>>106762385
>maids are making advanced technology
>Eli won

File: S3RENEBEAN420.gif (3.59 MB, 275x480)

3.59 MB GIF

>>106762385
I TOO HAVE ENTERED THE ZONE OF Ai SCHIZOS AND OPEN-SOURCE SOFTWARE CONFIGURATIONS

>ani
so is there a specific reason i should use anistudio instead of comfy or indrema?? :o
(not trolling)

File: optimizedSERENEBEAN.gif (3.88 MB, 271x480)

3.88 MB GIF

well???

>>106762432
>>106762454
%100 local
Good UI philosophy
Julien is a good guy

What are the newest AniStudio news?

He is in Japan with a new team of devs,

in a stallmanic sense will i have my 'freedoms'???

>>106762543
Yes, unlike Comfy, Anistudio don't use telemetry

>>106762491
>5 months later still in alpha.
Our boy hungry trani is still full of beans!

>>106762935
Please beta test Anistudio for us, our team is working hard.

>>106762436
If you're still here, it was this post. Haven't tested it extensively enough for myself though.
>>106443690

Ani is a good guy i support him

>>106763085
Prove it.Show feet pics of the dev team. This is standard procedure for verifying developer dedication.

>>106763085
I'm downloading it now for real. If this thing bluescreens my machine I'm coming back here to schizopost so hard the mods will have to create a new containment board.

>>106762385
So does this thing have LoRA training built in yet or do I still need to fire up Kohya's

where pw :[

>>106763451
LET HIM SLEEP!! >;c

still more active than /s*g/ kek

>>106763697
Based

Look cool, but where are the zipped releases?
GPU dependencies page 404's, not yet ready?

>>106763951
>2025
>thinking in GPU terms
This is an imaginationware UI general

>>106763974
>>>/b/

I want to use it and be free of the python virtual environment web gui double whammy of bloat but for some reason sd.cpp takes 6 times as long to generate an image with SDXL models as pytorch.
They even got offered cash by the civit.ai nigger to fix it and nothing has been done https://github.com/leejet/stable-diffusion.cpp/issues/772

>>106762385
HAHAHAHAHAHAHAHA HAHAHAHAHAHAHAHA NO ONE FUCKING LIKES YOU SO YOU CREATED A THREAD TO SEGREGATE YOURSELF, you're a joke julien you and [redacted] aka comfy.

>>106762385
>Retard doesn't link to/releases in readme because only speshul smart ppl should find and use this software

>>106762385
I belive in your project

>>106762454
>>106762432
sexo

>>106762385
this just dies if i use it on my 50 series blackwell card

>>106765754
What a cute thread.

Why dint we post our Anistudio gens?

Julien is the good guy

>>106762385
>4chan's birthday
>1st Anistudio thread
BASED

>>106766173
we can't make them because the software doesn't work ;D

>>106764043
Stop lying schizo

>>106762385
Newfag here
This thing just crashes for me, comfy works on the same machine, any idea?

>>106766750
Okay i give up
What a pile of shit

>he fell for it

>ani studio just flew over my house!

>>106763951
2 more weeks

>>106762752
QRD on comfyui telemetry?

>containment general for julien
Based and valid

>another worthless general for the filter

Is julien still in japan? Where are the development news regarding AniStudio?

>>106766892
Actual schizos who wants to gaslight Anistudio
Anistudio works in my machine

>>106766750
>schizo
Did you try atleast tike the VAE techlet?

more like anustudio

>works on my machine
devnoob

>>106767724
Just tile the VAE bro

>>106762385
Why is it every time I ask for an .mp4 of AniStudio working, just a simple txt2img, I get called a schizo instead?

Is UFO footage easier to get than proof this thing actually works?

File: AniStudio_InterOpTest-00693.webm (1.56 MB, 736x960)

1.56 MB WEBM

why is there a thread for anistudio? oh well

>>106767525
no I landed home yesterday

>>106768049
the readme has a screenshot but could someone post the output where it crashes because everyone I asked directly to test got it working fine. dunno what's up with the same guy saying it crashes all the time unless he's using NvidiaTK v13. I only build with 12.9 so it only works with that

>makes new UI
>recycles the same handful of images he made in comfyUI
>pretends he didn't make this thread while drunk
I guess forcing your UI in other thread OPs is not enough.

