Blessed thread of frenship

What are the recommended uncensor loras for krea2 questionmark

>>109402331
I like this one

https://civitai.red/models/2725430/krea-2-nsfw-v4?modelVersionId=3147117

>>109402341
nta but is this really any better at nsfw than realism engine? Also can it handle anything thats not (realistic:1.2)

>>109402331
All u need is the textfusion refusal lora for strictly censorship skirting

>>109402331
https://civitai.red/models/2728234/krea2filterbypass

for some reason all of my outputs using the same exact seed amd settings look slightly worse now

catjack is still spamming bakes today?

>>109402375
There is just one version of that right

>>109402341
>pubes asians and negroids
why even bother when that much shit has been thrown into the jeetbake

can you leave /adt/ alone catjack? we love ani and don't care about your obsession

>>109402453
lolcow

>>109402453
>we love ani
we all do. it's just not worth the nuclear meltdown catjack has when we get rid of the rentries. it's an open secret catjack has some sort of mod support for doing all this and that he is mentally ill. kinda sad this is the situation

rentry.org/ranfaggot

who is catjack? sorry i haven't been keeping up with the ldg lore

>>109402486
>sonic and ben10 diaper scat enthusiast
based

there is nothing of value to learn about any of the schizo lolcow namefags

>singular schizo nukes engagement of the entire board to advertise his mental illness episode

>female pov hand

>look guys im smoking do i fit in yet? XD

>>109400770
>meaning that you dont need to wait for the bf16 model to load again after changing a lora's strength?
the model stays loaded but I there used to be a pretty long delay on my workflow that was doing RAW+TURBO as a two pass. like it needed to load/unload the lora between runs. but now it's pretty much as fast as when I have enough VRAM headroom to cache everything.

>>109402331
probably just using the text refusal removal lora and do a bit of detailing after. Lustify seems good as well in the previews.

>>109402569
.>>109402611
lovely

>>109402583
>>109402624
xe/xer is getting tired

>tumblrspeak
any translators?

>>109402611
Hey it knows exhaled smoke

>>109402656
but not smouldering ember smoke

is mr. catjack here? i have a question

>>109402667
me too

File: 1785342200850880.png (1.83 MB, 1256x840)

1.83 MB PNG

>>109402221
Posted in the fake thread, very similar OPs.
Is tdrussell here? I have a question, why doesn't Anima know Shihouin Yoruichi, a character from 10 years ago? Also, look at her face and feet, this is the power of Qwen VAE?

File: 0932570864258024762.jpg (19 KB, 387x516)

19 KB JPG

>>109402569
yeah, hi. can i get a cup of hot wate and an ice sandwich. oh and let me get one of those mystery napkins.

>>109402767
Bigruss here. I think you are a schizo retard, catjack. You did more damage to the reputation of the model than anything else. Hope this helps

>>109402807
>You did more damage to the reputation of the model than anything else
you forgot bluvoll

>>109402802
It drives me nuts that local models still can't do piano keyboards. I haven't worked on it personally, maybe extra time at the right sigmas for the specific seed/gen?

>>109402583
>
lmao

>>109402569
can you make it so there's a mysterious package under the menu (ie drugs)

>Train lora overnight
>Excited to try it
>1000/4000 steps
>VRAM maxed, 3000s/it, stuck on sample
It's over

>>109402972
is there a need for that many steps

im training my first lora atm with 2000 steps

>>109402976
>is there a need for that many steps
That depends on factors like the model you're training, your dataset, the scope of your lora and many others. You won't find a one size fits all answer

>>109402976
models like qwen and now krea seem to benefit from a lot more steps, like 4-10k for style loras and it keeps notably improving over the step count

It will probably function after 1k

File: orod_00183_.jpg (630 KB, 2048x1528)

630 KB JPG

>>109403058
>now krea seem to benefit from a lot more steps
Yeah I think so too. Also regularization dataset seem to work better when split into smaller batches / epoch instead letting all roll with lower loss weight

File: z_krea_2026-07-29_17-32-4(...).jpg (1.39 MB, 1534x2154)

1.39 MB JPG

character lora, style lora, still following prompt adherence without overfitting and making all humans the character

anons whats the prompt for th expression of OliRod in the OP

>>109403110
>making a playful, exaggerated scowl directly at the camera. Her eyes are squinte tightly, causing light creases along the bridge of her nose. Her lips are pulled back into a grimace—exposing bared white teeth with her top lip lifted slightly higher on her left—giving the expression an asymmetrical, cheeky sneer.

