Thread 106171902

Welcome to the Anime Diffusion Thread. This is a thread for discussion relating to leveraging generative AI technologies for anime-related content. Images, videos and sound all welcome. Cloud-based anime generation is also welcome.

Please keep all non-anime gens on >>>/ldg/ >>>/sdg/

Because this is a new general, we are going to borrow a lot of the /ldg/ OP until we get some more anime-focused guides.

>Generating Images
https://rentry.org/comfyui_guide_1girl
https://tagexplorer.github.io/

>Generating Videos
Guide: https://rentry.org/wan22ldgguide
https://github.com/Wan-Video/Wan2.1

Oldfag's wise experience: >>106082956 >>106082967

>[NEW MODELS!:]
Name: ItN's Vintage Merge
URL: https://civitai.com/models/1823418/itns-vintage-merge

Anime Pop Cute Illustrious-XL
https://civitai.com/models/1540768/anime-pop-cute-illustrious-xl

Name: Retirement Mix (NAIXL)
URL: https://civitai.com/models/1821991/retirement-mix-naixl

>[UPDATES!:]
Name: Smooth Mix (Illustrious) v5
URL: https://civitai.com/models/1695233?modelVersionId=2038033

Name: Hotaru Blend
URL: https://civitai.com/models/1696665/hotaru-blend

Name: NanoMix ~Multiverse Remix~
URL: https://civitai.com/models/932505/nanomix-multiverse-remix

Name: Mature ritual
URL: https://civitai.com/models/994401/mature-ritual-or-illustrious

>Anime Checkpoints and loras for Flux and SD3.5
[SD3.5] Anime Style LoKr - Large / Medium
https://civitai.com/models/1807230/sd35-anime-style-lokr-large-medium

[FLUX] Illustrious Anime Style Lora/Lokr
https://civitai.com/models/1059388/flux-illustrious-anime-style-loralokr

[FLUX] 1984 - Anime Style feat. IllustriousXL
https://civitai.com/models/1072833/flux-1984-anime-style-feat-illustriousxl

Anime3.5L
https://civitai.com/models/931532/anime35l

Previous: >>106131439

>>106171902 (OP)
Didn't even bother with the collage? You suck.

Shit thread

>>106172022
This general was not built on collage autism you boring retard.

>>106171902 (OP)
>カヒクン
???

>>106171902 (OP)
>Oldfag's wise experience
Cringe

Any good anime pooping gens?

Thanks for baking!
>>106172022
Op of the last thread
This is the collage I came up with >>106169763

I was on my phone (and will continue to be for many more hours more) and I'm terrible at editing images unless it's in MS Paint :(

>>106173100
Just use the /ldg/ script for collages.
https://rentry.org/ldgcollage

>>106173124
Okay, thanks! But I think I have to use cmd python to run this script, right? I can't do it on my samsung toaster :(
But for the next thread yes!, I'll make sure the .mp4 doesn't get left behind!

>>106171969
Haven't visited general since SDXL being flop
Is it locally generated? How much VRAM it needs? Looks very good, please spoonfeed me

>>106173352
Imposters will tell you that you can run these models with 8 or 12 VRAM, but in reality, the minimum required to generate videos is 24, and the standard is 36 or 48

>>106173368
>36-48
Expected as much, thanks

Forever 2023 bross, forever SDXL bross, forever stagnant bross how are you doing?
Another day, another checkpoint, Lora and infinite varieties with the same structural problems!

> https://github.com/Panchovix/sd_webui_loractl_reforge_y

I recommend using this extension for ReForge! It allows you to control the weight of the LORAs at every stage of the generation process.

Do you want to use one LORA at the beginning of the generation process another in the middle(lets suppose a style lora) and a different one at the end (like a detail lora)? With this extension, you can do that! And much more!

>>106173404
You are literally listening to a schizo shitposter anon.

I am the one who generated this >>106171969 and it was done on an RTX 5070 Ti. 16GB vram.

>>106173441
Don't belive his lies.

>>106173404
>the retard believed the schizo

>>106173579
Can you make them poop please?

