02 No.108068579 >>108068559
>it's not your lora, tdrussell owns it
Similar licence to klein 9b? Loras cant be sold
>>
Anonymous
02/05/26(Thu)18:33:41 No.108068596 >>108068525
he added it
>>
Anonymous
02/05/26(Thu)18:34:04 No.108068600 >>108068541
the main diffusion-pipe supports anima now
assuming the fork didn't fuck up the early implementation, it should be the same but I would use the official one
>>
Anonymous
02/05/26(Thu)18:34:18 No.108068602 >>108068579
It's just a troll.
>>
Anonymous
02/05/26(Thu)18:36:17 No.108068619 >>108068579
it can be sold but only by tdrussell. you have to purchase his licence first
>>
Anonymous
02/05/26(Thu)18:37:58 No.108068639 >>108068619
>it can be sold but only by tdrussell
and Comfy
>>
Anonymous
02/05/26(Thu)18:38:45 No.108068647 >>108068619
forgot to add you have to buy the licence first in order to sell it
>>108068639
kek
>>
Anonymous
02/05/26(Thu)18:41:36 No.108068671 arent jeets grifters the only ones who sell loras
>>
Anonymous
02/05/26(Thu)18:41:49 No.108068672 >>108068619
>>108068647
welp, there goes my side gig
>>
Anonymous
02/05/26(Thu)18:43:03 No.108068684 >>108068671
gofuckem will still be able to sell his workflows
>>
Anonymous
02/05/26(Thu)18:43:27 No.108068687 >>108068671
Probably the ones who advertise the most. I get private requests.
>>
Anonymous
02/05/26(Thu)18:44:51 No.108068698 >>108068671
Why would grifters sell the loras instead of selling the slop generated from the lora to boomer coomers
>>
Anonymous
02/05/26(Thu)18:45:54 No.108068704 >>108068545
Get real, she was generic AI sloppa shine style to begin with lmao
>>
Anonymous
02/05/26(Thu)18:46:09 No.108068706 >>108068196
>>108068209
>having to load separate models and separate workflows
ngmi
>>
Anonymous
02/05/26(Thu)18:46:34 No.108068707 >>108068698
patreon exclusives. you bait the minnows and dolphins with images, you make a decent amount with whales
>>
Anonymous
02/05/26(Thu)18:46:48 No.108068709 >>
Anonymous
02/05/26(Thu)18:51:28 No.108068746 >>108068510
Based. It's also more compact, maybe have inpaint, controlnet, and img2img subgraphs appear when you right click on the gen.
>>
Anonymous
02/05/26(Thu)18:51:44 No.108068747 >>108068707
From what I recall I thought patreon didn't allow selling "NSFW AI tools" or something, which probably includes loras. Apparently they allow NSFW AI slop though, as long as it isn't photorealism.
>>
Anonymous
02/05/26(Thu)18:53:25 No.108068765 >>108067459
Klein only knows one train interior. I get the exact same train in my prompts
>>
Anonymous
02/05/26(Thu)18:56:06 No.108068785 >>108068746
>inpaint
just use edit models, bro
>img2img
just use edit models, bro
>controlnet
just use edit models, bro
>>
Anonymous
02/05/26(Thu)18:56:55 No.108068793 >>108068510
>>108068746
weak comfyshills
>>
Anonymous
02/05/26(Thu)18:57:15 No.108068798 >>
Anonymous
02/05/26(Thu)18:57:56 No.108068803 >>108068250
>it's comfy when comfyorg controls ever aspect of the community and grifts
I'm not feeling very comfy with this
>>
Anonymous
02/05/26(Thu)18:58:59 No.108068813 >>108068785
>controlnet
>just use edit models, bro
It's a legit downgrade from controlnet because edit still alters the image composition slightly. Ideally you'll want to use controlnet with edit models together. Also you can still inpaint with edit models.
>>
Anonymous
02/05/26(Thu)19:00:25 No.108068822 Where else has the open source solution assraped the competition so hard?
>>
Anonymous
02/05/26(Thu)19:02:57 No.108068841 >>108068822
where else has it's users ass raped and price gouged with fanfare by the open source solution?
>>
Anonymous
02/05/26(Thu)19:03:05 No.108068843 >>108068822
I think P-Diddy
>>
Anonymous
02/05/26(Thu)19:03:37 No.108068846 >>108068793
Nah, look, this is a txt2img + hiresfix workflow. Look how compact it is, and I've still got space left to add an adetailer.
