first for debo needs a job

>>109258982
CJ needs a life lol

>>109258982
this

>mfw Resource news

07/12/2026

>LingBot-Video-MoE-30B-A3B — GGUF (8GB-VRAM ready)
https://huggingface.co/realrebelai/LingBot-30B-3B_GGUF_ComfyUI

>Anima-ControlNet-VACE-Depth
https://huggingface.co/TaihoC/Anima-ControlNet-VACE-Depth

>huihui-ai/Huihui-ThinkingCap-Qwen3.6-27B-abliterated
https://huggingface.co/huihui-ai/Huihui-ThinkingCap-Qwen3.6-27B-abliterated

07/11/2026

>MobileWan: Closing the Quality Gap for Mobile Video Diffusion
https://qualcomm-ai-research.github.io/MobileWan

>Krea 2 Character Training
https://github.com/KONAKONA666/krea-2

07/10/2026

>LongE2V: Long-Horizon Event-based Video Reconstruction, Prediction, and Frame Interpolation with Video Diffusion Models
https://cdfan0627.github.io/LongE2V-page

>Enhancing In-context Panoramic Generation via Geometric-aware Pretraining
https://zry000.github.io/Canvas360

>SAM-MT: Real-Time Interactive Multi-Target Video Segmentation
https://henghuiding.com/SAM-MT

>OPSD-V: On-Policy Self-Distillation for Post-Training Few-Step Autoregressive Video Generators
https://meigen-ai.github.io/OPSD-V

>DeltaV: Thinking with Visual State Updates in Unified Large Multimodal Models
https://github.com/Pengjie-W/DeltaV

>Leveraging Color Naming for Image Enhancement
https://namedcurves.github.io

>ARDY: Autoregressive Diffusion with Hybrid Representation for Interactive Human Motion Generation
https://research.nvidia.com/labs/sil/projects/ardy

>ComfyUI Workflow Documenter
https://github.com/kasumiworks/comfyui-workflow-documenter

>INT4 (W4A4) Inference in ComfyUI - Krea2 Turbo INT4 Convrot
https://github.com/viralvfx/ComfyUI-INT4-Fast

07/09/2026

>Krea 2 Style Reference LoRA
https://huggingface.co/ostris/krea2_turbo_style_reference

>Infinite Worlds with Versatile Interactions
https://technology.robbyant.com/lingbot-world-v2

>HIVE: Understanding Post-Hallucination Reasoning in VLMs
https://github.com/hefengcs/HIVE

>Tree-of-Thoughts Reasoning for T2I In-Context Learning
https://github.com/Pandastep/ToT-T2I-ICL

File: 1782569415098282.png (2.61 MB, 1088x1920)

2.61 MB PNG

>>109258982
Update filters, tardblocking is that easy

>>109258985
who?

File: 8725973580246979353.jpg (2.51 MB, 2880x2808)

2.51 MB JPG

>tfw the "mfw" spam hits

>>109258156
>>109258171
loras and strengths? impressive.

File: filters are life.png (53 KB, 220x178)

53 KB PNG

>>109259011
>tfw I filtered his ass

>>109259005
how is krea in sdcpp? I want to get rid of pytorch and all this python shit

>>109259005
Thanks, **.

The collages have improved significantly since my last complaint, which was two threads ago. keep up the good work anons

>>109259005
For real. I haven't seen debo's gens or his research spam in a long long time.

>>109258995
thanks!

stop harassing debo

>pathetic samefagging

File: debo_pw_k2_00075_.png (2.75 MB, 2048x1101)

2.75 MB PNG

>>109259168
np. happy sunday :)

Couples Wellness Spa gen anon, what's your prompt?

>>109259075
I don't work for you. Maybe stay in your discord with your opinions.

>>109259210
who do you work for?

>says the guy who posts gaycord screenshots of his crush

enough homodrama

>>109243860
>>109243900
>>109243941
>>109244019
no way this is local

why are there so many AI gen general threads and why do mods allow it?

>>109259258
>why are there so many AI gen general threads
lots of people using AI for different purposes
>why do mods allow it?
ask them on IRC

can someone share the debo news filter

>>109259258
sdg is a containment general and adt a spite bake
next question

>>109259332
this is a catjack containment thread tho

Krea 2 is goated, had my doubts, thought it'd be a long time before switching off of ZIT, but krea 2 nvfp4 is definitively better.