>>109403110
rawr ;p

File: 00056-852492888.png (2.44 MB, 1152x1920)

2.44 MB PNG

didn't realize how bad some of the captions for the bouquet(blue dragon) krea2 lora were for the first attempt until i tested it. krea2 kept generating dragon horns, dragon tails and literal bouquet of flowers wrapped together. Had to rewrite the triggers and remove any mention of dragon and bouquet from all the caption txt files.
regular form: BouqBDgirl
invisible ghost form: BouqBDghostgirl
https://gofile.io/d/IjmyaK

>>109403058
5000 steps is the sweat consistent spot for krea2.

>>109403190
kinda reminds me of a dragon quest character for some reason

File: 00070-3271568156.png (2.04 MB, 1920x1152)

2.04 MB PNG

>>109403205
yes. the original character designers of Akira Toriyama for her and the blue dragon characters. That guy is the artists for dragon ball and dragon quest series.

>>109403097
Nice work car jack.

cozy breas

>>109403097
as everyone already knew, parameter count won. anima chose a 1b parameter tiny model and is dumber because of it, and it's still slower than krea 2!

can you fuck off from /adt/ schizo? thanks

>>109403385
>and it's still slower than krea 2!
Can you post some it/s comparisons between the two?

>>109403429
Verify this, /adt/ has at least 3 posts per hour, but now we have 50 posts in half an hour, that's strange.

File: krea 2 style pose (front).png (675 KB, 991x1059)

675 KB PNG

>>109403385
>dumber
false. Let's look at a krea (always meaning krea 2) style pose.

I welcome /adt/. They should delete their general and come here.

File: krea 2 style pose (back).png (758 KB, 991x1059)

758 KB PNG

>>109403466
and here's what is necessary to achieve that "pose" that krea 2 (korea korea) produces:

>>109403466
>>109403475
that's a good pose, though? very cute.

>>109403480
Most humans will never be able to assume that pose. It's a contortionist/photoshopped pose.

>>109403458
Ani literally revived the thread in two posts

Maybe zoomers are in the condition of thinking that's natural or good.

It's not.

Kind of like foot binding, certain cultural trappings lead to unnatural behavior that's at best delusional.

>>109403486
Skill issue. Stretch more I'm posing like that on my bed typing on my laptop right now.

>>109403466
>>109403475
what is this schizo retard bullshit? blud downloaded daz studio just to seethe

chat is this true >>>109399509

>>109403502
yes it's retarded. we need /gai/

>>109403486
so what exactly is the point you're trying to make here? we're not allowed to gen any poses that require flexibility, or else you start throwing a tantrum?

>>109403489
Yes by chimping out because his rentry was put in their OP keke not really organic posting desu

>>109403511
you have to gen exactly what he wants at all times or else he's gonna throw a melty

>>109403502
we are blessed compared to /h/ shitposting TBQH

>>109403486
You sound like a feminist crying about unrealistic body proportions in anime and video games. Cry moar.

>>109403385
I'll just assume then that you are lying since you cannot post any speed comparisons.

>>109403466
aren't you supposed to show a Krea 2 gen that actually has a human in that pose?

>>109403502
based on the order in which the generals are named one can deduce that the poster is incompetent

>>109402384
desu the fusion refusal lora is better than the filterbypass

>bro installed daz3d just to troll
classic case of the troll becoming the trolled

File: z_krea_2026-07-29_18-22-5(...).jpg (1.95 MB, 1536x2304)

1.95 MB JPG

>>109403097

>>109403499
???

what's your complaint? Everything I said is perfectly correct. krea sucks (meaning krea 2 sucks)

File: file.png (1.88 MB, 981x916)

1.88 MB PNG

>>109397919
I am redeeming this one will bring me many rupee

>>109403655
it pumps out gen after gen that's totally inaccurate. You would never mistake korea korea for a real photo:
>>109400560

look at that. What's that supposed to be? It's some kind of retarded abomination. The feet are glued on. that's not what humans look like.

