#🍥|anime
1 messages · Page 144 of 1
im tempted to try sdxl with that as well
since there's no need for hires fix
so probably gonna be good

Double thumbs and a pincer
i stopped caring too much about hands
don't make me more hungry than i already am
make it make the most delicious meal, don't need the waifu!
Im getting hungry as well so 
we need waifu cooking our meal 
That’s gonna be super cursed, I can already tell
But well, I’m making new engines
So gonna wait
In case anyone needs this kinda stuff 
Can extract that from any video if you need anything else 
Howw
It can't miss how a lasagna looks surely
i don't know
it's probably gonna look like those ai video about Will Smith eating lasagna
okay
build engine, one for 512x768
and another for 1024x1536
so when hires fix with 512x768 at 2x
it will be fast
Im tempted to say...
Not sure how to get the hands more consistent 
here's one 150 steps with 10 hires fix at 37.3 seconds
maybe with a base video with less agressive hand movement
you said before

this
is this with animatediff? or deforum?
i think that's 512x512

if im not wrong
taking notes without details
AnimateDiff
1 min 16 seconds for 150 steps with 70hires fix
with controlnet input
since you have a video input, isn't it better with deforum? 
Idk, havent used deforum before 

might be worth checking it out
I've gotten "pretty good" result with deforum and using video input
but that's ages ago,
so deforum should be better now(?)
from what ive seen animatediff looks way more consistent
is it? 
idk all the deforum videos ive seen have this weird jumping look
then i might be wrong 
okay i think i get what you mean
the very low fps jumping
A tutorial on how to get consistent animations using Deforum and Controlnet units.
HOW TO SUPPORT MY CHANNEL
-Support me by joining my Patreon: https://www.patreon.com/enigmatic_e
SOCIAL MEDIA
-Join my discord: https://discord.gg/ZuGj5nJGut
-Instagram: https://www.instagr...
like this
While i get a really nice output from animatediff besides the hands
trying different samplers, if that doesnt help I might need another input
Heres my current workflow for those interested
@lyric zealot so how long does it take for a 20 steps, 512x768 hires 2x with trt now ?

this is 20steps and 20 hires steps
27.3 seconds
should pin that 
Maybe if I got a final version, still working things out
But maybe someone else plays around with it and can improve on
if only i can not be lazy and actually use comfyui
https://pixai.art/?tag=challenge&challenge=13 anyway here, you might be interested
Cool and all but I run locally and dont need the credits 
Also I dont wanna advertise for random websites
can't you enter the challenge using local stuff?
I didn't read much
just went and look at some of their generations 
yaaay, lyricsprompts fixed
TBF the entries all look really boring so far
"just generic waifu"
lyrics waifu best waifu!
nice. tensorrt also works with region prompter
this might interest you 
@wispy canopy what times are you getting now compared to before
it's about 2.3x faster if i adhere to a single specific prompt length
but the profile selection goes wonky when it's a different prompt length
tell me in seconds and minutes
30 seconds for a 50 step restart 1024x1024
lemme run without
vs 60 without
so generally, a 2x speedup
very worth imho
32 seconds
If you always use the same model/lora/resolution combo it seems nice, but if you play around a lot its mid
hmm... this. i don't like -> same prompt seed -> different outcome
perhaps region prompting not working as nice as i hoped
lets turn that one off
I mean thats obvious, since its different approach isnt it?

ah. region prompter dies
you know i'm prompter only 😄
using only a single model... so uhh.. yeah, for me it's pretty good ^^
rookie xD
58sec rendering time. for me it works quite well ^^
you need new engines to run different batch sizes ?
for each variation of prompt length/model/resolution you need a new engine
:<
can you pad everything to a certain length ?
yeah, but profile selection goes wonky, not sure why
basically -> having multiple engines so it selects the proper engine for prompt size
@native halo time to pray for the pumpkin god
@wispy canopy look at my image
it only took 11 minutes 
dpm adaptive just keeps going until it detects no changes anymore
it's pretty good for upscale
What models are yall using these days btw, gotta looks into whats good for animatediff
by that you mean hires ?
yes
Every smol sus model here