>>109259389
>nvfp4
gross

>>109259395
deformed legs

>>109259395
check if she has the diaper butt.

>>109259400
no blackwell? lol

File: ComfyUI_temp_kjtpq_00039_.png (1.91 MB, 1024x1024)

1.91 MB PNG

so if I want to make a lora for use with a Krea 2 finetune, should i train the lora on that turbo finetune itself, or still just on the unmodified raw krea 2?

>>109259258
because they haven't created an /ai/ board, which is long overdue.

>>109259514
im having racist thoughts after opening this image

File: ComfyUI_01961_.jpg (1.14 MB, 2016x3600)

1.14 MB JPG

>>109259480
It's generally best practice to train loras on the unmodified krea 2 raw and then apply it to the turbo/turbo based finetune

>>109259488
everyone involved in the finance side of 4chan is an incompetent retard and can't afford to make new boards

seriously being unable to turn the traffic numbers into more than moots spare laptops in decades is so pathetic it's unreal

>>109259609
it's .org not .com you fucking jew

>>109259615
being so opposed to funding sources the website doesn't function is not some noble endeavour, it's mental retardation

File: krea-2-turbo-int8-convrot(...).jpg (2.33 MB, 2560x1610)

2.33 MB JPG

>>109259527
poke the snake until ketchup

>>109259075
Desu too fried

>>109259540
>tfw gf doesnt allow you to buy a siamese cat but as a favor she makes one with her feet

>>109259651
snakes dont have red blood

File: 5992a568-a49e-4b6d-9b6f-3(...).png (2.91 MB, 2006x1129)

2.91 MB PNG

>>109259640
would you pay for the 4chan battlepass if it funded /ai/

>>109259702
source?

>>109259702
source? workflow?

File: attempted-a-soyjak-of-the(...).jpg (56 KB, 640x577)

56 KB JPG

>>109259395
>>109259514
>>109259651
>>109259395
No way NO fucking way NO FUCKING WAY.
hat's an ACTUAL Hollywood celebrity?! A REAL one?! THANK YOU THANK YOU THANK YOU
THANK YOU COMFY, THANK YOU LOCAL MODELS, THANK YOU KREA, THANK YOU GOD for blessing us with the sacred right to generate celebrities.
PLEASE I AM BEGGING ON MY KNEES, DO RIHANA FEET NEXT

>>109259651
What is your upscaling process for these images? I'm doing a simple t2i with krea2 but sometimes it seems like getting worse in many ways.

File: 1769432903082786.png (1.18 MB, 720x1400)

1.18 MB PNG

>>109258966

Running ComfyUI locally here:.
So what's the big difference between using an edit model like Qwen-edit and an inpainting-capable model? I was interested in playing around with edit models for the heck of it after relearning how to inpaint using Anima but even fp8 Qwen edit models are like 20 GB+. Even when I asked Kimi (yes go ahead and laugh) for recommendations it kept insisting I download the bf16 version that would have been like 40 GB. My rig is more than Pitbull running it but I'm questioning whether or not it's necessary for me to download all of that and potentially waste 20 to 40 GB of storage? Is Qwen edit really that good? What I need a model like that to do things like pic rel or to do more advanced things like redesigning my bedroom or something? (My parents used chat GPT's image added capabilities in order to help them redesign their kitchen and backyard patio so I'm wondering if something like this would

A) be possible to do locally
and B) whether or not you need fatass models like these

https://huggingface.co/Comfy-Org/Qwen-Image-Edit_ComfyUI/tree/main/split_files/diffusion_models

>>109259760
>Steven Trooniverse
just for that I won't help you

>>109259760
Barely skimmed through your yap but inpaint models are obsolete. Just instruct the edit model of your choice (Qwen edit, or Klein 9B) to change a specific thing and not change anything else, most of the time it will follow.

>>109259746
you're so cool and counter culture anon

File: Screenshot from 2026-07-1(...).png (5 KB, 1089x35)

5 KB PNG

Is this a good tLoRA training speed for Radeon (7900XT)?