>>109403848
that's not the pose you showed

>>109403785
i wouldn't post these if i were you

File: 00057-852492889.png (2.32 MB, 1152x1920)

2.32 MB PNG

>>109403765
i just uploaded the lora in civitai. check the triggers in the description.
https://civitai.red/models/2818497/bonquet-blue-dragon-krea-2-lora?modelVersionId=3179148

>>109403875
>someone posts something vaguely interesting
>told to stop posting
what causes this

>>109403892
it's just some dork who got raped by krea. take no heed

File: Screenshot from 2026-07-2(...).png (96 KB, 1476x997)

96 KB PNG

she nears

can one of you make a better version of this?
https://civitai.red/models/2787319/krea2-anime2real

>>109403916
(roughly 2 gens away)
but I'll have failed gens and maybe inpainting.

But I'm not inpainting too perfectly, just enough that the lyrics are basically understandable. Odd pronunciation is irrelevant.

>>109403922
krea is badman

how do you change camera position in Krea 2 turbo? the usual "top-down view", "view from above", etc do nothing

>>109403945
do those do anything in any model?

>>109403945
here's why:
>>109403475

>>109403968
obviously

>>109403879
hi debo :D

>>109403989
proof? with which model

>>109403945
good luck
krea is so cucked you might need a lora for that

>>109404005
using the bypass loras is pretty much mandatory with krea

File: Krea2_turbo_07679_.png (1.53 MB, 1024x1024)

1.53 MB PNG

>>109403994
you never used any Danbooru based model?
all cloud models since DALL-E 3 too
Krea likely isn't as versatile and probably will fail with more complex prompts but it can be done

>>109403945
start the prompt with crazy ridiculous shit like `drone shot from up above` then describe the rest and leave the subject for last.. it'll prob work

>>109402976
Most people are just parroting parameters from other people used since the SD1.5 days.

You can do as little as 400 steps at a 0.0003 learning rate.

>>109404062
Nice. Rarely do you see gens like this.

>>109404130
i agree 100% this is front page material

lol found the problem, when i describe any background "forest/mountain/wall etc. behind" in my prompts it simply ignores everything about the camera position and it always generates the background you specified in the prompt instead

>>109404062
the handle comes up to her chin.

>109404152
>im a nigbophile

>>109404152
please anon, she's self conscious about her height

>>109404163
hey can you handle >>109403922

File: anima.png (703 KB, 1878x940)

703 KB PNG

What am i doing wrong? All my results with anima checkpoints have artifacts and very low quality.
I followed the instructions and enabled Anima on the settings, plus i am using the suggested VAE and text encoder.
I am using forge classic/neo.
>Use comfyui instead
No. It keeps breaking extensions almost monthly because of stupid UI changes no one asked for.

>>109404238
>What am i doing wrong?
not using comfy
>Use comfyui instead
yeah, use comfy instead

>>109404148
yea so exploit this knowledge and describe ground when you want topdown and describe the sky/ceiling when you want views from bottom

>>109404238
>No. It keeps breaking extensions almost monthly because of stupid UI changes no one asked for.
never had a problem with this and ive been updating every other day for years, are you updating the extensions too?

>>109404238
does it do it without the edit lora enabled?

>>109404238
>Use comfyui instead
unironically your best bet is to switch to noodleui

>>109404270
>>109404251
ComfyUI is a massive security risk and too complex for normal people.

>>109404280
>ComfyUI is a massive security risk
Not really, but okay

>>109404280
>massive security risk
do you run it outside a container and open to the internet without auth or something?

>>109404280
posting on 4chan is a massive security risk for normal people

>>109404288
>a container
not secure, see OpenAI's latest drama

I'm instantly getting viciously attacked by these commercial shills.
>>109404288
>>109404290
>>109404287
Your post doesn't make any sense.

>>109404302
Your post doesn't make any sense.

people run comfyui using linux? giga kek

>>109404287
>he doesn't use free compute from exposed cumfart endpoints
https://www.shodan.io/search?query=comfyui

>>109404302
>commercial shills
comfyui is free and open source

>>109404318
Don't think you understand the situation and how everything works.