I'm back to using HolokukiV2 and GishikiV1.1
i wonder why 
gishiki is very good 
i'm on brightprotonuke, my own sdxl mix
So lots of custom models 
literally that
im not gonna link that since there's nsfw image on preview
slight horror

who puts their password as a models name
also its 5gb
wtf
"incoming kiss" is in prompt 
huh
wait

now put fangs
and vampire costume
also 
the outfit got really weird

I think you need smile
holy 
What ? 
Extremely accessible house
SUPER CURSED, DON'T OPEN
Kek
nah, that's not max cursed by far
this better?
what you see looking for a house online
reality
my model can't make long fangs 😮
she seems interested
time to mix until it can
dude, i can't even get 1.2 working 😢

I can get 38 now 
only missing 5way bosses
I block legion kek
fun mechanic if you only run open maps
bad for stuff which is either rooms or long corridors
or city square lol
why??
legion has incredible bad spawn on city squary -> it's not permitted to spawn near the plaza part, so it always spawns near the edges giving a very bad amount of mobs
This model is very cute
did some blight ravageds earlier... it's pretty cool if you're 100% certain you can win them regardless of tower placement 🙂
sorry, went to eat dinner 
I ate mini burgees
i still have ate nothing
burgee pov @lyric zealot
13:30 pm, and ate nothing yet
this is actually kinda cute 
scanamaies
Muriica
you're using tensorrt?
@wispy canopy lmao
OMG GENIUS!
my brain is mush already I need sleep
what time is it?
white 
9am
what if i say im turning the bot on 
uhh... you just woke up?
@wispy canopy I have one credit left, give me an idea
lmao, this is the first time i watched the whole gif
hamburger explosions?
@wispy canopy definitely the most powerful yu-gi-oh fusion
she wanted someone to feed her so badly that she broke her arms to do that

always better than getting fused with melons
true true 
wtf
yeah, i think that toolkit has automatic pruning enables (it's enabled by defautl in settings)

???
imagine
you're playing your little game
all alone
and start to hear her on your ears

not cursed enough
wtf
@lyric zealot
don't prompt many hands on dalle
It really does it
Perfect chair for gaming ?
those are the dudes/dudettes that beat you in games
How can ben win every event ? This is how.

cute picture to aliviate the cursed

i should not be allowed near prompting devices
that's actually sick
eeeeeeeelf
link it if it's sfw

The Rinner is so amazingly powerful.
Also @nova remnant things improved enough at your side?
#🍥|anime message This is amazing to check out.
i can't try those yet, i don't think i can turn off my tft
How does she keep on being so fashionable?
tensor power
zyoom