>>109259773
>or Klein 9B)
How would you use it for that? I have both the 4B and 9B model downloaded on my machine and last time I checked there were standalone diffusion models but no dedicated edit models.

File: krea-2-turbo-int8-convrot(...).jpg (2.7 MB, 1800x2560)

2.7 MB JPG

>>109259759
No upscale, loras

>>109259773
Also wouldn't that result in it just regenerating the entire thing? Or can you have it only edit a specific part of the image? It's my understanding that even edits like >>109259760 essentially regenerate the entire image but the differences beyond the thing you told it to edit are so tiny they're basically imperceptible to the naked eye.

>>109259800
looks better than in real life.

>>109259792
Reference latents, they are trained to use them.
Comfy has example workflow.
>>109259801
It regenerates the entire thing but the degradation is low thanks to top notch vae.
If you still want to preserve the original image you can use resizing and masks to overlay only the edited part on the original image.

File: 2026-07-12-173702_524x422(...).png (79 KB, 524x422)

79 KB PNG

>>109259055

File: krea-2-turbo-int8-convrot(...).jpg (2.62 MB, 1800x2560)

2.62 MB JPG

>>109259813
Yeah probably

Celeb worshippers are a fucking disgrace to /ldg/

>>109259820
I should add it's (at least Klein) is prone to color shifting the image slightly so it can make this kind of workflow difficult sometimes.

So can I make my own porn videos with a 5080 yet or am I gonna have to wait until 6080?

>>109259801
You need to inpaint to prevent it from regenerating the entire thing.

>>109259843
the obvious answer is the 6900 XTX

File: 00016-3096293100.jpg (467 KB, 2816x1664)

467 KB JPG

>>109259480
stick to using the raw mode anon for training loras. everyone that has tried to use the turbo model has gotten burnt hard with bad results.

>>109259843
I can make them on my 3060 so if you can't on a 5080 that would be a skill issue.

>>109259836
god bless AI

>>109259843
>6080
its most likely going to be 16gb so i wouldnt get too excited

are img2img models capable of making adult female breasts smaller? am I a promptlet? it made her boobs bigger without much effort, but going the other way it didnt work at all.

File: krea2 lora training questions.png (1.16 MB, 888x1184)

1.16 MB PNG

How is Krea2 lora training? What's the consensus now? I assume we use raw for training? I am also hoping that it will perform well under 12gb with weights quantized to 8bit. I am planning to make a style lora. Does anything differ for Klein than usual style lora training stuff?

File: kuruminha___full_body_by_(...).png (275 KB, 627x1275)

275 KB PNG

>>109259800
can you be base enough to make a Kuruminha lora for krea2 anon :)

>>109259870
The Australians taught us adult females have large breasts and anything less is pedophilia

>>109259884
*Does anything differ for Krea2

>>109259889
I thought that might be the case when I doubled down and asked for flat the image came out corrupted... shame

File: krea-2-turbo-int8-convrot(...).jpg (2.65 MB, 1800x2560)

2.65 MB JPG

>>109259884
>How is Krea2 lora training?
It's great, no optimal settings yet afaik

>>109259885
You must provide dataset + captions

>>109259884
>12gb

Use Musubi Tuner with offloading of all blocks.

>Does anything differ for Klein than usual style lora training stuff?

Klein doesn't pick up likeness well.

>>109259863
>I can make them on my 3060
Prove it.

>>109259884
These training settings work pretty well:

raw, lokr 4 (1.5 gb loras if that's too much try 8), automagic3, 1024 only, weighted, default lr

Identity neutral captioning of the background, pose, clothing, hairstyle, expression..

>>109259959
>*accidentally shoots 3 innocent people*

>>109259890
You can use basically the same settings you use for other models except that time sampling has to be shift.

File: ComfyUI_Flux2Klein9b__00241_.png (3.49 MB, 1184x1776)

3.49 MB PNG

>>109259967
She is a trained professional.

>>109259984
how is she going to know who to shoot if they all have the flux chin?