>>109404325
Explain then

>>109404314
I run a windows vm in linux that I run comfy in

>>109404297
yeah im sure some tranny malware coder will make a custom comfyui node that will spread around without detection to thousands of people none of which will use ai to audit the code for months before u also do the same and then the tranny will use a docker sandbox escape 0day in order to use linux/m$ privesc 0day in order to ransom all your 1girls back to you for hrt money.

>>109404335
I feel like you're not taking my 2girls collection seriously

>>109404335
stop telling people about my custom node installed on 2,672 machines.

>>109404331
the application must run like shit

>>109404318
that is the freemium option. it's actually a corporation that wraps apis and sells them back for you. that is what most of the employees do at comfyorg

it's hilarious hearing about all these linux trannies getting backdoored

>>109404368
>and sells them back for you
never paid a dime

>backdoored

>>109404368
--disable-api-nodes
in your path

>>109404374
They take ownership of your workflows. It's a part of their tos

File: ComfyUI_temp_bxord_00002__.png (2.73 MB, 1152x2016)

2.73 MB PNG

>>109403882
You know, if you are training a character lora you shouldn't caption stuff that is inherently from your character such as skin color, eye color, hair color, body shape, etc-

Only caption stuff you want to be able to test flexibility later, such as outfit, hairstyle, etc.

Otherwise your prompt gets bloated with stuff that the lora should already know about the character, also I don't know how big your dataset is but 5000 steps for a character lora seems overfit

A good character lora should be flexible and versatile

>>109404383
show outgoing network packets then

>>109404381
it's ok, I only had to get the Bitcoin miner on your computer

>not having your own custom fork of comfy with all the junk removed
smdh, yall niggas not even real

>>109404395
>OOMs your bitcoin miner because u didnt use --disable-dynamic-vram
kek, nothing personnel, kid.

File: ComfyUI_00314_.png (3.77 MB, 1536x2160)

3.77 MB PNG

Why did Hitler stop working for my krea2?

>>109404394
didn't we do that already? They phone Google, ultralyrics sends God knows what, the manager calls home and there was one other thing I might be missing

>>109404392
That poster ignores any and all advice from anon and then spergs out about it. He is very annoying.

>>109404409
They patched him out

>>109404396
you could topple an empire if you made the fork public

>>109404392
>You know, if you are training a character lora you shouldn't caption stuff that is inherently from your character such as skin color, eye color, hair color, body shape, etc-
but if you do you should enable tag/caption dropout for those descriptors

>>109404396
>>109404426
Contact info@kemonotech.ai with this asap.

>>109404444
no faggot

>>109404260
>are you updating the extensions too?
Extensions on comfy are not actively updated apart of a very few ones. Things like regional prompting stopped working on comfy half a year ago thanks to the unnecessary UI updates.
>>109404268
>does it do it without the edit lora enabled?
Yeah, even without the lora the result is the same.

>>109404462
>Extensions on comfy are not actively updated apart of a very few ones
so you dont actually update, found the problem

>>109404462
>unnecessary UI updates
what makes them unnecessary?

>>109404483
he doesn't like them

File: ComfyUI_temp_qhipp_00028_.png (3.31 MB, 1344x1728)

3.31 MB PNG

>>109404441
caption dropout doesn't work when if you cache the text embeddings, and since most users who train on Krea2 are vramlets, this is bad advice

>>109404481
>so you dont actually update, found the problem
There is nothing to update because the extensions are not actively updated by the developers of them.
Go on, try to do an image with regional prompting on the latest comfyui and see it's not possible.
>>109404483
>what makes them unnecessary?
They add nothing meaningful other than some pretty logo and moving things around.
A cute "comfyui" logo appearing does not justify half my extensions suddenly not working.

>>109404508
>A cute "comfyui" logo appearing does not justify half my extensions suddenly not working.
that wouldn't make the extensions stop working thobeit

>>109404508
>A cute "comfyui" logo appearing does not justify half my extensions suddenly not working.
you're right but that's not what's breaking anything.