>>
Anonymous
02/05/26(Thu)19:04:22 No.108068851 How much of a performance uplift can I expect going from a 3060 Ti to a 5070 in Stable Diffusion WebUI's? Any optimizations I can unlock?
>>
Anonymous
02/05/26(Thu)19:05:19 No.108068862 >>108068851
'bout three fiddy
>>
Anonymous
02/05/26(Thu)19:05:45 No.108068868 >>108068851
yeah some speedups only work on more recent cards like the Nvidia quants
>>
Anonymous
02/05/26(Thu)19:07:33 No.108068882 >>108068868
How fucked am I going to be versus the 5060 Ti because of the 12GB of VRAM versus 16GB? I ended up picking the 5070 since it was only $50 more at that point and because I passed on a used 3090 earlier since I had reliability/power concerns.
>>
Anonymous
02/05/26(Thu)19:09:08 No.108068895 >look at this thing other UIs do better
>>
Anonymous
02/05/26(Thu)19:09:25 No.108068898 >>108068882
Always go for as much as vram as possible.
>>
Anonymous
02/05/26(Thu)19:10:33 No.108068906 ayo dis nigga bout to get MORE rtx 6000 pros
>>
Anonymous
02/05/26(Thu)19:15:17 No.108068944 nat the lich lora on anima
>>
Anonymous
02/05/26(Thu)19:15:48 No.108068947 >>108068906
Decade of chroma letsgo.
>>
Anonymous
02/05/26(Thu)19:16:31 No.108068962 >>108068746
>>
Anonymous
02/05/26(Thu)19:17:33 No.108068975 well, now I know Anima (NVIDIA Cosmos Predict-2) is bad at anatomical drawings.
>>108068502
crazy hand like hold on
>>
Anonymous
02/05/26(Thu)19:17:42 No.108068976 >>108068906
>daddy Yoland in the comments
kyaa~
>>
Anonymous
02/05/26(Thu)19:20:11 No.108068992 >>108068538
Flux 2 is super pro lol. I can tell it's not a real photo, because of my experience with photography (ignoring that it's a fantasy color), but most people would prefer something like this over a real photo. Why? The ears or the nose would be blurry. Portraiture is hard.
>>
Anonymous
02/05/26(Thu)19:23:14 No.108069016 >>108068944
>>
Anonymous
02/05/26(Thu)19:24:50 No.108069025 >>108068746
sloop
>>108068962
KVNO
>>
Anonymous
02/05/26(Thu)19:25:57 No.108069036 AMDs average bonus was 115% holy shit
Lisa Su just said that she asks AI everything even personal questions
Amd is going to go even faster in 2026 and to try to scale up compute 10,000x
>>
Anonymous
02/05/26(Thu)19:28:01 No.108069054 >>108068906
So Comfy is basically doing charity work for model makers now, why doesn't he support LAXhar? LAXhar made Noob with only 80k, imagine what he could do with 1M.
>>
Anonymous
02/05/26(Thu)19:28:38 No.108069061 >>108069054
tell LAXhar to apply for the grant then
>>
Anonymous
02/05/26(Thu)19:30:06 No.108069070 >>108069061
I don't have a cord and i'm shy :$
>>
Anonymous
02/05/26(Thu)19:32:37 No.108069086 >be a nonnie
>don't have discord
why
>>
Anonymous
02/05/26(Thu)19:33:44 No.108069097 >>108068947
WE FEAST!
>>108069054
Who and what?
>>
Anonymous
02/05/26(Thu)19:35:25 No.108069108 >>108069054
Why'd you give money to someone who'd waste it on an sdxl clone again?
>>
Anonymous
02/05/26(Thu)19:35:55 No.108069115 >>108069036
Well, until models start to learn by themselves and not by data we could have infinite compute that they would still be dumb machines limited by our knowledge, a good AI would be an AI that would also work in the past, if you trained a LLM on 100 B.C. textbooks/knowledge it would be retarded and would not be able to develop modern science.
>>
Anonymous
02/05/26(Thu)19:36:55 No.108069121 >kyaa~
>>
Anonymous
02/05/26(Thu)19:40:02 No.108069154 I hate furfags
>>
Anonymous
02/05/26(Thu)19:47:10 No.108069220 >>108068898
Even over generational improvements such as Ampere to Blackwell?