>>109259996
easy, just shoot the darkies

File: ComfyUI_Flux2Klein9b__00243_.png (3.54 MB, 1184x1776)

3.54 MB PNG

>>109259996
they glow in the dark

>>109259884
loras seem to fix the humerus issue. But not the ridiculous fake bokeh lol

File: 1775227879982753.jpg (48 KB, 513x802)

48 KB JPG

Why do my krea2 pictures have this weird pattern sometimes?

>>109260032
You must be kidding right??

>>109260041
try negpip with (hairy man:-1.0) in the prompt

>>109260032
you can tell the difference between real and fake bokeh? damn, you must be a professional photographer or something

>>109260050
I think he is a professional comedian or something

File: Artimafacts.png (163 KB, 477x350)

163 KB PNG

>>109260041
Funny coincidence, just genned image
I'm using filterByPass2 lora with weight of 3 and this is the first time I'm seeing these
But it's because of the lora

>>109260041
replace the bypass with a nsfw lora

File: 68879.png (2.49 MB, 1296x1760)

2.49 MB PNG

>>109260047
nu uh

>>109260067
https://old.reddit.com/r/StableDiffusion/comments/1ukh334/i_extracted_the_values_of_krea_2_safery_filters/

>>109259960
Never trained lokr, I stick to bog standard lora.
I am also not sure what you mean by "Identity neutral captioning" here.
>>109259982
What values do you recommend? I already use minor shift like 1.3 for training typically.

>>109260067
I use Realism Engine and with the bypass off lots of things stop going through

>>109260091
who is this?

>>109260087
standard lora works fine as well. by identity neutral captioning i mean no stuff like eye color, body type etc.

>>109260091
why fat

File: HMk8UvJXYAASJUC.jpg (612 KB, 3018x2155)

612 KB JPG

>tfw want to do the lora training for lily

>finished collected nice dataset for in a her default outfit and multiple alternative outfits.

>tfw reminded of the brutal autistic captioning that comes with her complicated outfit design:'(

i hate the effort maxing that has to go into this shit but i don't want jeet slop results. wish there was a way i could do this on my phone while I'm at work.

>>109260113
but since you're training a style you can just ignore that.

>>109260047
how do you add a negative prompt to krea2?

>>109260113
>by identity neutral captioning i mean no stuff like eye color, body type etc.
do trainers not give you the option to drop parts of the prompt?
having some captions mention those details helps changing them later

>>109260125
isn't that easy after the first couple of reference images you pretty much have 90% of the captions ready for the other images

File: 1767690233773018.jpg (334 KB, 1296x1616)

334 KB JPG

wildcard prompting is the most lame shit ever

>>109260139
https://github.com/blue-pen5805/ComfyUI-krea2-negpip

>>109260150
and 1girl portrait spam isn't?

>>109260103
troll post

>>109260177
i don't follow zoomer topics i really don't know

>>109260187
how do you know it is zoomer related then

>>109260150
never tried it. what is it? some sort of predefined prompt variables or something?

>>109260204
I feel fortunate the most dedicated idiots that have issues with me are alcoholic low functioning losers that have been exposed a lowcows that can't even match the average /ldg/ poster in skill
They also like to ERP pretending to be girls with other men and can't even read filenames to realize who's posting what.

>>109260221
Is it physically possible for you to stop obsessing over a man? Very tsundere faggot behavior

>>109260212
the self-hating projecting zoomer will say "who is this" to this one as well

File: HLD6hIYXsAEBdkf.jpg (386 KB, 1440x2560)

386 KB JPG

>>109260141
it starts off easy captioning the facial details, skin details, body features, body shape afterwards i can copy and paste them. But the pose, facial expression, lightning, mood, camera angle, background, aspects of the outfits appearance from viewing angle of image always different from one image to another. i want the lora to be very versatile and not to rigid but the mimic the particular video game graphics style of stellar blade. Eve and Raven's designs are even more complicated and require more clever words to describe their default appearance and outfits.

>>109260247
to be fair both look off

>>109260267
yeah, anima is deprecated.

>>109260261
Maybe Gemma 4 could be useful.