File: 00111-4283581244.png (3.57 MB, 2688x1536)

3.57 MB PNG

>>109404238
i will never understand the appeal of this 2B parameter model or why Haoming02 found it so appealing and prioritized so much time adding features for it on forge neo.

>>109404523
its sota for tranime

>>109404530
illustrious is sota for anime

>>109404545
anima aesthetic 1.1/noob/illust share the top essentially but anima is overall the highest quality, although all should be used

>>109404523
Anima beats Illustrious in every metric

anima can only gen 1girls, which is a huge problem for me since i rely on creating unique art to make a living

nobody here wants to help

>>109404545
lol no, thats novelai

File: ComfyUI_00321_.jpg (605 KB, 1536x2160)

605 KB JPG

>try krea2 without any prompt

>>109404578
rat

>>109404573
where do I download that? is that a finetune of noobai?

>>109404587
just wait for the leak, it always leaks

training protip that makes a lot of sense but i didnt see it online and only learned it by experimenting:

if you have some kind of a shittier dataset with some hand/body horror, artifacts, detail problems and inconsistencies, and you cant fix them easily, train the concept/style lora at like 512 resolution only.

lower res will hide the problems while conserving most of the style, compositions, colors etc of the image, merging this knowledge into the model without destroying base model quality by not swaying it towards learning these small problems.

>>109404571
with?

>>109404622
see >>109403922

>>109404610
another protip: dont use synthetic datasets for lora training. as the saying goes, slop in slop out

>>109404603
i heard it leaks in 2 more weeks, stay patient localsisters

>>109404630
just use the krea image edit lora

>>109404631
You are pretty clever, anon.

>>109404578
I've been telling you that it's basically loras. the closeup stuff doesn't really belong with the rest of the image.

>>109404640
it can't do anime to real

>>109404631
Retarded take, if the synthetic data is an output of a Lora or something like that, it's fine.

>>109404631
sometimes i want the specific look of a specific lora but with a twist, or i generated thousands of images with a specific lora in an unorthodox way and want to train on only the best of those images, or i collected someone elses ai images online which are made with a private lora i want to recreate

>>109404630
there are many versions of this type of lora, try all of them or spend the hours you want to make the best dataset to train a new one

>civitai jeets telling you to use synthetic data so they can yoink your lora and reupload it for buzz

>>109404672
>xe cares about buzz

>>109404672
whos telling you to use synthetic data?

>>109404672
Have you actually tried retraining your loras with the "best of" outputs of a previous version of the lora? It does work like RL while you -can- make it not slopped as long as you don't add slop to the dataset. If you are saying every synthetic data is slop, then every lora output is slop too.

>>109404672
implying using only non-synthetic data stops them from doing that

>>109404650
it can do real to anime, so i don't see why it shouldn't also be able to do anime to real when paired with a realism lora

>>109404734
reminder:
>>109403475

>>109404700
base models trained on synthetic data like ernie can only produce slop.
lora training is no different, if anything, it just amplifies the slop even more

>>109404740
There's no spine bend in that pose. She's flat on her stomach

>>109404734
This is collagen material. Let me guess the prompt...
>1girl, bed

Were none of you here when the Z paper dropped? Chang already proved real data is the way, synth data is doodoo unless you have trouble recognizing images as ai generated I suppose.

>>109404740
A reminder of what? You whining like a feminist bitch on xitter?

>>109404755
in this case, her knees are 2 ft into the bed. maybe it's a water bed.

Psa: three discord faggots are the drama niggers
>>109404708

>>109404759
real data is the way for realism in a small model in the current day.
if you dont want realism then it doesnt matter.
if you live in 2030, then it wont matter either since the models will be smart enough to delineate between slop and kino while training.

File: ComfyUI_00095_.jpg (749 KB, 1664x2080)

749 KB JPG

>>109404757
Testing my Ideogram lora with some old prompts. Trying to make it better at amateur photos. Picrel is without lora

>>109404783
yeah we need some way for the model to know which images are good during training. maybe some kind of score system? then the user can just tell the model what score the output should be

>>109404775
who

>>109404795
It does work. It removes one layer of slop.