same
@nova remnant How are you doing that?
Still fighting with 
But getting there 
hi smol sus, i have to ask, what is the key differnce in generating images with 150 steps cause any steps above 50 is not really visible
and some models might distort the image with very high steps
my mind is stuck at wanting to get tensor rt to work
how did you come up with that image? combination of extensions or just prompts?
yep
combo of the prompt style I had saved plus a short text prompt
and trying a merge I made as well as adetailer, cd tuner extension
delinquent street punk in emo style clothes holding up the middle finger, looking at viewer, 1boy, bob cut, (sidelocks:1. 3), tongue out, closed mouth, squatting, torn clothes, piercings, jewelry, tattoos, observatory, finely detailed background, amazingly intricate background, incredibly absurdres, best quality
Negative prompt: lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, artist name, one tail, no extra legs, kkw-extreme-neg
Steps: 75, Sampler: Restart, CFG scale: 7, Seed: 3325340270, Size: 512x768, Model: HelloAnime_furryV_Deliberate_Merge, VAE: counterfeitV25VAE_v25.pt, Clip skip: 2, ADetailer model: face_yolov8n.pt, ADetailer confidence: 0.3, ADetailer dilate/erode: 4, ADetailer mask blur: 4, ADetailer denoising strength: 0.4, ADetailer inpaint only masked: True, ADetailer inpaint padding: 32, ADetailer model 2nd: hand_yolov8n.pt, ADetailer confidence 2nd: 0.3, ADetailer dilate/erode 2nd: 4, ADetailer mask blur 2nd: 4, ADetailer denoising strength 2nd: 0.4, ADetailer inpaint only masked 2nd: True, ADetailer inpaint padding 2nd: 32, ADetailer version: 23.10.1, CDT: "1,1,1.7,1.9,-0.6,0,0,0,1,1,0,10,5", TI hashes: "kkw-Extreme-Neg: 43eba3ed8ad5", Refiner: cyberrealistic_v33 [7a4dbba12f], Refiner switch at: 0.7, Pad conds: True, Version: v1.6.0
im gonna run that too
taking this line delinquent street punk in emo style clothes holding up the middle finger, looking at viewer, 1boy, bob cut, (sidelocks:1. 3), tongue out, closed mouth, squatting, torn clothes, piercings, jewelry, tattoos, observatory, finely detailed background, amazingly intricate background, incredibly absurdres, best quality
got it from sigocha on prompthero
tongue out, closed mouth ?
edited..
delinquent street punk in emo style clothes holding up the middle finger, looking at viewer, 1girl, bob cut, (sidelocks:1. 3), tongue out, squatting, torn clothes, piercings, jewelry, tattoos, observatory, detailed intricate background
their neg prompt that I didn't use : Negative prompt: nipples, (worst quality), (low quality), (normal quality), lowres, normal quality, ((monochrome)), ((grayscale)),extra fingers,fewer fingers,strange fingers,bad hand
i used my own negs
n: easynegative, bad-hands-5, (low quality, worst quality:1.2), (watermark:1.1)
starts to play with the prompt
well that went wrong.. what?
I only added on tokyo street, posing , and changed it to 1girl miku. wtf
hmms and turns off the cd tuner ext
try this
1girl, homeless, watches a television in flames in a toxic wasteland, natural rust and grime detail, rustic sophistication. rain, night,
This is neat.
she's homeless but somehow there's a tv in a wasteland 😜
sadly in flames
a dystopian scene
i know 🙂 i liked the concept of that art
yeah looks cool, could be great concept art for a videogame
yep
:p ... aren't you using adetailer?
yes
i keep the adetailer settings to its default
a beautiful adventurer girl wearing gold jewelry in the streets of a city in the Western Sahara, fluid motion, stunning shading
if you apply high res fix by 1.5x or 2x that would influence the image detail a lot too
fingers got messed up lol
Who's this woman?
she looks dangerous! kinda like it can get you in trouble and sent you to the horny jail 😂 😂 😂
nah she's covered
there are so many of them in the Ai land.....
fortune teller dressed all in black, tarot cards on the table, tarot card in her hand, magical room with pots of plants and light coming through, bookshelf behind her,
She looks well-endowed, does she have a name?
not really, but what do you have in mind, what would you like to call her?
cute
@bitter latch intricately detailed, illustration of a young woman, lofi atmosphere, in her bedroom, a large window, picturesque countryside landscape, aesthetic clothing, serene, contemplative mood,
or maybe...
1girl, in her bedroom, aesthetic clothing, illustration, lofi atmosphere, a large window into picturesque countryside landscape, serene, contemplative mood
novel illustration, 1girl, brown hair, low ponytail, aesthetic clothing, in her bedroom, a large window into picturesque countryside landscape, lofi, serene, contemplative mood
this is nice, is that from Dalle?
yeah, with that same prompt
very nice detail there
Yeah Dalle has ruined me.
illustration, 1girl, brown hair, low ponytail, aesthetic dress, bare back, in her bedroom, rear view, a large window into picturesque countryside landscape, lofi, serene, contemplative mood
That’s true. But that also depends on the sampler, like example Euler a, it will always reintroduce some noise back into generated image, so there’s no convergent point
But that doesn’t really necessarily means higher steps means better quality
It was just really a fun thing to do, when I set my preview period to really small, and being able to see the image slowly takes shape
And it was actually fast enough for that to not be boring
And for other samplers, using high steps can almost guarantee generated image is at convergent point
And, holy wall of text
good point 🙂
by good point i meant for the whole views you shared not just one line
Well, really for the fun of it, using like 60 steps might be enough for most samplers
yes unless there is some need for very large resolution on intensive production level work
kinda off limit for most use
i usually go for 30 steps, and i keep youtube playing for music videos while generating images lol
So far so good, by chance?
some anomalies with the prompt i wrote which is clashing with her and the setting but not by chance
i cant resonate with highly futuristic look missing organic touches, whether its the girls character or the setting
what the difference between having styles in the prompt box and having styles in the menu bar (right side)?
Having it added to your prompt box, you can make changes to it, like increase the weight and stuff
ahh ok sounds logical
i mostly edit them in my csv file and i kinda developed the habit to keep the most frequently used generic ones in the menu bar to right
so basically this
custom made 🙂
i wanted to keep them tidy
@nova remnant ever tried DPM adaptive?
not yet, is it good?
It is good
But it might take a while to finish the image x3
Try it
how good tho?
@lyric zealot go sleep 
I almost don’t use style now, but should probably put in more effort for it
It's a confusing sampler
It does steps until it's satisfied
i will try it out .. give me few mins
i dont use style too often but some of them are interesting to play with
btw how do you prevent cowboy belt and hat when using (cowboy shot) ?
Make one for the pink hair yellow eyes :p
Your model doesn't know what cowboy shot is very well
Try other words
It's more of an anime tag, not a real thing, realistic models do that for me D:
like half body?
Yeah
this are very basic elements and needs to be properly implied into the engine
Try medium shot instead of half body
I think it's better, it refers more to the camera angle than the body
@nova remnant this is not half body tho lol
(half body:1.1), 1girl, contemporary, indoors, large windows, golden hour,
What I said above xd
well yes i often used medium shot but that too isn't perfect
Put close up together
close up, medium shot
Shouldn't need to increase the strength of them
and guess what ... medium shot is not a recognized tag
at least not in the autocomplete in webui
The auto complete is for anime
yeah and im doing anime images
If your model doesn't do cowboy shot well, it's probably not trained with them
xd
Not 100% at least
@nova remnant did you try adaptive
yeah going to now ....
there is only one ... DPM Adaptive
Yes
yes
DPM adaptive is a first-order DPM solver with an adaptive noise schedule. This means that it does not guarantee to finish within the number of sampling steps specified, but it can be more accurate than other samplers, especially for high-resolution images. DPM adaptive is also relatively fast, especially compared to DPM++ 2M Karras.
DPM++ 2M Karras is a second-order DPM solver with the Karras noise scheduler. This makes it more accurate than DPM adaptive, but it is also much slower. DPM++ 2M Karras is also more likely to produce artifacts in the generated images.
ok im trying it now.. rendering in the process
Tell me how many steps it does x3
For me it took so long
you might not belive this
but its slower than 2m karras
by now i'd have completed 4 images with 2m karras
It's not slower, it's doing many steps