>>106768114
ok schizo

It's a great idea to bundle / contain all the tranistudio spam here
Based schizo

>>106768087
Come on, man, you have your own general here, please never come back to our threads.

>>106768299
post gens from anistudio and I will

>>106768087
>why is there a thread for anistudio? oh well
To celebrate it being renamed to maidstudio.

>that drawing
>no maid dress
Sad times.

>schizo
>makes thread to shill
>has no support
>only cope is to call others schizo
>>106768320
He has Greek taste

if you fix it so it works on blackwell cards i'll use it over the python slop

>>106768314
But it crashes on booting the app
Are you trolling?

>>106768465
He can't vibe code it yet

>>106768491
got any cmd outputs to back that up? why is it so hard for you to post the error outputs? are you retarded or something?

>>106768533
nta but on the release that's on the github I get told there's no CUDA kernel to run and it crashes when I queue up an image
when I built it from source for CUDA instead it used CPU

File: Screenshot_20251002-110157.png (396 KB, 924x1749)

396 KB PNG

>>106768539
>I get told there's no CUDA kernel to run and it crashes when I queue up an image
can you be more specific?

>>106768576
the cuda binary for linux

>>106768583
I probably messed up the build. the shell script might be forcing CPU and AI didn't catch it. I'd suggest using the cmake commands directly to make sure cuda is forced for now or try vulkan to see if it's the same issue

>>106768592
Vulkan crashes at the end of image gen but at least seemed to send the model to GPU and use it for compute
seemed slow though compared to the speed I typically get on invokeai with my RTX Pro 6000 Blackwell

[SD_LOG] stable-diffusion.cpp:2064 - sampling completed, taking 9.84s
[SD_LOG] stable-diffusion.cpp:2072 - generating 1 latent images completed, taking 9.88s
[SD_LOG] stable-diffusion.cpp:2075 - decoding 1 latents
ggml_vulkan: Device memory allocation of size 6979584016 failed.
ggml_vulkan: Requested buffer size exceeds device memory allocation limit: ErrorOutOfDeviceMemory
ggml_gallocr_reserve_n: failed to allocate Vulkan0 buffer of size 6979584016
[SD_LOG] ggml_extend.hpp:1418 - vae: failed to allocate the compute buffer

Segmentation fault         ./bin/AniStudio

i can assure that nothing should exceed 96gb vram for this kind of gen

>too mentally ill to listen to feedback
>goes on drug induced spergouts when criticized
Remember the first time anon gave you feedback and you had a melty?
This is how you got soft blocked from /ldg/ :]

>>106768592
and here, for cuda fast binary on linux:

ggml_cuda_compute_forward: GET_ROWS failed
CUDA error: no kernel image is available for execution on the device
  current device: 0, in function ggml_cuda_compute_forward at /home/julien/AniStudio/external/stable-diffusion.cpp/ggml/src/ggml-cuda/ggml-cuda.cu:2522
  err
/home/julien/AniStudio/external/stable-diffusion.cpp/ggml/src/ggml-cuda/ggml-cuda.cu:87: CUDA error

>doxxing himself again
You on pills again Ani?

>>106768620
yeah, vk needs to tile the vae, it doesn't operate like cuda where it trickles down efficiently for the decode. it just hamfist it all at once like it's a shader. stay within 1024x1024 as well

>>106768649
ok so it looks like it is just trying to find the lib in the wrong place, I thought I fixed this but it's still hard baking the paths so building manually would be the only thing that works for now. I'll see what I can do when I get home

>tfw this thread has more technical discussion than /sdg/ and /ldg/ combined
pottery

He can't really fix anything because he only cobbles together other people's code and uses ai to do the heavy lifting.
>>106768719
>lying
Kind of like how he always call anons giving him feedback trolls followed by seething?

>>106768592
>I probably messed up the build
Oh so yet again anon was right
Maybe learb to listen to potential users bext time instead of schizomaxxing for months

>>106768690
>I thought I fixed this but it's still hard baking the paths
Not gonna lie that makes you look like an absolute noob who vibe coded a shitty wrapper to larp as an actual dev

>>106768783
well no, anon was complaining about not being able to start it at all and still hasn't posted the output at all. nice try though

>>106768783
He needs to lay off the drugs first

>>106768804
First time dealing with trani?