File: 1762827895059621.jpg (376 KB, 1256x1672)

376 KB JPG

>>109260071
Howdy

ive got some kinos on standby, but im not going to post them in front of these bozos. i'll wait until more sane people are online instead

File: krea-2-turbo-int8-convrot(...).jpg (2.64 MB, 1800x2560)

2.64 MB JPG

>>109260247
It's fine I'm still standing

>>109260208
Shit like
1girl, big breasts, {blond |red |blue }hair, standing
Your prompt randomly changes based on predefined set > more diverse output for your coom

>>109260276
Anima is actually a decent model, all it needs is better anatomy, hands, faces, prompt adherence, composition, lighting, textures, proportions, perspective, text rendering, pose understanding, clothing details, hair, eyes, fingers, backgrounds, spatial reasoning, color accuracy, shadows, reflections, depth, fine details, style consistency, and overall image coherence

>>109260316
I need to try it. Laziness is a virtue.

>>109260328
goes to show how bad Illustrious was that Anima surpasses it despite all that

>>109260345
kek

>>109260328
>>109260345
>>109260354
too bad krea exists. anima isn't needed anymore

>>109260277
i've used gemma 4 26B-A4b q5 and q6 multiple times locally via koblodcpp but its has its annoying limits and starts to break down after 4-5 image captioning requests. It's useful and stable for the photorealistic people than for anime or 3d render characters. Had a better time using 3.1 flash on google ai studio but even that one begins to break later when context window session exceeds 40,000 tokens.

>>109260345
>>109260354
hey I wanted to tell you how much better the thread when Ani was around. he was right about everything :)

File: 35 stars status?.png (140 KB, 1150x312)

140 KB PNG

>>109260375

>>109260380
why would the context window increase at all for that task?

>>109260380
your supposed to start a fresh session per image, ideall a script that does it automatically, the context shouldn't build up past the system prompt + the image tokens + output caption

>>109260380
are you using multiple images in the same instance? if so then you're making a mistake

>>109260400
I want to AIDS free!

>>109260380
Are you retarded? Why do you keep captioning different images in the same context? You are supposed to make a new request for each image. You are degrading the caption quality for the latter images for no reason.

>>109260333
for wildcards to work they have to directly be inserted into the text encoder because if you feed it "wildcarded" text from outside it'll take everything, so you gotta use some custom nodes like https://github.com/ltdrdata/ComfyUI-extension-tutorials/blob/Main/ComfyUI-Impact-Pack/tutorial/ImpactWildcard.md and connect that to the text encoder

feed this to a cloud model and it'll give you a decent baseline to work with

gen me massive wildcard prompt, focus on quantity (so 1-3 words for each wildcard) {red|blue|blue yellow} 500 words, separate each. 0) locations prefix - (i.e royal, high class, urban, normal, pristine) 1) locations - indoor 2) locations - outdoor (separate indoor and outdoor locations by using a wildcard within a wildcard) 3) country name (just 1 word each, avoid poor ones, can also do specific NA states but the more known ones 4) time of day (i.e 8pm rainy , 3pm sunny, 11am overcast, 9pm golden hour) 5) camera angle but avoid closeup portraits or just portraits in general so it doesnt zoom in 6) camera styles (i.e iphone snapshot, 2000s sitcom style, handheld cam, digital camera}0) 

https://files.catbox.moe/lremgc.jpg

>>109260423
I'm not touching Impact Pack but I'm sure I'll find something else. Thanks.

>>109260427
Now that's something I didn't want to see.

>>109260386
knew you couldn't resist calling me ani. you are too fucked in the head because if this thread lol

anyone got tips on how to make a krea 2 model "zoom" out so its not always portrait? fuck i just wanna see some femdom with heels on ffs

>>109260467
mention the heels, mention more things that force it to show more of the surrounding
get negpip and put close-up in the negative

File: hello ani.jpg (24 KB, 796x110)

24 KB JPG

>>109260465
that's because you are ani

File: krea-2-turbo-int8-convrot(...).jpg (2.82 MB, 2560x1800)

2.82 MB JPG

>>109260467
Prompt for location first, then poses etc, then details

File: ComfyUI_02408_.jpg (885 KB, 2024x2696)

885 KB JPG

>>109260467
Give the heels more description. Krea tends to prioritize what you describe the most

>>109260472
An anon post from January is your proof? Anon is right, you are mentally ill

>>109260476
John Lennon?