oh?
Look in the CMD
its taking too long to make 4 imags and still going
Kek
@wispy canopy hey
How do you use adaptive for hires only?
btw by using DPM adaptive i had 30 steps, high res fix by 2x with 10 steps, and adetailer on
It doesn't care about how many steps you set
It does steps until nothing good happens or something

lol really?
cause im getting the idea the image textures are pretty amazing albeit damn slow
settings -> user interface, it's pretty far down
dpm adaptive does "whatever it takes", it doesn't respect your step setting
i finished as well 
the output isn't 100% what you'd like but textures are very good, and the speed is just too much
@nova remnant adaptive only sounds like a good idea if you're using trt xd
it does not guarantee to finish within a specified number of sampling steps, but it can be more accurate than other samplers,
Would it also do as many steps as it wants in hires?
yes, but because you don't denoise as strongly, it'll converge quicker, it might only need 20 steps, for example
Do you like this model more than dark sushi 2.5d something one?
both are great models
No idea, but maybe putting belt in negative?
i use them both
I should really do for those OC I have
don't use cowboy shot: head and upper body
or medium shot
yeah i went back to using medium shot
Cropped legs

Cut legs, legs out of place

It just sounds like you're removing the legs lmao
check out the author's collection, he/she has decent mods up https://civitai.com/user/aji1/models
Kinda, like just don’t want the legs to be in frame
Or kinda like those promotional poster, where their legs are literally cropped
i dont have a big complaint about legs when medium shot includes them
@nova remnant try thick thighs in your next prompts, try "portrait" see what it gives
use cowgirl shot instead of cowboy shot 
i mean who wouldn't like beautiful legs of beautiful girls
lol
thats a whole another territory
restart seems kinda slower than dpm++ 2m karras
Restart... Restarts
It goes to like 80% and back steps
i love nenenain 
She can hit me full force

so even w/o going into technical detail, dpm++ 2m karras is by far the fastest sampler with decent output quality
Unipc ddim and euler a

i have doubts about their quality next to 2m karras
Do a plot with them :D
Including 2m karras
well i say that from actual outputs i saw
prompt issue
150 steps solves all problems