>>106173368
Imagine being fooled by schizo to this extend kek

The Wan 2.2 guide is confusing.
I got past the prereqs and am now on the installation step. What do I actually download now? It says to download these models, but there are 11 links and each is like 16gbs in size and takes forever to download. I have 24 gb vram, but will probably want to go down from q8 to q6. The problem is I don't know what the difference is between these model files and dont want to spend the next week downloading the wrong ones

>>106174274
You have to download all of them if you want both I2V and T2V (image2vid, text2vid). That's just the reality of the new 2.2 version, you need high noise and low noise models.
The files from lightx2v and below are much smaller files. Again, they are necessary for the workflow to work properly.
There is no need for you to go down to Q6. I have 16GB vram and I can gen in 100 seconds with the models from the OP. Alternatively you can try the AIO workflow or the bullerwins workflow.

Um, I-I’m a big guy, s-so, I have a... a 3060 12gb, I guess, and, um, w-which version of WAN 2.2 would you, uh, recommend?

Let the one who has never believed in a schizo out of naivety cast the first stone.

>>106174274
>24gb
q8 is perfectly fine

How is this thread /g/ related?

mom i posted it again

>>106174507
Whichever quant fits in your vram capacity. Try Q6 with some dram offset:

https://huggingface.co/bullerwins/Wan2.2-I2V-A14B-GGUF/tree/main

>>106171902 (OP)
>Generating Videos
>https://github.com/Wan-Video/Wan2.1
Shouldn't it be updated to 2.2 now?
Also grim if you need 24gb at very least

>>106175171
>Shouldn't it be updated to 2.2 now?
You're right, should be fixed next thread.
>Also grim if you need 24gb at very least
You don't. Takes me 100 seconds to generate 5 second videos on my RTX 5070 Ti with 16gb vram.

>>106173100
>>106173168
enjoy the collages anon

>>106175966
Mami's mamis.

I'm jealous of all the anons who can genn videos in a reasonable time.

>>106176211
it's not hard just use the self-forcing workflow with lightx2v and sufficient dram cache.

>>106176211
heres 1 4 free :)

>>106176456
thanks even tho I would've liked one with a happy ending more.

>>106176721
im sorry but he 41% :(

Can you gooners just fuck of from /g/?
Thank you.

>>106176870
gooner website
gooner sub

>>106172107
No, it was built out of spite

>>106176908
Whatever helps you sleep at night ldgschizo.

>>106176870
Let me guess, you miss 24/7 Elon threads, don't you?

>>106177233
Damn, the consistency is getting scary good. What's the workflow? Is this Wan 2.2

>>106177710
Wan 2.3

>>106173924
Still waiting on those pooping gens anons. Don't make me train a LoRA myself :3

>>106176870

This thread is 50% VRAM arguments, 30% people not reading the OP, 10% fetish posting, and 10% actual content.
Peak /g/.

any good mahiro animation?

>>106177233
loool
that was actually funny. nice one anon

>>106178778
nice

>>106178979
>do not look at her feet do not look at her feet do not look at her feet

>Asked GPT 5 for generate me a SDXL scene---

Negative prompt: nsfw, nudity, minors,

>>106179214
ty brah

there aren't gay anons or anonnetes? only anime girls, where are my anime boys...

>>106180606
all i got are shitty qwen test prompt gens

>>106180623
and they are cute? :3

>>106180755
oh no, they're more joke gens like bodybuilder dudes in schoolgirl clothing. i didn't read the first part of your previous comment.

>>106180821
w-without shirt?

>>106181134
when I make enough money to sub to nai again

I have a question. I want to replicate an image generated by Tensor on my local hardware. I have the checkpoint, LORA weights, seed, prompts, and so on from Tensor.
My big question is: is Tensor's seed system A111 webui like Forge or ReForge, or does it use the ComfyUI seed method?

bump

>>106172568
CHAINSAW HEEEEAAAAAAD
https://www.youtube.com/watch?v=FmqHs_6VSzQ

>>106184151
very nice

>>106184305
Hands look fine, what os your negatives?

>>106173168
You can do the collage on Android with Termux:

pkg install python ffmpeg
pip install pillow
run the /ldg/ collage script with local paths
It’s slow but works.

try to top this comfy

Goodnight anonies
Tip: at /edg/ they have tutorials on how to make Loras. We can use their OP there since their field is similar to ours, but NSFW.