>>109260467
just prompting the bodyparts you want to see usually works.

>>109260478
yeah see its all shit like this. cuts off at hips. maybe i wanna look at elsa's feet

redoing the color song. I discovered errors in some colors...

File: least obvious ani samefag.png (686 KB, 784x598)

686 KB PNG

>>109260479
>Anon is right

>>109260328
>>109260345
>doesn't have a good finetune with quality LoRAs
skill issue

>>109260514
not sure which side you're taking

>>109260458
https://files.catbox.moe/qph02e.jpg

>>109260500
At least post a gen faggot. Nobody wants your sperging with your dead meme folder

>>109260534
Thanks that made me forget what I saw earlier.

>>109260550
no, his gens are very bad

>>109260478
Do you mind if I get the prompt for this? Any loras used?

I see a lot of "realism" and "detail" krea2 loras but whether I include them or not the results don't really change

after fooling around wiith krea2 and testing these loras I'm still not sure about what is the best way to do things here
it seems like anything bit more complicated is breaking down and it ignores lots of stuff
just unreliable
sure it probably works well enough for a single subject

How tf do you describe to anima when a karate gi is only worn on one shoulder?
Am I such a promplet or does anima just suck

>>109260673
sampler issue. prompt issue.

>>109260697
?

>>109260550
>ani meltie
love to see it

int8convrot I got from who knows where:
>Great likeness, higher quality, shitty hardcore nsfw, doesn't work on native load diffusion, slower.

int4convrot shitty merges:
>Worse likeness, worse quality, better nsfw hardcore, faster, works on native load diffusion.

Why do you do this to me?

File: krea-2-raw-int8-convrot_00042_.jpg (2.51 MB, 1800x2560)

2.51 MB JPG

>>109260731
make your own?

>>109260731
Here is an idea, use the non-shitmix 8bit quant and use nsfw loras as needed.
>doesn't work on native load diffusion
?
If it's an old checkpoint you need the conversion script
https://github.com/BobJohnson24/ComfyUI-INT8-Fast/blob/main/convert_to_comfy.py

>>109260797
>use the non-shitmix 8bit quant and use nsfw loras as needed
I tried a bunch but they don't get the same results.
>If it's an old checkpoint you need the conversion script
That's what I'm using for it, it's slower though.

File: Screenshot from 2026-07-1(...).png (88 KB, 1171x1009)

88 KB PNG

new fag here, what does this mean?

>>109260832
it means your model is losing knowledge, STOP THE COUNT!!!

>>109260832
The loss curves are memes that you can ignore but lr graph is either bugged or you are doing something incomprehensible.

>>109260850
I'm sorry Mr. President, I can't stop it, it's too late!

>>109260832
>>109260860
Ignore what I said, had a brainfart there, everything is working fine
Ask an LLM to explain because likely no one here is going to give you a lecture over the theory.

>>109260832
all useless, the only useful info is validation loss, which is just a proxy for quality

File: Screenshot from 2026-07-1(...).png (102 KB, 1156x1092)

102 KB PNG

>>109260850
>>109260860
>>109260887
why my runs look so different then?

>>109260887
I never got validation loss to work properly (maybe skill issue on my part, I dunno). I use samples instead.

>>109260832
I tried to make sense of these but when the curves looked suboptimal while the samples kept improving I stopped paying attention to them.

>>109260893
because you changed the learning rate, change the seed and they will also look different, nothing there will tell you anything

File: Screenshot from 2026-07-1(...).png (445 KB, 425x503)

445 KB PNG

>>109260904
>>109260895
purple line is putting out this so at least I know this is trash

>>109260916
yeah thats the only thing you can see, if it exploded, but that you can see from the values in the CLI, no need for the graph

File: ComfyUI_02491_.jpg (947 KB, 2024x2696)

947 KB JPG

>>109260638
>A realistic photograph taken from inside a house, looking out through a closed sliding glass patio door. A beautiful young woman cosplaying as Elsa from Frozen stands outside on the patio. Her large breasts pressing against the glass. She has long platinum-white hair styled in a single thick braid. Her top is pulled up, fully exposing her large breasts which are firmly pressed flat against the outside of the glass door. Each breast is covered only by wide strips of black tape arranged in a cross shape over the nipples. She looks directly at the viewer inside with a playful, teasing smile

https://civarchive.com/models/2737295?modelVersionId=3120557

File: Krea2_turbo_00358__75.png (3.04 MB, 1254x1884)

3.04 MB PNG

>>109260930
Yes ty anon

>>109260933
i would sue for sexual assault

File: file.png (777 KB, 1280x720)

777 KB PNG

>>109260935
nadal is that you?