(yay more updates ||trash|| incoming)
Also youtube's newest thing
blocking you from watching videos if you have adblock

lmao
which one
im using their free version
also i think youtube getting in the way got sorted out
free version
?? of adblock ??
im not getting any messages that asks me to stop adblock anymore
yeah
let me get the ink..
and not getting ads while playing videos
youtube prolly got user backlash by trying to break adblock
Karma is a wonderful mistress. 😊
you should be using ublock origin
i dispise everything about this economic model
afaik all other adblocks may be borderline spyware
i get no ad
adlblock free version works fine
hmm youtube still blocks me
with that one
odd
adguard
I found for ublock origin, you just have to right-click the element and block it
I'm not a regular youtube watcher tho
capitalism
yes and its even more dangerous, billions of money goes into diabetes care, its more profitable to medicate and sell meds than to cure
thats just one example
the system doesn't incentivize to find cure
wyz is here 
run
I did like 600 steps once, on a deforum project but I think after 200 it's more wasting energy and time than nothing
no notable difference after 150-200 steps onwards
maybe some sampler/scheduler combination does "something" but most eyes wont notice it
True, but just wanna do it for the fun of it
Or maybe a always generating image
Where steps are unlimited

can't find the link on huggingface that talked about maximum use case of sampling method, i remember it was 50
Not if you use Euler a and just keep adding noise back 
But well, like I said, more steps doesn’t necessarily means higher quality so

yeah i get what you mean
Chaotic maximalist illustration, iridescent purple and black tight space suit worn by a woman with pastel blue hair, in the style of dark scarlet and light gold, liquid minimalist-core, meticulous design, dark sanguine and pink, velvet, red and gold, ray-tracing
That looks powerful.
What model ? 
Private model
hmm ok this line"in the style of dark scarlet and light gold" keeps confusing it
replaces with in style of Gerald Brom, dark scarlet and light gold accents
oh I had my cyberpunk style active. oops
Randomtora are you on a laptop
yes
newer models maybe
also
looks like my adblock still works in incognito lmao
probably an extension doing something with it, all others are disabled in incognito
ok adguard works in chrome but not in brave, well it does but there's something in brave still blocking and letting youtube know. wierd.. shrgs
: A bright vivid chaotic cyberpunk female, makeup, smooth, mesmerizing otherworldly and enigmatic charm, mysterious, a celestial being, vibrant blue hair, a masterpiece of nature and magic, electrifying big eyes, atlanean world in style of Gerald Brom, dark scarlet and light gold accents , liquid minimalist-core, meticulous design, dark sanguine and pink, velvet, red and gold, ray-tracing
neat style
seems th e merge I made doesn't like the heun method
or maybe it's some other setting
ahh denoising was to low

Level Up your Stable Diffusion Art with these easy img2img techniques, using creative stock photos, photoshop creations and the ControlNet extension in Automatic 1111. Best of all, there's no need for long complicated prompts! It's easy as dragging and dropping! This is the easiest way to get amazing details, control composition and adjust ligh...
the concept shared in that video link above, you can apply it to create some fascinating stuff
was using img2img where i loaded a background then also enabled controlnet with another image of a girl and used lower denoise value to generate the output with a brief prompt description
this was the background pic
that was the background yes, and for controlnet i loaded an image of a girl
if you had her in a cyber bodysuit or something, it'd be like Snake's suit
yeah if i used a background of that sort
a large background and a character in the foreground
by loading the background in img2img section and chracter in CN
does anyone use sillytavern? im having a hard time installing it
also i could just use the background image and write a prompt for a character in it, but no CN
which would come out something like this
1girl, long hair, brown hair, beautiful loose clothing, bare legs, standing on a boardwalk, island resort, ocean side,
its crazy good how the Ai algorithm put everything into perspective... altho i could improvise on that image a bit to make the girl look more prominent
for those that might wanna try TensorRT but failed https://github.com/NVIDIA/Stable-Diffusion-WebUI-TensorRT/issues/27#issuecomment-1767570566
Hello, i have fresh installed using installation method provided and the extension will install, but the tab for TensorRT will not show. *** Error loading script: trt.py Traceback (most recent call...
hi smol sus, thanks for posting it, im about to dig into it. 🙂
well good luck 
GAAY 2
oh?
thanks 🙂 glad you linked it
i was trying to see if there's new update about it