>>
Anonymous
02/05/26(Thu)19:53:06 No.108069267 >>108069036
Just got the most direct insider confirmation that AMD doesn't care about consumer GPUs that I'll ever get at my current underling status
The head of compute (Jack) got asked by Lisa to answer questions about the DRAM supply crunch and he was like "The consumer GPUs that we care about a lot, we are allocating parts for"
So they don't care about any consumer GPUs except probably their latest workstation cards kek
I literally don't care if you believe me or care or not I'm just fulfilling my cultural obligation to share information no matter what
>>
Anonymous
02/05/26(Thu)20:35:39 No.108069515 >>108068906
>>
Anonymous
02/05/26(Thu)20:47:40 No.108069603 >>108068906
Is this assuming the diffusion-pipe guy has already gotten a mill to train?
>>
Anonymous
02/05/26(Thu)20:52:31 No.108069638 I don't like how the login system works on civit.ai with the whole send a link to your email thing, does your email have anything to do with the account that you login to afterwards? Is it tied to it at all?
>>
Anonymous
02/05/26(Thu)21:12:11 No.108069763 >>108069603
>one MILLION fucking dollars for anima
Why do people keep repeating this? The million bucks is clearly the total budget for the grant program. Hell, something like Z-Image was trained from scratch for only like 600k, obviously the training costs of a 2b parameter finetune aren't a million dollars.
>>
Anonymous
02/05/26(Thu)21:17:32 No.108069790 lmaoooooooooo
how many days will it take for the piece of shit training to actually be fixed, waitGODS win again
https://x.com/bdsqlsz/status/2019350879472873984
>>
Anonymous
02/05/26(Thu)21:19:58 No.108069802 >>108069790
How are the chinks so retarded they can't train on their model? there's alredy loras for ZiBase
>>
Anonymous
02/05/26(Thu)21:20:40 No.108069804 >>108069638
I don't trust using anything but a burner mail for that site, even though it doesn't say anything in their terms of service about selling user data I am certain that with download history and shit they are one big major privacy concern considering accounts are in fact tied to emails.
>>
Anonymous
02/05/26(Thu)21:21:30 No.108069811 >>108069802
iirc the loras are fine but to make a full-finetune -- picrel
>>
Anonymous
02/05/26(Thu)21:22:39 No.108069822 >>108069811
There's also special ppl who can't even train a lora for it.
>>
Anonymous
02/05/26(Thu)21:32:05 No.108069870 flux2 klein lora training fucking sucks donkey balls
so fucking inconsistent
>>
Anonymous
02/05/26(Thu)21:37:14 No.108069905 ComfyUI straight up broke two of my subgraphs, lol.
>>
Anonymous
02/05/26(Thu)21:37:52 No.108069910 >>108069515
rofl
>>
Anonymous
02/05/26(Thu)21:39:20 No.108069920 >>108069905
>do retarded stuff
>oh no shit broke!
lmao'd, thanks for the free laughs retard, keep at it.
>>
Anonymous
02/05/26(Thu)21:42:04 No.108069938 >>108069920
I didn't do anything, nigger.
>>
Anonymous
02/05/26(Thu)21:42:23 No.108069941 >>
Anonymous
02/05/26(Thu)21:47:01 No.108069976 >>
Anonymous
02/05/26(Thu)21:55:32 No.108070028 >>108069941
Badass
>>
Anonymous
02/05/26(Thu)22:03:13 No.108070076 >>108068089
>>108068036
His general
His UI
His model
His rules
:^)
>>
Anonymous
02/05/26(Thu)22:06:02 No.108070095 >>108070076
Well yeah, there's a lot of people who only care about Anima genning I guess.
>>
Anonymous
02/05/26(Thu)22:07:44 No.108070105 >>108070095
woops meant anime kek
>>
Anonymous
02/05/26(Thu)22:08:46 No.108070119 >>108068036
There's an/ldg/ Discord?
>>
Anonymous
02/05/26(Thu)22:10:07 No.108070131 >>108070119
Yeah, and Comfy is the mod :^)
What a time to be alive.
>>
Anonymous
02/05/26(Thu)22:14:00 No.108070155 Proof?
>>
Anonymous
02/05/26(Thu)22:15:39 No.108070173 >>108070076
If you're in his discord can you ask him to hurry the fuck up and get tdrussell to release the next intermediate checkpoint?
>>
Anonymous
02/05/26(Thu)22:17:07 No.108070186 >>108069938
>git pull
>dindu nuffin, i was a good boy
>>
Anonymous
02/05/26(Thu)22:17:35 No.108070189 >>108070173
he already confirmed he's funding the apache2 anima. it's over circlestone sisters
>>
Anonymous
02/05/26(Thu)22:20:10 No.108070209 >>108070173
Still don’t get why Lodestone gets invited to projects when his only résumé is being a rich furry funding failed models.