File: 1774920129370219.jpg (1.57 MB, 4505x1927)

1.57 MB JPG

Gonna order a vacation to hotel varmland with this one

File: test.jpg (467 KB, 1920x1080)

467 KB JPG

>>109260930

Thank god for anima aesthetic, it saved anima for anyone whos not trans.

b model also seems slightly better actually, more creative while being more stable, at least on my longer complex tag prompts, although i use all prompts on both

>>109260982
>humerus

>>109261005
>>109261058

>>109258452
Going back to literally any model before Krea 2 is like a bodybuilder trying to scratch his back with no backscratcher.

Also thank god for RES nodes, any kino wf you create and gen thousands of images for, you can queue for res2/5/6s and instantly get x3 more gens which will be quite different and obviously higher quality most of the time too.

Any other easy ways to generate more kino with slight wf changes?

Different resolutions/aspect ratios/general aesthetic loras, what else?

>>109261063
Krea 2 is like a bodybuilder - short humerus, can't scratch his back.

Krea 2's humerus tell is so funny, it's as bad as Fluxchin.

>>109261050
nobody cares. everyone here is using krea or id4

>>109261102
>id4

>>109261102
35 stars status?

>>109261102
>nobody cares. everyone here is using krea or id4
I don't.

I will never download ideogram, and I will never download krea 2.

I'm mostly using ace step 1.5 XL Base right now.

>>109260666
Slop

The trait of indians is
>everyone's doing it

lmao they're all women.

File: file (1).png (46 KB, 1240x443)

46 KB PNG

>>109261058
youre schizo lol

It's pretty funny CJ gets trolled this bad with the truth

>>109261126
Ask if the humerus looks normal in length.

>zitchuds using humerus cope

>>109261109
Not my cup of tea but anons use it regardless. Last thread had a couple of anons showing their bounding boxes

>>109261126
>forms his opinions from hallucinating AIs
>thinks he's not the schizo one
lmao

>>109261139
What's a bounding box?

File: 1765709209454415.jpg (3.09 MB, 3856x1288)

3.09 MB JPG

I will literally never ever forget this krea vs anima comparison I did the very first time I used Krea 2.

I mean I dont even have to say anything.

>>109261156
>I mean I dont even have to say anything.
yeah, it's obvious enough, Anima destroys Krea 2 here

>>109261102
id4 was for predditors who never touched regional prompting which was available for years and krea was just for predditors in general, people who want to type sonic and mario fistbumping in front of the white house and get an uncanny valley pokerface resemblance of those characters with the most slopped overcontraster style imaginable in 1:1 aspect ratio gen.

>>109261156
both look different but good enough for the prompt, anima is more unstable but krea is more slopped

Do you ever get flashbanged by how insanely good a gen is as you cycle through them and your brain registers it in the first moment? If not, you need to up your prompt game, anon.

>>109261156
Anima is for the same group of redditors who think slapping a film grain filter on an image makes it look real.
Badly drawn slop does not make it "soul"

File: it do be like that.png (267 KB, 586x522)

267 KB PNG

>>109261210

File: 1761475786775560.png (25 KB, 145x101)

25 KB PNG

>>109261210
see? kreatards can only make slopped throwaway memes while soulkino gens need some rough edges and creativity that you cant unslop krea to, it always overly centers the subjects and makes it look corporate with unnatural colors and film grain type of pattern they tried to force everywhere since they knew the model is bad at tiny details like skin in realism.

>>109261233
truth nuke

I asked Mythos 2.0 and it said Krea 2 sucks, and stop using it.