>>106186835
CHRISTINA

>>106174049
thanks for the nightmares asshole

what is the current meta for Illustrious based models?
I can generate bobs & vagenere very fine, but that's not my focus.
Currently using Illustrious XL 1.0 but I feel it is limited in characters and dataset.
I don't want to go Flux or other stuff since I am a comfy 8GB Vramlet.

>>106178778
Love this style

>>106187615
nice, one of my favorite characters

>>106173168
hello the collage is FUCKED up
the pictures are all horribly cropped STOP

1 post per hour...

>>106187555
What’s your use case?
Mine isn’t goon either.
Reality is most anime ckpts on Civit are hyper cooked for gooning, so if you want SFW slice of life you need to seaech for a checkpoint actually tuned for that(very dificult)
The most rated/downloaded isn’t the best, the top rated are usually goon models.
Or do what I did thd last thread, I used a corpo/heavy model to plan the scene composition (Qwen for me it's the best at anime), GPT, Flux or Google Image ultra 4. And then use the SDXL ckpts with ControlNet that exels at generating Anime characters to generate on top of the image. You will have geometric pattern problems and colour pattern problems, but that is the reallity of SDXL.

>>106188395
2 more bakes

Is fully running wan inside VRAM really that much better? I'm a 12gb vramlet and the speed is okay, might as well buy more ram sticks and raise the output quality

>>106188434
thx for the answer.
I really like my 80s scifi anime with space ships and space stations. but most models/loras seem to focus on the characters. I don't mind a skimpy girl astronaut here and there but I do really like my classic animes.
SDXL so far works pretty well. Isn't qwen very vram hungry?

>>106188460
Any guide i have the same vram but 32gb of nornal ram

>>106188497
Yes, I understand that you want to give the background of the scene the same importance as the character. For SFW stuff like what you want to do, I recommend lmarena.ai.
In the search tab, you have to click on the "image button." There, you can choose any corpo model I mentioned for free.
You request it, generate it, and save the image to your local hardware.

Vramlets concentration camp when...

>>106172604
>>106172604
fuck off

>>106188622
For greater accuracy, specify your character's body type, attributes, and hair and eye color and length. This will make it easier for SDXL to understand! So, if your character has short white hair and a polearm, specify this to Qwen so that it can be included in the draft.

>>106176908
kek

more flies with honey than vinegar anon ;3

>>106188598
Just used the OP guide and it filled by both VRAM and 32gb ram to the brim, sometimes even digging into pagefile and slowing everything down

>>106188692
cool! what model?

>>106188694
disgusting shiny slop

>>106172604
Good gen anon, emulates 2000s pomemon animation perfectly

>>106185708
anistudio already mogged this

;3

>>106188739
kinda neat
a bit sloppa tho

>>106189621
im gonna slop ur ass

>>106189671
u clearly are using dall-e and unskilled

>>106189686
nta stupid animetard

anime?

u n t i l
t o m o r r o w <3

How you thought about >>106190556 180 degree 3D VR locally generated anime?

>>106189621
You're right, I tried to fix it up a bit.

>>106190686
ignore all hateraid on these \g\ boards
there are schizos on the loose <3

>>106190658
It's technically possible to tune square prompt into SBS video, but artifacts will be headache inducing
Also right now you have like 5-10 seconds animations, while you want probably a minute or 5 minute long
It will also be blurry, vr videos need 8k to be good, while animations are upscaled 480p/720p

>>106190686
the skellies look super sloppy, inpaint them bruh

>>106188792
>ani
beauagahahah

>>106190861
nice

>>106173482
C-catbox?

>>106190747
Thanks, it's difficult to notice every mistake when the image is turbojank like that.

now you are thinking about hags
https://files.catbox.moe/n9z9no.png

>>106187630
thankies. I think its just the 'oekaki' tag doing it. im not using any artists or loras

>>106191710
you can make a personal artist lora ^^

>>106192369
>>106192379
>>106192393
No one asked

>Can redraw entire sections of AI-generated images to be anatomically accurate, make fucked-up objects normal, etc. with no problem, it's actually kind of fun
>Realize what I'm doing is actually drawing
>Decide to go try it
>Brain locks the fuck up the instant I get anywhere near how-to art books, etc., can't take any of it in

Does anyone else experience this, or just a similar "decent at computer graphic work but awful at rawdogging drawing" situation? Why the fuck does it happen? It's not that different, and I understand the principles, but fuck, man.