>>106768690
im pretty sure my source build for CUDA didnt really work well either but I can retry in a bit
I do have a feature suggestion: I'd love to have a font size option rather than needing to adjust imgui render scale because that is super blurry and im on a 4k monitor so default font size is a bit too small

>>106768859
yeah, imgui rasterized the font by default so it's the only thing I could do as a placeholder for now. I saw an extension for anti-aliasing on the fonts but I think just having more scaled font options would be for the best. as for the inference backends I'll probably just have different options in the settings and curl the official shared libs to switch on the fly like the plugin hot reloading. that should solve the lib path issues, speed up builds and generally be more convenient for testing

>>106768649
>doing imaginary debugging with a linux user that happens to have his first name
Holy schizo Batman

>>106768927
i understand you guys have some beef with this dude but nah i'm not him
the path is clearly hardcoded somewhere for that build unfortunately
source builds fail for me but my system has the latest build tools installed (gentoo) so some things may be too new for what the codebase is built around

>>106768898
thanks for keeping at it ani, a lot of this stuff goes over my head since I only know python. I guess this will be my first taste of c++ and honestly it's overdue, I've been putting it off because I didn't think it would be worth it but you showed me the light

>>106768649
I’m not a leer haxor coder, but notice how it calls for Ani’s external hdd and not your own cuda location?
Had this happen too, I believe it was the same error at the start of genning

It will not find it bc its not looking in the right place
This was my previously given ‘most crucial’ part of feedback, since I personally have no fucking clue how to fix it

>>106769114
if he is switching to using the official binaries that shouldn't be an issue, just have to wait for the next update I guess

>>106769124
I will, for real
To my simple small brian this seems like 3 lines of code but it must be more difficult

>>106769124
ani has been hyping the speed like crazy so waiting a day or two is fine by me.

holy samefagging kek

Ok, I want to use Anistudio and post gens in /ldg/ with the Anistusio prefix is there any step by step tutorial?
Txt2img and some inpainting windows 10

>>106769290
download the windows release then just unzip and run

>>106769314
Ok thanks

>>106768649
AAAAIIIIIIIIEEEE WHY HE IS SIMULATING A DEBUGING

File: 23434.png (15 KB, 303x214)

15 KB PNG

>>106769314
Whhich one? I have an nvidia 3060 12gb

>>106769632
either of the cuda ones

>having hardcoded paths in his vibe coded app is a problem which takes him months to solve

where do I put "start"?

Help, i don't know where the "generate" buton it is!

File: 1.png (1.06 MB, 1024x1024)

1.06 MB PNG

>>106766560

[INFO ] stable-diffusion.cpp:2072 - sampling using Euler method
[INFO ] stable-diffusion.cpp:2121 - generating image: 1/1 - seed 1
  |==================================================| 14/14 - 1.19s/it
[INFO ] stable-diffusion.cpp:2158 - sampling completed, taking 16.64s
[INFO ] stable-diffusion.cpp:2166 - generating 1 latent images completed, taking 16.65s
[INFO ] stable-diffusion.cpp:2169 - decoding 1 latents
[INFO ] stable-diffusion.cpp:2179 - latent 1 decoded, taking 45.21s
[INFO ] stable-diffusion.cpp:2183 - decode_first_stage completed, taking 45.21s
[INFO ] stable-diffusion.cpp:2475 - generate_image completed in 61.99s

best quality, 1girl, erika \(pokemon\)
Negative prompt: worst quality, old, early, low quality, signature, username, logo, bad hands, mutated hands,
Steps: 14, Sampler: Euler, Schedule type: Automatic, CFG scale: 5, Seed: 1, Size: 1024x1024, Model hash: bdb59bac77, Model: waiNSFWIllustrious_v140, RNG: CPU, Emphasis: Original, Version: classic

Time taken: 9.4 sec.

A: 7.46 GB, R: 7.91 GB, Sys: 9.4/11.6309 GB (81.1%)

You're right I am schizo it's more than 6 times slower.

O SHIT IM GENERATING

Oh no! what happnd after %100 the program and CMD closes abruptly

Okay, sorry, but the program closed itself after generating 100%, at least I tried.

>trani is crying

>>106769870
Do you really edit console output schizo? What has ani done to you?

>>106770040
He's a good boy who din du nuffin I just want him to fix it being slow so I can use his software.