>>
Anonymous
02/05/26(Thu)22:22:24 No.108070223 >>108070186
Who said I did pull you fucking moron.
>>
Anonymous
02/05/26(Thu)22:23:06 No.108070231 >>
Anonymous
02/05/26(Thu)22:23:59 No.108070239 >>108070223
>he didn't pull
Lmao
>>
Anonymous
02/05/26(Thu)22:28:14 No.108070279 >>108070209
People who think lode of shit stones will see a cent of that money are delusional.
>You gave one million dollars in grant money for a furry porn model?
Even if I liked the guy, the entire idea of giving him grant money to throw furry porn into the money pit is ludicrous.
>>
Anonymous
02/05/26(Thu)22:29:20 No.108070287 >>108070209
>rich furry
he is not rich
>>
Anonymous
02/05/26(Thu)22:29:34 No.108070289 anima preview can't render the colors or geometry on eva unit 01 correctly. doesn't really get @shinkawa_youji either. hopefully the fully baked version improves on this
>>
Anonymous
02/05/26(Thu)22:30:43 No.108070299 >>108070289
>fully baked version improves on this
>>108070189
>>
Anonymous
02/05/26(Thu)22:31:16 No.108070310 >>
Anonymous
02/05/26(Thu)22:31:48 No.108070317 >>108070279
i agree, but someone bought him 4x 6000 pros and the 4x 4090 rig he had recently (and into which he placed the 6000 pros), and all his other 'experiments' so he's getting money from someone, and it aint the rich furry who paid for chroma-1
>>
Anonymous
02/05/26(Thu)22:32:03 No.108070321 >>108070209
lodestone has interesting ideas but so far the anima guy is probably the only competent baker weve had
i bet ani really underestimates what it takes so im excited for another ponyv7
>>
Anonymous
02/05/26(Thu)22:32:44 No.108070330 >>108070209
based ani
>>
Anonymous
02/05/26(Thu)22:34:53 No.108070349 >>108070209
ani is fulfilling his promise
>>
Anonymous
02/05/26(Thu)22:35:02 No.108070351 >>108070314
Please more.
>>
Anonymous
02/05/26(Thu)22:35:20 No.108070354 >>108070209
why is julien such an attention hungry faggot all the time
>>
Anonymous
02/05/26(Thu)22:36:30 No.108070363 >trust the words of the guy whos never delivered
kek you will reply with cope
>>
Anonymous
02/05/26(Thu)22:39:20 No.108070386 >>108070131
Link?
>>
Anonymous
02/05/26(Thu)22:39:31 No.108070388 >>108070314
bro, you can't just post nudes of people on the internet. that shit's illegal.
>>
Anonymous
02/05/26(Thu)22:41:53 No.108070408 Ohhh nonono, anime sisters Comfy will not bankroll another animu model.
>>
Anonymous
02/05/26(Thu)22:42:02 No.108070410 >>108070209
>trying to prop yourself up while shitting on another dev
holy fucking shit what a faggot, hope he gets raped irl
>>
Anonymous
02/05/26(Thu)22:43:00 No.108070421 >>108070209
Fluffyrock was a very widely used SD 1.5 model both by itself and in merges cause it was trained up to 1088x1088 buckets. Also Chroma is overall a very good model if you weren't expecting baselessly for it to be some kind of le ebin realistic or le ebin anime model.
>>
Anonymous
02/05/26(Thu)22:43:01 No.108070422 this is from the gay isekai
>>
Anonymous
02/05/26(Thu)22:43:03 No.108070423 >>108070408
DRAGGED AND SHOT!!!!
>>
Anonymous
02/05/26(Thu)22:44:36 No.108070434 >catjak and comfy double team trani
ran won btw
>>
Anonymous
02/05/26(Thu)22:44:57 No.108070440 >>108070410
Who is this guy even? Is it even true that McMonkey is working with him?
>>
Anonymous
02/05/26(Thu)22:46:25 No.108070452 >we want to spend tens of thousands of dollars to train the exact same anime model
>"for what purpose? how is it different?"
>it will have apache 2 license so SaaS platforms can print money using the model and not pay us a single penny
>"excellent justification let's go for it"
why
>>
Anonymous
02/05/26(Thu)22:46:38 No.108070456 >>108070209
>failed models
Chroma was far from failed.
