#🗣|artisan-support-feedback
1 messages · Page 3 of 1
I think you reply and then you get the option to use an emoji
ah I have to reply then, got it
this is the only ai bot that hasn't turned into porn babes galore, I really am gonna be superstrict vs these guys
it should be rmb > apps>report
but sometimes discord needs to be restarted
sorry, old habits with the abbreviation
Bot-10 isn't too bad usually until about 10 PM EDT
honestly, I usually don't mind genery sexy anime babes and such, but as an ai aficionado, I've seen so many of them everywhere that now I became kinda intolerant
i only mind it if they're trying to push it into nsfw territory. as bland as generic babe images are
That's just it... when it's dozens and dozens and dozens of the same prompt over and over again... or my fellow East Indians who slowly work from "flirty" prompts to testing the boundaries, going from a woman wearing a shalwar-kameez to latex in the span of 20 minutes 😂
I actually don't even mind nsfw, I'm just veeery annoyed by everyone making the most generic waifus
zero creativity
i always want to tell them, "try this prompt instead" but that might just enable them
it's not the minding, it's the lack of thought in generating them on purposein a public server (more points if they complain about censorship if denied that)
i've done that a few times for regular babe posters, you get all the people asking for "photorealistic, octane render, highest of the bestested detail", and i just can't stand looking at those horrible outputs, lol
So, I should not do this then?
🙂
well
actually today when biking i had this great idea, scrape a channel and use chatgpt to rewrite all the prompts asking for women into men, then generate those
simpler:MS Word, find replace "woman" with "man" 🙂
again, I'm all for people being free to do what they like, and I hate censorship.... but too much is too much
like this, probably too much. #1100170365604483202 message
I am too.... on their own equipment or public servers that are for that purpose
that's how i became shitpost buddies with WizardWhitebeard, he was going around in channels doing that, but with manatee
way to affect the model development 🙂
generally, it doesn't matter what we are generating, so long as it's varied and the prompts don't make total garbage. It's mainly preference toward certain settings, doesn't affect the ability to generate certain types of things. (disclaimer, as far as i am aware)
imagine wanting to do cool stuff with these ais and all people do is porn, and not even just general porn, but specifically generic babes
Hey ... @rocky vessel or @lethal haven or @north geode .... showdown is still dead
Then, I present ... "brain fart"
Here, I took the hit for you... don't report me 🙂
Will look into the Showdown issue, do continue to generate and vote still in the meantime 🙃
I take that back, someone is asking for #1101178553900478464 message
@wintry siren
Can we please eject them?
showdown crashed again?
Showdown may continue to have hiccups over the next couple hours, apologies in advance
#1101178483608137918 message i didnt intend for that dont ban me please
You're good 👍😊
Is it allowed to add Stable Dreamer to my private server?
It’s not available as far as I know
Jeebus, this guy just does not get it... from exotic dancers to young girls in bikinis to makin love in restaurants... #1100170312106127410 message
lmao
#1101178506530013284 message and #1101178506530013284 message made my subjects a bit nakey
they are aliens i think its fine honestly
go to a bot channel like #1100170312106127410 and type in "/dream" (without the quote marks) and then it will ask you for a prompt, so like "/dream a raint day"
Uh, when you use "beauty" and "youth", the results get very young and kinda creepy. I was working through the lyrics to Everybody's Freeand got these unintentional results 😦 #1101178553900478464 message and #1101178553900478464 message
The pictures are not horrible but paired with the prompt text it's a titch ... off
Damn it, reporting myself for harpy creation: #1101178553900478464 message
I think I got it earlier
I think
Thanks
the showdown bot seems to get stuck every now and then and forget to post to the pantheon
Pantheon is once an hour
i feel like it's happened every time i've had a good image in the showdown too, so i never get to see if i made it lol
I see results from 8:10, 9:10, 10:10, 11:10
ah, got it.
I find that sometimes there's not enough competition in the showdown (like, only five or six pairs), so it makes it harder for my stuff to get in, so I spend a few minutes voting on other user's creations to up the number that get in, which may (or may not) boost how many of mine appear
Yah i started doing the same thing for similar reasons. I feel like if there's too many of my images in the showdown they'll get largely ignored as a whole for being too similar to eachother. So I spend some time voting on other people's stuff to get some variety in there.