>gets kicked out of 3 different generals for calling everyone giving feedback a schizo
>still an attention hungry faggot
>bakes his own general for maximum attention to him and his commercially licensed wrapper
>...
>needs to ack "the schizos" were right all the time
And that's why you're a lolcow

>DEBI ABD RANFAG SITTING IN A TREE

Why is there a general specifically dedicated to image and video diffusion UI with NOTHING, I repeat, absolutely NOTHING uploaded?

ani should create a discord for us i think
anyone willing to be a mod so we could keep schizo anon out?

>>106769870
use vae decode only and flash attention

we need a discord ASAP

>>106769870
are you even using the cuda backend or vulkan? comfy can't use vulkan at all

>>106770618
>comfy
Why do you mention Comfy, this is your general, why isn't there a gen posted and we're already past 100 posts? your UI isn't working

>>106770634
BECAUSE IT CRASHES AT 100% YOU BRAINLET. I get a black preview box and then the console window just fuckin' dies. No image saved, nothing. It's like the program is ashamed of its own output and commits sudoku.

>>106770340
And have him call us all schizos in real time? No thanks. The thread is containment enough.

>>106770393

./sd -m /media/models/Stable-diffusion/expmixLine_v3.safetensors --sampling-method euler --vae-conv-direct --diffusion-fa --cfg-scale 5 -t 16 --steps 14 -H 512 -W 512 --seed 1 -p "best quality, 1girl, portrait" -n "worst quality, old, early, low quality, signature, username, logo, bad hands, mutated hands" -o /tmp/1.png

Heres my full command what am I missing?

>>106770618
CUDA backend on RTX 3060 12GB, I'm comparing to forge classic I don't particularly care for comfy either.

>>106770654
I have this problem with Ani's latest binaries too but the previous ones he put up worked and performed the same as using sd.cpp directly.

>be me
>see /anistudio general/ has 100+ posts
>get excited, think there's finally some cool gens
>open thread
>it's 90% schizoposting about a black image and path errors and schizo auto debugging
>close thread
My disappointment is immeasurable and my day is ruined.

>>106770668 (Me)
Whoops that was a different command checking if sd 1.5 models were slower too (15~s on sd.cpp 2.1s on forge classic), this is the one for the image and output I posted earlier.

./sd -m /media/models/Stable-diffusion/waiNSFWIllustrious_v140.safetensors --sampling-method euler --vae-conv-direct --diffusion-fa --cfg-scale 5 -t 16 --steps 14 -H 1024 -W 1024 --seed 1 -p "best quality, 1girl, erika \(pokemon\)" -n "worst quality, old, early, low quality, signature, username, logo, bad hands, mutated hands" -o /tmp/1.png

>>106770668
>>106770678
>he
come on man

>>106770677
It's getting renamed to MaidStudio. Does that make your day any better?

When given the spotlight they desperately desire they often prove to the masses why they have the lot they have in life.
I expected better from you Julien, you don't even know how to use linux yet you spent years larping claiming that you're some important figure in this community.

>>106770677
But anon, it's not about the AI art you generate, it's about the friends we schizoposted with along the way...

can i generate animated juliens with AniStudio?

>>106770668
conv direct uses the vae convolutions on cpu so that's one that isn't very good for such a small model

>>106770862

[INFO ] stable-diffusion.cpp:2072 - sampling using Euler method
[INFO ] stable-diffusion.cpp:2121 - generating image: 1/1 - seed 1
  |==================================================| 14/14 - 1.17s/it
[INFO ] stable-diffusion.cpp:2158 - sampling completed, taking 16.33s
[INFO ] stable-diffusion.cpp:2166 - generating 1 latent images completed, taking 16.33s
[INFO ] stable-diffusion.cpp:2169 - decoding 1 latents
[INFO ] stable-diffusion.cpp:2179 - latent 1 decoded, taking 2.05s
[INFO ] stable-diffusion.cpp:2183 - decode_first_stage completed, taking 2.05s
[INFO ] stable-diffusion.cpp:2475 - generate_image completed in 18.51s

Much better results without it but still about twice the time of forge classic. Guess the "should improve the performance" line about that flag from the sd.cpp readme doesn't apply to SDXL.