>>
Anonymous
02/05/26(Thu)22:46:40 No.108070457 >>108070440
>Who is this guy even?
A serial liar who shouldn't be taken at his word
>>
Anonymous
02/05/26(Thu)22:46:53 No.108070460 >>108070440
if he did then why isn't the desktop app just swarm?
>>
Anonymous
02/05/26(Thu)22:47:12 No.108070464 when did ebonics become the official language here in /ldg/?
>>
Anonymous
02/05/26(Thu)22:47:55 No.108070470 >>108070452
there is a lot of things that could have been better.
>>
Anonymous
02/05/26(Thu)22:48:06 No.108070473 >>108070440
read the last link in OP newfren
>>
Anonymous
02/05/26(Thu)22:49:51 No.108070482 >>108070460
No i meant McMonkey and the AninAnon guy
>>
Anonymous
02/05/26(Thu)22:50:27 No.108070486 >>
Anonymous
02/05/26(Thu)22:51:06 No.108070488 >>108070456
The only legit failbake recently was Pony V7, which seems to have been captioned in a weirdly rigid way that makes it almost impossible to use
>>
Anonymous
02/05/26(Thu)22:51:20 No.108070490 just a bit of banter :)
>>
Anonymous
02/05/26(Thu)22:56:10 No.108070519 >>108070470
like what
>>
Anonymous
02/05/26(Thu)22:56:38 No.108070521 >>108070434
>ran won btw
won the "most schizophrenic" award from the mods
https://desuarchive.org/_/search/boards/g.desu.meta/tnum/108058266/deleted/deleted/
>>
Anonymous
02/05/26(Thu)22:57:07 No.108070526 >>108070408
Didn't know catjak was real, why catjak and animanon hate each other?
>>
Anonymous
02/05/26(Thu)22:57:54 No.108070529 >>108070521
you're such a lolcow you don't even know how to use the archives properly lmao
>>
Anonymous
02/05/26(Thu)23:02:48 No.108070556 Can a publicly funded model even be nsfw? This will end up with a 1M safetycucked model.
>>
Anonymous
02/05/26(Thu)23:06:18 No.108070585 >>108070556
It will 100% be a talking point when it’s released and people use it for porn. But throwing that money who’s explicit goal is to make niche porn is inviting trouble
>>
Anonymous
02/05/26(Thu)23:06:33 No.108070587 >>108070519
the vae, the encoder, the dataset
>>
Anonymous
02/05/26(Thu)23:08:20 No.108070600 >>108070526
I don't think ani cares about catjak at all. it's just catjak schizing out 24/7
>>
Anonymous
02/05/26(Thu)23:08:38 No.108070604 >>108070585
>explicit goal is to make niche porn
wut
>>
Anonymous
02/05/26(Thu)23:08:59 No.108070605 >discord leaks
>the schizo slows down his ruining of the threads moments after he discover comfyorg is funding bakers
>gets caught asking for funding
i am noooticing, where is the comfypasta ani? is it on hold until you get denied again?
>>
Anonymous
02/05/26(Thu)23:09:26 No.108070608 oh one of those episodes again
how lame
>>
Anonymous
02/05/26(Thu)23:09:53 No.108070609 COMFY GIVE THE MONEY TO NOOB TEAM,
>>
Anonymous
02/05/26(Thu)23:10:40 No.108070611 >>108070587
i think ani just wants to piggyback off the model and slap that its apache without improving or changing any of that
but i know for certain hes gonna fuck up the dataset
>>
Anonymous
02/05/26(Thu)23:12:17 No.108070622 >>108070587
>vae
Qwen-Image VAE is near-lossless when it comes to human perception. It is a massive step up from SDXL. Flux2 VAE is only very marginal improvement over that.
>encoder
Maybe. But even Qwen3 0.6b is giving huge improvement in natural language understanding over SDXL. I see the value in making all parts of the model as small as possible.
>dataset
We know hardly anything about the dataset other than it's booru plus a bit of non-booru. Unless you are complaining that it doesn't include e621 furry diaper shit porn, in which case I say thank god it doesn't
>>
Anonymous
02/05/26(Thu)23:13:17 No.108070633 >>108070611
no I think he get's it. this is the perfect vibe check for comfyorg and what they care about more: grifting or FOSS
>>
Anonymous
02/05/26(Thu)23:14:09 No.108070639 >julien
>>
Anonymous
02/05/26(Thu)23:16:10 No.108070653 Give me a single reason I should put trust in Ani
>>
Anonymous
02/05/26(Thu)23:16:14 No.108070654 They should train it as an edit model too. I've found out that complex edit tasks make it easier for the model to generalize despite the 2x cost per pair.