>>106193190
ok this is kino, the character is self aware of the scene

>>106193216
i tried braiding my pubes once and came to a similar predicament

>>106193190
>i you want it THAT bad OK
>hnnng
>*braps*

>>106193190
Model?

>>106193528
wan2.2

Alright, fine, you're convincing me about Wan. How much VRAM do I need for 2.2? Is a 3060 gonna be enough if I have a shitload of regular RAM? Can you even offload it?

>>106193890
the more vram the better but you can get by with 8-12gb so long as you offload to dram. Just try try the rentry guide's installation + workflow and gen at 480p. If you get oom then try the AIO workflow or one of bullerwins quants.

>>106193890
wan2gp repo. wan workflows in comfy will break the install

>>106193925
I'll give it a go. Does it have negative effects on the video quality?

>>106193916
Will try this if it turns out wan2gp messes with the quality. Thanks!

>>106193967
>I'll give it a go. Does it have negative effects on the video quality?
no.
Lightx2v has an effect that might not be desirable but the speed increase is too big. You can't go back to non-lightx2v after using it.

BUMPY ;3

Okay, let's be honest and real.
This is an anime difussion general, right?
I understand that you like to generate anime images through difussion, and that's a very valid and understandable point.
Now, here's my question, why in /g/? What does this have to do with /g/?

Qwen is more heavy than WAN, hilarous

I hope you are not sleeping on making thousands of dollars generating slop

>>106194227
more flies with honey than vinegar anon;3

>>106193967
>Does it have negative effects on the video quality?
no. it's just a slow startup but it's at least stable so you don't have to fix it every update like comfy

>>106194249
I tried and only made $7 kek. I don't know how to market this shit

>>106194227
blame hiro for making a dozen /v/ spinoffs but still no AI board

>>106186835
>>106187615
Nice, we are so close to realizing AI Kurisu gfs

>>106193064
https://files.catbox.moe/ti75a0.png

>>106194123
Why do you always make her put her arms behind her head in your videos?
Don't answer me with that honey thing.

>>106195605
Very nice

Are requests allowed?

mmm, some anon leveled up in video genning and isn't sharing his workflow.... :3

>>106190720
I really would like this on a local model.

>>106196391
to make girls brap? weird fetish bro but alright

>>106196217
yes

>>106196497
Then I am cordially requesting some Nico/Maki.

just dropped $300 to upgrade from 32GB->128GB RAM because of these threads
hope it was worth it

What do you use to butcher characters out of a picture into individual images? I've just been using the background assistant/brush in windows photo viewer, but it's not cutting it anymore.

>>106196709
i've always had the best luck with the magic wand tool in gimp/photoshop

>having to install conda
>it's another fucking nightmare which installs 1 or 2 more python interpreters in bullshit envs
Python tooling just makes me vomit.

>>106196808
Python is a fucking trainwreck

>>106196808
uv solves everything

>>106196816
Now I have two more python interpreters and two nested environments, but the pip command that's next still fails to find a specific version of torch.

>>106171902 (OP)
If you want to share this thread in other boards...

[>>106171902 (OP)]

>>106196853
>JUST DOWNLOAD MORE BLOAT BRO TRUST ME IT GETS BETTER!!1!!!11
python shitters everyone

Does anyone have the Python script to make De3u work with gpt image 1 from the proxy?

>>106196808
with no exaggeration I have over a dozen instances of pytorch installed for different programs and those are over 5GB each

>gigabytes of garbage downloaded
>several pages of python errors on start
Is there just some docker image for this fucking crap?

>try "pinokio"
>installs another couple of python installs and envs
>but now I can watch the shell output through a browser window

How do you manage your checkpoints and Loras Anons? I have many, but I only use the same.

I've only ever made one good lora and it constantly taunts me by being more flexible than it has any business being compared to the rest, and the worst part is I didn't save the parameters anywhere, I don't even recall how it was trained (collab, local, civit), I'm so freaking miffed.