Uh oh stinky stinky stinky
What would your comicbook shop friends think of this Mr know it all

File: file.png (376 KB, 1384x866)

376 KB PNG

>>106769870
>>106770931

[SD_LOG] stable-diffusion.cpp:2064 - sampling completed, taking 4.84s
[SD_LOG] stable-diffusion.cpp:2072 - generating 1 latent images completed, taking 4.90s
[SD_LOG] stable-diffusion.cpp:2075 - decoding 1 latents
[SD_LOG] ggml_extend.hpp:1428 - vae compute buffer size: 6656.25 MB(VRAM)
[SD_LOG] stable-diffusion.cpp:1447 - computing vae decode graph completed, taking 0.67s
[SD_LOG] stable-diffusion.cpp:2085 - latent 1 decoded, taking 0.67s
[SD_LOG] stable-diffusion.cpp:2089 - decode_first_stage completed, taking 0.67s
[SD_LOG] stable-diffusion.cpp:2372 - generate_image completed in 5.69s

it's odd because I'm getting ~7 second gens on comfy with a 4090. is there another optimization flag for forge you are using? I am at least using xformers and FA on comfy for comparisons. all the image saving logic is all messed up too with my current changes. I have a lot on my plate rn.

>>106771020
oh right, 14 steps, 5cfg. here is the output:

[SD_LOG] stable-diffusion.cpp:2064 - sampling completed, taking 3.49s
[SD_LOG] stable-diffusion.cpp:2072 - generating 1 latent images completed, taking 3.55s
[SD_LOG] stable-diffusion.cpp:2075 - decoding 1 latents
[SD_LOG] ggml_extend.hpp:1428 - vae compute buffer size: 6656.25 MB(VRAM)
[SD_LOG] stable-diffusion.cpp:1447 - computing vae decode graph completed, taking 0.69s
[SD_LOG] stable-diffusion.cpp:2085 - latent 1 decoded, taking 0.69s
[SD_LOG] stable-diffusion.cpp:2089 - decode_first_stage completed, taking 0.69s
[SD_LOG] stable-diffusion.cpp:2372 - generate_image completed in 4.36s

>>106771020

export COMMANDLINE_ARGS="--listen --api --cors-allow-origins=* --theme=dark --enable-insecure-extension-access --cuda-malloc --cuda-stream --fast-fp16 --model-ref /media/models/ --skip-prepare-environment --skip-install --skip-python-version-check --skip-torch-cuda-test --skip-version-check"

>>106771088
ok so therte are ggml flags I haven't played around with but yeah you are playing with more flags. I want to map out the relevant ones that matter but I'll have to go through the ones outlined in this issue:
https://github.com/leejet/stable-diffusion.cpp/issues/772

>>106771115
It gives me 10.6s without the --cuda-malloc --cuda-stream --fast-fp16 flags but there is still about 8 seconds extra time taken by sd.cpp there.

Are you going to apologize to the anons that are helping you or are you going to do that autistic loser thing where you can't own up to being wrong?

File: Screenshot_20251002-160533.png (175 KB, 1080x1175)

175 KB PNG

>>106771212
I am thinking it's the memory bandwidth limitations or FA just having a larger impact with more cuda cores, one of the two. I see the author is also testing on a 4090 but that doesn't help diagnose the bottleneck on your card specifically. I do want to say thank you for taking the time to test, it really does help give me a better idea of what's going on with different hardware

>>106771255
>I do want to say thank you for taking the time to test
HOLY SHIT HE ADMITTED IT. HE FINALLY ACKNOWLEDGED USEFUL FEEDBACK WITHOUT CALLING SOMEONE A SCHIZO. MARK THIS DAY ON YOUR CALENDARS BOYS.

>>106771301
>inb4 "he called us schizos"
We are on 4chan. We are all schizos.

Okay, so the performance is real then. It's not just a vibe code issue, the underlying library itself is slower on certain hardware configs. This is good to know.

For the anons getting crashes at 100%, can you check if there's a `./output/` folder created in your AniStudio directory? The saving logic might be failing silently if the path doesn't exist or is read-only.

>>106771255
It's an EVGA 3060 XC I usually run it with power limit on 100w with 175MHz offset on core 500mhz on memory overclocks, bumping the pl to the stock 170w brings sd.cpp down to 15.84s from 18.51s but forge classic without the extra flag only goes to 9.1s from 10.6s.
Not sure if that helps but figured I'd post the results for you since I was checking and it had more of an impact on the sd.cpp time.

i hope you dumb shits enjoy the spyware you downloaded

>>106771321
No this particular clown is a schizo, he destroyed his career prospects by leaking his personal information in a catbox and then getting caught posting loli and shota. There's a reason why none of his former peers want to be associated with him.