>>109261210
the white goes on the bottom of the cane lmao

File: debo_pw_k2_00082_.png (2.92 MB, 2048x1101)

2.92 MB PNG

>>109261244
>insults retro meme
>has CDs everywhere
choose a lane, buddy

>>109261264
Obviously he assumes the unc in the meme is saying whats in the bubble unironically, but hes retarded anyway and cant debooonk the arguments >>109261233

File: debo_pw_k2_00083_.png (2.67 MB, 2048x1101)

2.67 MB PNG

>>109261275
oh shit, i misread the meme. thats perfect actually
but now everyones gonna call me a retard :(

>>109261244
I mean, I won't say that ideogram and krea 2 don't create images.

I just never have saved any of their images, that's all I need to know.

:(

I actually don't hate star wars type retards, but God sent me to Earth to tell them how retarded they are, it's a bitter pill.

I'm worried about CJ

>nigbo

>>109261050
Really, is it good or is this a shitpost?
I'm too much of a non-autistic newfag to understand

>>109261319
It's a 4gb model, just try it with the prompts u usually use
https://civitai.com/models/2458426/anima

>>109261328
I have a marked preference for avoiding fatigue

>>109261348

> indian <

It's not any good...

Bros how do I stop being a promplet

File: 1763746022412910.png (179 KB, 541x480)

179 KB PNG

>>109261319
I'm sure you can get an *okay-ish* output after about 3000 gens and then you can cherry pick the good one and post it here and say "See anima can do realism." even though realism is the most useless form of AI art and really is only used by pedos, prn addicts, and weirdos who want to trick people

>>109261156
Krea 2 looks objectively better but is extremely slopped, Anima has soul but is objectively worse
grim

File: 1782507704022331.jpg (833 KB, 2712x1160)

833 KB JPG

>ruins your hobby

>>109261373
>objectively worse
but it's subjectively better and that's what matters

File: 1755974379977595.png (621 KB, 690x879)

621 KB PNG

>>109261366
incorrect sister
it is only MOSTLY pedos, porn addicts and weirdos
I like it because you can prompt edit it into making butiful painterly artstyles

File: Krea2_000167.jpg (1.01 MB, 2560x1440)

1.01 MB JPG

>>109261351
No idea what your dipshit comment means, but I'll just keep generating random images with this lora I found

>>109261419
It means it sucks.

>>109261394
those mtg players are missing 20 years and 150lbs

File: Krea2_000182.jpg (977 KB, 2560x1440)

977 KB JPG

>>109261458
I don't remember asking for your opinion

>>109261394
>they mostly have short humeri

no.

way

>>109261470
It's true though, the concepts you're doing are cool but the style itself is very generic
Could benefit from a more unique-looking mix

i'm having more fun making utilities with claude than 1girl gen'ing. think its about time I leveled up

>>109261519
We don't even have one gamer girl, sorry.

>>109260125
you are not using enough ai bro. use gemma 4 to caption it and prompt it with context about her outfit. make the ai code you up some python loops and just locally generate the txt files with captions and make your instructions very specific

>>109261394
>Implying MTG hasn't been irredeemably ruined by DEI troons and Fortnite wannabe crossover slop already
Funny joke anon!

>>109261339
me and who?

wheres the kino at?

>>109261366
>having a mini melti over realism in anima despite nobody mentioning realism
?

It is interesting though that someone whos low IQ and not into pushing the limits of the models towards actually generating realistic things (the actual thing thats hard) is also at the same time somehow against the best anime model. Do you just want to generate the globohomo quirky corpo art styles that any model can do or?

>>109261644
In my folder, waiting for a pixelspace edit model so that I can start making a VN with all of my kinoest 1girls instead of wasting that soul on posting them in this ocean of piss.

>>109261663
make a front side and perspective image of your 1girl and plug it into the kinoplexatorium

I just wish for an extract and run UI like Comfy portable, but with everything ready to gen.

File: 1759815878183955.jpg (971 KB, 4096x1664)

971 KB JPG

>>109261648
> Do you just want to generate blah blah blah blah blah i suck cocks

Don't test me - I made pic related with illustrious, you're no where near my level little guy.

Has anyone tried Prompton? The japanese site people use to sell or commission ai art? It's like the AI version of skeb. Doesn't seem like you even need to be Japanese to sell your stuff on there. You think it's worth it to make some money off this hobby?