>>106194249
It's like selling your soul for 400 bucks a month. I mean if you are a thirdie that might be worth it, better than back-breaking jobs for that pay, but otherwise it's pretty bad. Realistically you won't get a paying audience with cute 1girls and nice-looking gens, you'd have to spam shitty interracial and other fetish slop to get the money.

>>106173482
>no head tilt
0/10

>>106197960
>WHAT?! YOU GOTTA HAVE LE HECKIN SHAFT HEAD TILTERINOOOO!!

>>106196391
wan is local

>>106195659
>>106198950

>>106196703
>128gb
what platform u on?

how to get these girls to stop yapping?
>>106199546
W10, AM4 Ryzen 5 (so 128GB will max it out) with 4090 24GB

>>106191189
Im curious about your promtps, they are very original. Couls you share one?

>>106183751
post the uncensored version somewhere

>>106200001
>illustration in the style of "the greatest guilt trip in the history of humanity" by h.r. giger with an anime girl. dreamlike. surreal. full shot. psychological. subtle silhouette effect. sitting. front shot. black and white
This is how all of them are framed. I come up with fake painting titles and let the image generator figure the rest out.

>>106200121
>yap while playing violin

>>106200350
As an anime girl, I do this.

Another day, another gigabytes of uncached downloads wasted on a slow connection just to result in a python error.
Has anyone managed to install this on linux?

Can you get this piece of shit to stop spamming in /a/

>>106194227
He's doing it in /a/ too for some fucking reason. He's posting every webm in here too, he never gets new material. I guess a lot of genners just want attention for the shit they make.

>>106202874
And /v/ too.

>>106202943
he should set up a bot to auto post on all boards 24/7

To the guy making the weird, slightly creepy, vaguely grotesque gens --
Keep it up, they're visually interesting

>>106202974
What will that solve?

>>106203139
Yes, I agree.
Personally, I think he's putting up a barrier between what is anime and what isn't.
What is anime? Just cute 2D girls with big eyes doing cute things? Or isn't it the lighting and color or the theme or what the image wants to convey?

I would like him to take one of those gens he makes with Dall-E and try to run it through one of those SDXL anime checkpoints, giving it anime-style coloring. I'm interested to see what happens when the colors, lighting, and characters are anime-style, but everything else borders on insanity.

>>106199029
mp4 of this please

>>106203183
These three in particular evoke a strange feeling:
>>106161076
>>106150659
>>106122227

>>106203201
The anime models themselves aren't too bad at creepy stuff

>>106203274
yes I love them, I would like to see a "remaster" of that pictures althought their dalle slop style adds to the schizo vibes

The only emotion that melted slop should evoke is disgust, not sure why you retards are sucking such sloppers of and encouraging it like it's still 2022 when we were in awe of human-like blobs the AI was producing. It looks terrible and the whole point of visual art is the opposite of that even on the lowest level (unless deliberately stylized). Even redditors have moved on to have higher standards of quality for their slop, but 4chin is full of facebook-tier boomers these days I guess.

>b-but muh grand idea, muh prompterino is art on its own
Don't care, de-slop your ugly shit before posting. There are services available online if you're a vramlet.

>>106203812
Is this about the dalle spammer?

>>106203812
what do you mean?

>>106203812
Hmm, I don't know. At least, I don't see it that way. Just as with today's retro pixel or '90s anime aesthetics, I don't see the "Dalle 3 slop aesthetic" as being so out of context. And I'm saying this as someone who is just getting started and has a critical eye.

One thing I was pondering today while using openAI image services is that with all this copyright and NSFW censorship, I think AI has been creatively delayed by at least 5 or 10 years for text and image and everything

>>106203918
The issue with his pics is that everything is melted to shit. You can barely see what's going on there, the faces are fucked, the quality is awful. I was sure it's just some troll from ldg spamming the thread desu.

>>106191189
I dont know, in the begining i thought that he was trolling but then i thought, "if i have to fix or highres fix a picture of him, how would i do that? How do i prompt that in sd xl? ">>106190686 or this?

>>106203159
>What will that solve?
huh?