>>106771344
>For the anons getting crashes at 100%
that is just the vae pounding their gpu too hard. my suggestion is to just use tiled vae. for the saving logic, I am appending a default \output.png somewhere in the new binary but that is appended after the actual filename.png so it just fails (I am just dumb and missed this). I'll get back at it and hopefully have the new binaries out tomorrow since I have to split things into shared libs anyways. at the very least people should be able to run the application and pick a backend at runtime when I am done

>he destroyed his career prospects
bruh I just got off the horn with google lmao. they want me to work on the world model

>>106771420
>bruh I just got off the horn with google lmao. they want me to work on the world model
I would love to be a fly on the wall when schizo reads this and punches holes in the wall

"Ok, so when can we have an Anistudio update with all these implementations, it still seems very crude/basic to me. Also the pixel art aesthetic isn't for me, I prefer something more neutral or modern, not pixelated

>he's drunk again
Stop shilling in the other thread and nobody would bother you, better yet sober up before you fail a QA interview for Google.

>guys i'm a real dev now
>not only am i working since 3 years on my wrapper
>no
>i also read github issues and interpret them

>>106771464
He just revealed he's been unemployed since leaving SAI and needs money. Never seen anyone in this field use a interview and not a actual offer as a flex.

>>106771472
he just got back from Japan. you haven't left your shit smeared office chair

>totally working UI
>143 / 8

File: huggingface.png (5 KB, 199x89)

5 KB PNG

>>106771472
oh interesting

>>106771484
Is that supposed to be a flex on a tech bro salary?

>>106771420
Does that mean we could use the webui API as a backend for your GUI instead of sd.cpp? That would certainly sidestep the performance issues.

>>106771519
I would honestly leave that for community plugins but the example plugin and python interop code should be enough for people to tinker with other backends. dealing with ggml is enough on my plate but making an API plugin shouldn't be too bad desu.

>tfw the "unified C++ solution" can't even save a .png file
This is so sad.
Alexa, play Despacito.

>>106771499
It explains why he has had so much free time to beg but it also makes his progress even more embarrassing. Look how bitter he still is, he has to post in both threads and go on a bender trying so hard to be accepted.
He's lied so much at this point it's kind of heartbreaking seeing him continue down this path.
>>106771484
If you don't have enough money to relax and travel as a single man in his 30s on a tech salary you might need to go into assisted living.

>>106771472
>>106771499
wait did they fire him? not up to date with the trani lore

>>106771484
>he

>>106771572
He thought comfy would take him and use his frontend.
As you can see that never happened.

File: anistudio_0001.jpg (755 KB, 1152x1920)

755 KB JPG

BEHOLD. THE FIRST (OFFICIAL) ANISTUDIO GENERATION IN THIS THREAD.

Prompt: masterpiece, best quality, 1girl, blue hair, maid dress
Steps: 20, Sampler: Euler A, CFG: 7, Size: 1152x990

Took about 22 seconds on Vulkan. It works!

File: AniStudio-01816.png (1.33 MB, 1024x1280)

1.33 MB PNG

>>106771644
based vk chad!

give me some other prompts!

>>106771679
Please make a maid with huge boobs

>application has been worked on for almost 2 years
>lacks most UI features
>can't do high resolution pass
>most of it is him copying other people's code and using ai
Comfy has nothing to worry about
Just stop advertising in the other threads and we'll wish you the best, also stop seething at comfy.

File: AniStudio-01250.png (1.67 MB, 1024x1344)

1.67 MB PNG

>>106771679
https://tagexplorer.github.io/
I normally just experiment with different artists but meido gens are always welcome!

>>106771723
seconding this!

File: anistudio_0002.jpg (180 KB, 1024x1536)

180 KB JPG

Prompt: masterpiece, best quality, 1girl, blue hair, maid dress, big breast
Steps: 20, Sampler: Euler A, CFG: 7 Size: 1536x1024

Took about 40 seconds It keeps working!

do you hear that? the beating of local gen's heart? I am getting that same excitement from years ago, I can't entirely put down the feel in words

File: AniStudio-00903.png (1.17 MB, 768x1024)

1.17 MB PNG

>>106771764
nice booba

>>106771777
local is the essence of freedom and freedom is the essence of ani! (ok I'll get back to work now, this thread was a lot more productive even though I'm pretty sure it was a troll bake)

>>106771764
40 seconds for a 1.5 model gen, my grandma's toaster runs A1111 faster than this.