so about clicking the generating engine button do i need to pick any option from the drop down or it doesn't matter initially?
depends how big you wanna gen
they have max px sizes in there
I can't tell what's new about adetailer
yeah .. so i should pick the dynamic 768 perhaps?
you can click on any, and then press advanced settings
i use 576x768 with ddim + linear(nearest exact)
ok thanks doing it now
gives good quality
the new adetailer adds the little face detection square on my preview, before actually doing the inpainting
i thought it bugged
but then realised it's there all the time now when i gen
it's kinda neat feature, useless, but just fun to watch
useful to know if stuff went wrong
I used to output those red-box images
it's nice it just shows you midgen now
i chose the dynamic 768 and clicked on advanced settings, so it lists several min max options, about min width it has 768, but can i lower that to 512 and also for max 1024 size can i increase that to say 1530?
true true
@lyric zealot @stable crane @nova remnant
hipeople
yoyoy
I'm done gaming I'm freee
yup, you can change anything there. What you have selected at the start is the preset, once you clicked advanced settings, then that preset doesn't matter anymore
hellooo 
nice, just making sure i dont break anything lol
We love how fabulous this woman looks.
do i click on force rebuild btw?
i done setting the sizes
ok
Export engine
remember to close all stuff that's taking up your RAM
not vram
RAM
you need that to build the engine
i got an error saying no onnx found,
the message window closed on its own
look at your cmd and the output there
this part?
there should be some details over there, and over at cmd
it's an old generation, but 
eeps
i think one of the ext is conflicting
No ONNX file found. Exporting ONNX...
Disabling attention optimization
ERROR:root:
Traceback (most recent call last):
File "C:\StableDiffusion\stable-diffusion-webui\extensions\Stable-Diffusion-WebUI-TensorRT\exporter.py", line 74, in export_onnx
inputs = modelobj.get_sample_input(
File "C:\StableDiffusion\stable-diffusion-webui\extensions\Stable-Diffusion-WebUI-TensorRT\models.py", line 977, in get_sample_input
latent_height, latent_width = self.check_dims(
File "C:\StableDiffusion\stable-diffusion-webui\extensions\Stable-Diffusion-WebUI-TensorRT\models.py", line 265, in check_dims
and latent_height <= self.max_latent_shape
AssertionError
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "C:\StableDiffusion\stable-diffusion-webui\venv\lib\site-packages\gradio\routes.py", line 488, in run_predict
output = await app.get_blocks().process_api(
wait wait

i was sure i got something like this the first time
but not sure if it's an actual error like yours
or
it was just exporting ONNX
cmd is just nothing?

got those errors on cmd i posted up there
basically this ...
No ONNX file found. Exporting ONNX...
Disabling attention optimization
ERROR:root:
Traceback (most recent call last):
File "C:\StableDiffusion\stable-diffusion-webui\extensions\Stable-Diffusion-WebUI-TensorRT\exporter.py", line 74, in export_onnx
i think nubby had this issue too, i see the predict in my error too, could be conflicting with autocomplete extension
Disable everything

We need less errors and more cute girls
ok ... no error when i choose default preset and click the export, its working
getting files and building in console
It's lottalewds
wow
i was able to generate this using TRT model i have on top, but hires fix crashes it every time,
and its pretty fast
your hires fix ending resolution needs to be included in your engine
how do i edit the TRT model i have active now?
also do i have to make new TRT models for each of the SD models?
yup
Imagine not have 1tb+ for SD
Could have come in handy
i have 500 gig ssd, i cant waste space lol
imagine not having enough vram 
I'll have your bot

so there is no way to edit the TRT model thats currently usable?
I need to make friends with someone who just happens to not need a GPU anymore by chance and I get it
good plan

not that i know of
since it's like building an engine, you can't edit the engine
let me know if you find someone like that, i also wanna be their friend
so here is the thing i noticed ....

Can you disable it
the first time when exporting preset its prolly necessary to choose Default to not get errors
Or still can't

i have no idea how to disable
once after that i can now create dynamic custom models via advanced settings w/o error