>>109404795
>Ideogram
uh oh
>>109404740

File: file.png (124 KB, 402x526)

124 KB PNG

>>109404828

>>109404708
Who was trAni replying to here?

>>109404844
Who literally cares

>>109404795
have you [validated] your pose using OpenPose?

>>109404860
the same retards that Ani is telling to touch grass

>>109404842
you point this out after posting an ugly ass blurry gen >>109404816
lmfao

>>109404828
jeep compass xl

>>109404808
This has been done before. There is an RL training pipeline I forgot the name that uses a reward model to "tell" the trainer mid-training that certain outputs are more desirable than others, so the best output is kept as one of the outputs to train in the next "round" and the other is discarded. But spoiler alert: the reward model was trained to give a higher score for slopped hyperaesthetic shit since this is what normies consider to be the pinnacle of aesthetics

>>109404775
They are bumping our dead general, let them be

>>109404900
This is why everything is so shit. Statistical average of average taste.

>>109404931
It's like the film grain is embedded in her skin

>>109404944
dermal grain is a serious condition, please don't make fun of her

Are small models dead?

>>109404931
is this with the two keksampler autism?

>>109404964
people still train for SD 1.5

File: ComfyUI_00099_.jpg (1.11 MB, 1536x2048)

1.11 MB JPG

Lora
>>109404817
It helps a lot making normal looking people. Backgrounds can get a bit messy though

File: ComfyUI_00100_.jpg (847 KB, 1536x2048)

847 KB JPG

>>109404998
No lora

>>109404964
No see Anima

>>109404970
yeah but every new model is a quadrillion b blob

>>109404998
Yeah it works, cool

>>109404920
We really need a reward model trained on feedback from people with an actual fucking taste. Like, make it always give a higher score for photos that actually resemble regular phone-taken photos and not "cinematic" shit with background blur, give higher scores for paintings that actually look painterly with paint strokes and not digital art slop etc
These models defaulted to please "product-oriented" people so this is the reason everything look so fake and plastic out of the box

>>109405014
yeah... because that's how the models get better...

>>109405032
isn't that what pony did but with pony furfags

>>109404998
that leg ain't right, when you turn your foot your calf has to turn too but it doesn't seem to be turned in that image

>>109405044
afaik for Pony they did not do RL, they just fine-tuned on most of e621 and booru

>>109405040
one could suck that thang all day long

>>109405032
we need to finall make finetunes of ok models and not waitfag for the next potentially better model

>>109405072
he had a discord where you rate images which is how the score_9 system came about or so I thought I don't know nor really care that much

>>109404931
I have a song coming out soon that will help you spot the signs of fas.

how are people able to use a vae from a different model? i thought the weights were specific to the original model

>>109404610
Body horror and inconsistency in a dataset won't necessarily result in a bad lora. I trained a lora from a manga with only bits and pieces of the character available at inconsistent resolutions, and it turned out fine.

>>109405055
Your foot has a bit of rotation and it looks like she's rolling her ankle outwards rather than turning her foot anyway.

File: 260729160308_routput-0001.jpg (697 KB, 1650x2464)

697 KB JPG

>>109402569
one raw potato please

we've come a long way from complaining about fingers and toe counts to now complaining about whether or not an ankle is rolled or if the anatomy of the underlying bone structure is correct. impressive.

>>109405069
Have you validated that pose with OpenPose? I believe you are guilty of numerous violations of proportion. as we all know, krea (always meaning krea 2) is abominable at the task of realistic human beings.

>>109405133
It's just one dude nitpicking every gen

>>109405133
yeah and the next model might be 0.1% better than the previous one so lets wait for that instead of finetuning that one promising model already

>>109405131
>his nails are glued on
ahhhh

>>109405149
nobody is stopping you from doing what you want
complaining that others aren't doing what you're also not doing is pointless and retarded.

>>109405120
>rolling her ankle outwards
fuck does that mean, show your real foot doing that right now

>>109405170
nta but I can do it but it's pretty uncomfortable. everyone here is somehow an expert on anatomy yet have never even drawn the human anatomy.