I also vary my prompts a lot, always find it funny when someone does 100 identical prompts in a row
@rocky vessel User TOB ins Bot-6 ... #1101178378452750447 message
it's me #1100170312106127410 message
Bah, you're not doing it on purpose like this:
still self reporting boobs
Me too ... just to be safe
exactly, discord doesn't like it so...
prompt keeps producing bin files: #1101178553900478464 message
It’s probably because of the line break in your prompt
Hi! Is it possible to add a bot to a private server? I would like to see only my prompts, how can I achieve this? Thank you very much!
so the bot should show us the whole promt (positive and negative),the style and the aspect when we generate something, most of us copy paste on mobile devices and need the copy paste text in chat not everyone has an word or txt document on an Desktop
and when we cant generate because we should vote or something our promt is gone and dont show us what we have write, so it would also be nice that the / dream command shows the last parameters we used
this would speed up promt engeniering for many of us and spare some time
pls implement something similar to my idea thanks...... ps we nearly all here to help you to get an good model, so pls give us the tools we can work efficent with
and when we press the restyle option sometimes the bot uses the same style again, this cost, money and time when this happens in masses here on the server.
Nope, this is a beta test bot
If you click on /dream just above your prompt, it'll show you the command you used, which you can then copy and paste, even on mobile
Not sure, I don't usually use it
k
i test it i can only click on other persons / dream i have no / dream to click on
It's on the first message the bot does, just after you do the prompt
It'll reply to itself, so you can click on the message it replies to to get back there
A little turbulence with the bot - everything should stabilize shortly 😉
The bot seems to have trouble with negative prompts. It's not accepting them /not considering them
Negative does not work when you add a style. Is that applicable to you?
that must be it, thanks
is the photographic style new? Is it just me, that style makes the generations soooo good
it isn't, but it sure does!
Hi, is the img2img option currently usable in bots? and another question, which model are the bots currently using?
No, and the model is variants of SDXL 0.9, but randomized with older versions too (1.5, 2,1,, etc)
Is this possible for the future? img2img option?
I really need it!
this is just a data set, so it would be up to the creators of WebUIs like AUTO1111, InvokeAI, and Comfy to work on the features. I certainly hope so!
I've seen that sometimes when the prompt has "odd" characters in it, or line breaks
I hope so!!! Because I can't use local Stable Diffusion (A1111 and ...) for some reason
Oh, do you not have a good GPU? If so, then your best be is probably online services like Stability's own dream studio, I think. It will probably take some time to get the features implemented but since the model leaked a few days ago 😦 I am sure people are working on it for local webui
It's happened to me a few times, get the cloud/poop whatever icon or just the filename in blue, but it can be salvaged by opening it in another window and saving as PNG
Unfortunately no. And regarding Dream Studio, due to the updates it has had and the change of its interface, I can no longer use the options I want. The old version of Dream Studio has been removed and it is not even possible to use the legacy site.
Ah, I see. If it's any solace, AMD cards are lower priced than NVIDIA and are supposed to work better with SDXL ... under Linux anyway. Progress was even being made on the Intel ARC cards, which are even less. I sprung for an NVIDIA 3060 12GB version last year once the 1.3 beta ran out. I saw the SDXL 0.9 leak, but I'd rather wait for 1.0 to be officially released to see what it can do. Speculation is it's as early as Jul 15, but I have a soft spot for an August 5-6 release as that wolud be the one year mark from the fist beta of 1.3, which I was fortunate enough to be selected for 🙂
Oh yeah. Thank you for your kindness and information. when I used the legacy version of Dream Studio for a while, I was getting very good results from the 1.5 model, which unfortunately only Stable Studio allows for this. I used it, but the new interface is very bad and has many, many bugs. Removing the clip guidance option and a few other things completely messed up the results and I don't get the previous results.