>>
Anonymous
02/05/26(Thu)23:17:39 No.108070666 >>108070209
>tdrussell has no base model experience but you funded him so i don't think competency has anything to do with it
what a disgusting snake, which is fitting since all he sells is snake oil, similar to those furry bakers that keep scamming the local community.
>>
Anonymous
02/05/26(Thu)23:17:54 No.108070668 Anima won't last long-term, max one year before it's outdated, is only slightly better than Neta Lumina, that's it.
And remember, the full Anima release will be disappointing just like Z base and everything else /ldg/ gets hyped about.
>>
Anonymous
02/05/26(Thu)23:19:18 No.108070685 >>108070668
scammer-kun the common base models that get released aren't being trained on boorus
>>
Anonymous
02/05/26(Thu)23:24:05 No.108070720 >>108070685
Modern local models can't be finetuned easily, they're more restrictive and something always breaks when you try anyway. There hasn't been a single good finetune after SDXL. All the Flux, Qwen, and whatever other finetunes have been completely broken.
>>
Anonymous
02/05/26(Thu)23:28:16 No.108070743 Has any LoRa trainer supports Anima yet?
>>
Anonymous
02/05/26(Thu)23:28:49 No.108070748 >saar
>>
Anonymous
02/05/26(Thu)23:29:16 No.108070754 >>108070666
>i don't think competency has anything to do with it
lol, lmao even
clearly anima dev has some level of competence given that the model is actually pretty good
there were no less than four (4) attempts at post-SDXL anime models and none are even close to Anima
>>
Anonymous
02/05/26(Thu)23:31:47 No.108070774 whats the baseline lora strength you give for wan2.2 i2v loras that don't give a reccommendation?
>>
Anonymous
02/05/26(Thu)23:38:23 No.108070833 ani might actually leave this place for a bit since his attempts at shilling himself here only backfire. so the old bakers might need to make a comeback
>>108070754
now that the scammers know comfy is funding they smell blood on the water, and they will do anything to get a slice of the pie.
>>
Anonymous
02/05/26(Thu)23:39:34 No.108070843 >>108070754
Because none of them were funded with $1M USD, dummy , only Comfy’s high school friend’s model was ;)
>>
Anonymous
02/05/26(Thu)23:49:15 No.108070924 >>108070668
why would it be anything like Z Base? Release Anima will be more trained, at higher resolution.
>>
Anonymous
02/05/26(Thu)23:51:57 No.108070937 due to his rampart botting and trolling we lost many real anons and 2 bakers over the course of the last 6 months, so if he goes away (please god) the thread might finally heal.
>>
Anonymous
02/05/26(Thu)23:52:58 No.108070951 >>108070937
rampant*
>>
Anonymous
02/05/26(Thu)23:53:49 No.108070955 Fuck off with the score_9 tags already wtf was he thinking
>>
Anonymous
02/05/26(Thu)23:54:38 No.108070960 catjak is having a meltdown on the gaycord. imagine being mad at someone for working on a better anima model kek
>>
Anonymous
02/05/26(Thu)23:55:40 No.108070967 >>108070960
screenshots or it didn't happen
>>
Anonymous
02/05/26(Thu)23:55:47 No.108070968 >>108070937
the problem is you
>>
Anonymous
02/05/26(Thu)23:57:32 No.108070981 >>108070937
>and 2 bakers
nah theyre just sleeping rn
>>
Anonymous
02/06/26(Fri)00:03:14 No.108071031 >>108070720
About Qwen, Wan, etc., it's because of the size; you need more VRAM for a large-scale finetune, so nobody has done it yet. Flux is distilled, so Chroma takes a lot of time, and after the Pony flop, people are avoiding training immature models and waiting for the so-called "perfect" Zbase. But there is something wrong with its architecture or the weights. We have to wait until someone finally finds a fix. I mean, SDXL was broken too; the model was a flop with terrible anatomy. The model that saved us was Pony, and that was a miracle. Nobody could train a good finetune before it.
>>
Anonymous
02/06/26(Fri)00:05:10 No.108071048 >>
Anonymous
02/06/26(Fri)00:36:51 No.108071293 >>108070743
a fork https://github.com/bluvoll/diffusion-pipe
official https://github.com/tdrussell/diffusion-pipe
there's also a pr on sd-scripts for it
Reply to Thread #108066594