>page 8
looks like its about time for him to bump the other thread again

>>109405170
Using the rotation of your ankle and putting your weight on your instep and big toe and no I'm not posting a feet pic for you bro.
This is an extremely autistic debate over a minor anatomy issue

>>109405205
>minor anatomy issue
the leg immediately looks off, even broken

>>109405212
It looks like a slightly unorthodox but not impossible pose

>>109405163
it's just frustrating, SDXL showed what local is capable of, a meh model then starting its own ecology once some finetunes happened with interesting stuff and refinements constantly happening, and now we're restricted to sitting back and watching whatever model some bigwhig shits out, get excited for a month and then move on to the next new thing and forgetting about the previous thing
it didn't use to be like this
local isn't supposed to be like this

>>109405219
no, you're wrong

File: 4557845437894.png (18 KB, 900x806)

18 KB PNG

where is kino?

all my kino is currently NSFW

>>109405226
sdxl left a lot on the table, modern models not as much

>>109405202
>>109405297
>>109405126
What is the base model for this? Are these loras?

File: ComfyUI_00582_.mp4 (3.37 MB, 640x832)

3.37 MB MP4

graceful

File: krea2Clean_00170_.png (2.76 MB, 1448x1448)

2.76 MB PNG

krea is fucking nuts

>>109405398
thats a man

>>109405157
>
cmon babe let's see your humerus

krea is nuts 2
https://files.catbox.moe/pclicc.png

>>109405404
That's Princess Peach

>>109405006
>>109405069
>>109405297
Have you ever bothered asking your fellow anons in this general how they feel about your foot fetish gens?
Because I, anon, find them pretty gross and offensive.

>>109405415
yeah that's a super smash bro alright.

>>109405420
I approve of his feet gens. Sorry!

>>109405421
You're supposed to cosplay as your main when playing smash. It's normal

>>109405421
i laughed

>>109404392
Catbox please

>>109405297
anatomically inaccurate, as per the usual with krea (we always mean krea 2). krea just doesn't understand anatomy at all, and it shows with everything you've ever genned.

>>109405427
why?

they're malformed.

>>109405471
>>109405464
you're malformed and you don't hear my crying about it.

File: 53745.png (21 KB, 738x219)

21 KB PNG

should i run a TRIM command after i wipe all this out?

File: Krea2_turbo_01444_.png (3.96 MB, 1400x2096)

3.96 MB PNG

>lora trained
it's over

>>109405494
You should run
>oxipng.exe -o max -Z --zi 15 --alpha --preserve -r .
in your output folder to losslessly compress PNG files while preserving embedded workflows.

>.exe

>>109405549
the aryan method is to only keep the metadata so you can regenerate whatever you want to see again

>>109404734
did you prompt for redeye or is this some sort of old camera lora? i tried to get redeye a couple weeks ago and couldn't get it from prompting, it's pretty weak in the image but better than anything i managed to get

>>109405533
Please answer
>>109405372

>>109405579
Doesn't work properly since you would also need to keep and recreate exact comfyui ver, torch ver, gpu, gpu drivers, maybe even cpu depending on where seed RNG is calculated etc, and if one thing is off you will not get too close to the same image and even if you nail almost everything the image will have slight differences which will change some images enough, like images with human faces, to make it not the same face and thus worthless.

>>109405533
wow it sucks

what a shitty model krea is.

>>109405620
I am NTA but I'm 99% certain it's krea turbo

I think that anon also mentioned using a custom lora in the past

>>109405639
krea really is the worst model ever released publicly, and it's not even close.

what they did is they trained it to fool ai vision models, but vision models don't think in 3D, so it stitches together garbage.

But, because indians have never seen a woman in a bikini in 3D real world reality, it looks good enough to them.

>>109405659
I'm Indian and I prefer ideogram. You have no good excuse for being racist.

I've a woman in a bikini. In 3D!!!! (not 4D though)

>>109405659
Not Indian, and krea is currently the best model to gen with locally

>>109405696
It's a bobblehead pin the tail on the donkey model and it sucks and it's ugly. krea is never going on my drive krea 1 krea 2 krea 3 krea 1,000,000 I'm not puttingh that trash - or ideogram - on my drive. No.

>>109405718
your gens look like ass anyways, the model you can't even run won't help you

>>109405724
I'm genning ace step. The masterpiece is near complete.