>>106171902 (OP)
Hello, I probably won't be here when the thread reaches its limit.
BUT
I have already created a collage with the gens!
I didn't crop anyone! :D
Mr/Ms Baker, please use it for our next great #6 thread! :D

>>106204914
Mm the leters are small

>>106204914
xDDD

>>106204914
Now! :D

>>106204914
>omg iks deh Xd

>>106205166
>omg it migu!

>>106201548
WEW

I'm scared, anons... will we make it to next thread?

>>106206242
I kneel

..why /ldg/ doesn't love us?

I've been trying to find a way to get this but I can't find it.

>>106206791
https://github.com/Kosinkadink/ComfyUI-VideoHelperSuite

>>106206876
Thanks, that solved it

Need a tutorial and workflow, I have a 3060 12gb VRAm and 32gb ram..

I know what you're all thinking but my 3060 12GB is PERFECTLY FINE okay?? It's not about the size, it's about how you use it! And before anyone says anything, 32GB of system RAM is ABOVE AVERAGE - I've checked the statistics!

And no, I'm NOT upgrading to a 4090, my setup performs GREAT when it matters. The benchmarks don't tell the whole story!

>>106207521
WAN 2.2....
Just need that tutorial or workflow link pls (my VRAM is a grower not a shower btw)

>>106207530
Start with the guide in the OP, then tweak from there.

I have a 4070 Ti lying around doing nothing, would it be good at slop generation?

I'm trying to use Kijai's part of the rentry and got everything set up for I2V, but whenever I run the task it reaches this point and then disconnects, I tried messing up with the settings but it always disconnects.

Guess who just remembered his Chinese WW3 enabler pass just renewed 2 days ago?
>>106199983
Not a Konatafag but god damn.
>>106206232
Nani the fuck?

>>106207600
I-IT'S NOT SMALL IT'S AVERAGE!! 12GB IS LITERALLY THE MOST COMMON SIZE!

>>106207646
OOM?

>>106207683
I have a 4090

>>106207702
>I have a boyfriend
that's how it sounded like

>>106207726
well, I tried everything and it just keeps disconnecting. I'll just give up then.

>>106207702
Share your modified workflow, I'll take a look.

>>106207702
Enjoy your oversized power bill!

>>106207665
>Guess who just remembered his Chinese WW3 enabler pass just renewed 2 days ago?
?

>>106207798
Sorry I mean Hailuo, our future AI overlords.

>>106207957
Oh! Shizosloppy chan! How are you? ^^

>>106207845
New video gen model?

>>106208062
Not that I know of. Just a lot (a lot a lot a LOT) of wind in gens I never asked for. Not saying they ain't cute though.

Today I received 5 likes from tensor for this generation >>106192369. Is that good, bad, or nothingburger? Can I make a living from it, like this anon? >>106194249
Yes, I use Tensor because on my main GPU i gen videos, and I need to do soooomething in the meantime.

>>106208353
>Can I make a living from it, like this anon
No.

>>106208469
where's the anime titties

>>106208469
Cool.
I would play this vidya game.

Does anyone have a non catbox moe link for the ldg 2.2 json workflow? I think my isp isnt allowing download.

Newbie here, so I’m looking through the guide, but are the models there different from the standard ones recommended with a fresh install of ComfyUI, then heading into ITV 2.2?

Fuck me in the ass,I take back everything bad I said about AI.
I'm a tech illiterate retard and with only 2 minutes of google search I managed to get this bullshit https://files.catbox.moe/5ag9q3.mp4
I can picture what you are playing with being nerds and all, just imagine where we'll be in a year or two, my dick gets harder and harder at the possibilities.

>>106208056
Look at all this slop. I can't even handle it anymore.

>>106179994
>>106197784
very nice

>>10621067
Nice gen, anon,why does she have a rabbit tail? And the elongated limbs, are you using some kind of embeddings or something? I always say that you shouldn't use "good quality embeddings" or "poor quality embeddings" because they are an easy solution, but it's difficult to fix their problems!

Futari wa Precure!

>>106210990
nice images
>>106210667

Baking

>>106210990
atrocious hands

>>106211157
But its nice

>>106211184
>>106211184
>>106211184
New thread
New thread
New thread

>>106210764
The tail is attached to her bunnysuit. Her limbs are the correct length, if anything her arms might be short, try it on your own body. I am not using any embeddings.