>>106771873
a1111 can't use vulkan and neither can comfy. it's exclusive and dependency free

>be ani
>get called out for speed issues
>immediately have two anons post successful generations
The schizo coordination is impressive. Almost like it's the same person with two VMs

>>106771887
>be ani
I wish. he goes to japan and doesn't afraid of cpp!

>>106771887
He's not that smart. This is just the natural lifecycle of a lolcow thread. The true believers come out to defend their messiah after the initial wave of criticism.

>>106771931
t. lolcow

Alright, I'm out. This is like watching someone try to build a car out of mud and sticks. It's kinda impressive in a primal way, but I'm not getting in that death trap.
Call me when it can do a simple img2img without requiring a blood sacrifice to the CUDA gods.

>larper trani is back and very hungry for attention
fun times ahead

>>106771887
The selling point is being able to say you're not using Python slop while using C++ slop that's somehow slower. It's an ideological victory, not a practical one.

>>106772013
One is a corporate speculator dev of a UI, the other is a guy like one of us with a dream. But in the 4chan local models general they defend the first one...

>>106772048
Julien its like watching a toddler challenge a heavyweight boxer. It's cute.

>>106772013
comfy is regretting those words >>106771255

>>106770074
it did seem slow on my giga nvidia card compared to invoke as mentioned earlier
couldnt see output though because it died

File: ComfyUI_temp_snhsx_00008_.png (1.53 MB, 904x1328)

1.53 MB PNG

So, if I get it correctly, a maid or two came along and it's working for them too now?
How wonderful

>>106772116
ani was telling the truth the whole time. there are more gains to be had!

>>106772089
You could try sd.cpp directly if you want to see how it runs without the crashing issue https://github.com/leejet/stable-diffusion.cpp

and it's all on topic because the thread is about him
we should keep that kek

>did nothing
>has a general entirely dedicated to himself
why?

>>106772142
think i'll have to since compiling anistudio from source doesn't work for me with CUDA since I'm on GCC 15

>>106772245
ok turns out that's a stable-diffusion.cpp cuda thing
thanks nvidia

>>106772214
wdym did nothing?

File: maid.png (1010 KB, 1024x1024)

1010 KB PNG

>>106772245
>>106772261
I couldn't manage to get ani to compile either on CachyOS but sd.cpp compiles fine for me with CUDA.

>>106771883
>>106772116
Pic related 241 seconds using Vulkan on my maid PC (gpd win 4 7840u), beats forge and comfy by infinity percent because pytorch (rocm) just segfaults past kernel 6.6 on this thing.

>>106772307
if i swap gcc with eselect to 14 (installed because of nvidia cuda shit ofc because nvidia is always behind) I can get sd.cpp to cmake for once, so probably the same deal for anistudio if I want to retry that
waiting on a kernel build though

holy shit ani predicted saving amd vramlet. it's like a prophecy

>>106772339
>503
so happy!

File: AniStudio-00065.jpg (301 KB, 1280x1280)

301 KB JPG

Very intuitive, thanks for fixing it!

File: AniStudio-00024.jpg (3.11 MB, 6304x4456)

3.11 MB JPG

>>106772346
Handles big resolutions very well

remember this day lad. this is the day local won

File: AniStudio-00002.jpg (1.31 MB, 1920x2880)

1.31 MB JPG

new to anistudio? does it support v pred?

python is dead!!!! long live C!!!!

>>106772398
yessir!

File: 1750536444907140.jpg (391 KB, 1493x1152)

391 KB JPG

>>106772322
ok yeah with GCC 14 and building from source on latest master with

./build.sh --cuda --clean

I can finally get output as long as I dont have a VAE selected or use the fp16 VAE file instead of the default
I'm getting 5.79s for 1024x1024 at 32 steps Euler Karras compared to 2.625s for the same model, size, and basic settings in invokeai on my RTX PRO 6000 Blackwell
I did turn on flash attention but I'm not sure if there's other settings I need to hunt for that invokeai probably does automatically perhaps?

>>106772475
--fast-softmax

>>106772475
now do some janny shitposts

>>106772363
impressive

>>106772481
is this another build.sh script argument?

>>106772500

echo "Stable Diffusion Backend Options:"
    echo "  --cuda               Enable CUDA support (NVIDIA GPUs)"
    echo "  --vulkan / --vk      Enable Vulkan support (cross-platform)"
    echo "  --hipblas / --rocm   Enable HipBLAS support (AMD GPUs, requires ROCm)"
    echo "  --metal              Enable Metal support (Apple devices only)"
    echo "  --opencl             Enable OpenCL support (requires OpenCL SDK)"
    echo "  --sycl               Enable SYCL support (Intel GPUs, requires oneAPI)"
    echo "  --musa               Enable MUSA support (Moore Threads GPUs)"
    echo "  --fast-softmax       Enable fast softmax (CUDA/HipBLAS/MUSA only)"
    echo ""

>>106772521
oh sorry I should probably clarify: does this flag exist for just the stable-diffusion.cpp build or anistudio itself?
doing a clean build by force removing the build dir atm since it didn't seem to change anything when i tried anistudio with that flag

File: 1744412892815133.jpg (85 KB, 954x237)

85 KB JPG

>>106772569
i do see it here but even with it chosen alongside --cuda, I am getting a bit over 6s on a 32 step gen still
I assume stable-diffusion.cpp might still have some catching up to do with pytorch

>>106772569
https://github.com/FizzleDorf/AniStudio/blob/26d7ed2220960317a03c7c9b0140e0a7194be1e3/CMakeLists.txt#L23
it's in here and
https://github.com/leejet/stable-diffusion.cpp/blob/abb115cd021fc2beed826604ed1a479b6a77671c/CMakeLists.txt#L34

>>106772601
he was also talking about the other opts he hasn't played with yet >>106771115
it's only a matter of time

File: 1739087719613224.jpg (476 KB, 1493x1152)

476 KB JPG

if there's speedups that can still be made then that's still fine by me
I've been wanting a tool I can use for both images and video models that isn't node-based and avoiding python is also a pretty big bonus because I'm sick and tired of all of the fucking venvs I'm juggling around just trying out ai slopware
my main pet peeves/requests for now are:
>need to be able to increase font size since its way too tiny to read at 4k
>prompt spaces need to have word wrap or something at the very least and be resizeable so I can easily look at giant prompts and organize them
>readme doc for install needs to be updated in main repo readme so it doesn't go to a bad URL
and then maybe just some guidance on how to make things work well since it's a new UI to me and I'm not well versed in all of the aislop terms that have grown into a full dictionary these days

File: Maid1.jpg (2.26 MB, 3136x2048)

2.26 MB JPG

Turned into such a fun thread
Well done maids xd

>>106772760
also it doesn't seem to reopen the history of outputs for a given project when I close and reopen the binary fully
all options chosen for model, vae, size, etc. all also reset when the binary is closed

>>106772845
now stop the spam please

>>106772854
Anon, but they are local gens, wtf are you talking about?

is this a thread we should continue or do we join back with /ldg/?

>>106773171
>doesn't know that in this thread there was always one person
how naive you are ldgSchizo anon

Threads like this are why I love maidposters. A bunch of assholes were bullying Ani for no reason and then suddenly the whole thread was doing heart hands. The overwhelming power of politeness and positivity transformed the thread from something dark and boring into a shining maid cafe on a hill.

>>106773171
fuck ldg

Sorry Ani, /ldg/ lied to us again.
Now they’re using my gens, the ones you helped me set up today with your UI, in their threads. And you are not in the OP again. Sorry

>>106762385
umm, use case?

>>106774449
Berry nice indeed
>no verification required

Maidposters are incredibly skilled.

Anistudio update? /ldg/ says if you release another build with all the fixes, they’ll put it in their OP.

>>106776802
I was joking kek

File: AniStudio-00042.jpg (241 KB, 701x1200)

241 KB JPG

>>106772760
Ani, don't listen to the haters. You're building the future. Watching this thread go from "it crashes" to people posting gens was the most fucking inspiring thing I've seen on this site.
Keep cooking my man!

>>106777044
He fixed stuff in real time. That's a real dev for me. We see you, king. Take all the time you need for the next build.

Keep going, Ani. Some of us remember what "local" actually means.

File: Doom maid.jpg (380 KB, 832x1216)

380 KB JPG

>>106762385
Ani, you have the full backing of every maid on this site right now. How can we further assist you?

>>106778000
KEK
based (and sext) maids