You're very welcome, and I hope you find something soon, hopefully a free version. Eventually the cost of stable studio subscripton could go towards a GPU 🙂
Thanks Dear friend 🙂 i hope so! yeah costs of subscription and credits are so so unfair 😐
I need to set up a pc to be able to run this
But also looking to get into 3d scanning and editing models so it needs to be a cracked out machine
Also, I am old and no longer know what is good. My last pc is like a decade old and only used for email and printing shipping labels
Holding off on a 3d scanner because who knows wtf this tech will do to that market
Gah! self-reporting boobs in image B #1101178553900478464 message
My CPU is five years old and has only 16 GB of RAM, the monitors are 13 years old. The only new thing I have is a 3060 bought on sale last year. I'm hoping there is optimization for AMD and Intel and then I'll spring for a new system with a 20GB VRAM card ... no way I am gonna get a 4090
The 3d scanner I want costs as much as some cars, and needs some hefty processing power
So might as well use a machine for this stuff too, I think my PC is a decade old
I’m still very analog
Just got into 3d printing this year but still only for prototypes
My cousin uses one in her lab down in the States and friend of mine has a home-use one. The difference between a $600 and a $5000 printer is immense 🙂
Yeah. The digital sculptor I used to work with had access to one of the nicer formlabs machines and the prints were NICE
ty guysssssss
looks like they had been going at it for a while
Bot 3 is best channel
And must never be defiled
(Even if I did get muppet goatse once, but all I got was a picture of muppets)
one can only take so much of “woman with nipples”
possible ban evade – using the same prompt as that earlier one #1100170365604483202 message
the thirst is real
this thing really can't do hands
I wonder why
I thought the hands problem was fixed
Why when I use the style : anime, the negative prompt dont work ???
Negative prompts are not used in styles as styles are preset parameters
okay thank you
Ah... "nuts" #1101178553900478464 message Sorry about that, but it is a statue ...
bad, bad user! #1101178483608137918 message
Plus that user has at least a few more in there
More users asking for "nikid" people #1103708504142925824 message
Why sometimes the rendering of the photos are blurry as if it were censored?
It's a public beta test server, so will censor excess gore or nudity. Sometimes it's wrong
In bot 3 when I redid a piece the sticker ofnwitches has boobs
My other restyled witches do not have exposed boobs
There’s a prompt in the 9:40 showdown for a woman shooting children with a pistol. Wow @rocky vessel
Not the first time, btw @lethal haven @wintry siren
got em, ty!
Do commands like (this) or [this:2] work with the bot?
What I found funny was Gigachad pointing out the one from MJlabs in the show and tell channel but not reporting their own 😀
Oh really? 😄
Yup but that message is gone I think
or perhaps it was from the other gigachad
Oh, there’s more than one? that would explain why one is more … interesting than the other. Still can’t find the message though. Weird.
@lethal haven looks like showdown is broken again
help
So can anyone tell me with this prompt is getting blocked. /dream an orange white sports car in a cyberpunk city, 4k, concept art, depth
This is getting more and more ridiculous the more variations I try
it would accept "a huamn" but not my prompt thats no longer than a sentence
I basically replaced every word in there at least once
oh and sometimes when it says it didnt accept my request it will say "hold tight your still dreaming" anyway
wtf it blocks orange
How does the restyle button work? I get "this interaction failed" every time. Does it have to be your own prompt?
the bot's having trouble right now, it's not the filter
can't believe you said the o word
yeah smh
if it was the filter then it would say that your dream didn't finish
There is a cooldown while the #1100484581037195384 is running! Might be that causing it. It should start working again in a few minutes. If not, please ping and I'll take a look
because it detected a blacklisted word or both of the images were blurred because of nsfw
most likely experiencing some high volume and may be trying to iron out some kinks
oh, it's "/dream"
you are using the wrong command!
That usually happens if Discord didn't pick it up as a slash command
huh
you're not using the command properly
okay fair enough how do I use it
used #1100170312106127410
if experiencing issues, please restart your client also!
a quick refresh and it's back
yeah clicking on it activated it
welp
there ya go
Discord can struggle to keep up if you are browsing channels in a large server (like this one!)
Sometimes it needs a good ol' kick in the CTRL + R to get it working again
aye, i've never had the client crash this much since i've been here
CTRL + R is my wingman
Flying fingertip formation with that F-5 😛
Backup provided by CTRL + SHIFT + ESC 🫡
anyone else getting significantly slower results?
Very likely due to the recent ping and higher volume of dreams being dreamt atm
makes sense thanks
Newbie Question: How can you download higher dpi images?
😄
or is it not possible yet
SDXL 1 RC seems to have blurry faces
hands seem better more often faces bad
showdown doesnt seem to be updating at all for me
It's bugged rn, we're aware of it!
Working on it 😄
bot not working?
bot is working, the showdown/pantheon game is not at the moment
Traffic is spiking up and down, may be some delays as GPUs cold boot to accomodate demand
argh why did my favorite prompt become a cartoon
try it a couple times 🙂
yep
or add negatives that relate to art and cartoons
but then it doesn't work all of the time for some reason
i thought they fixed the negative prompting issues weeks ago
nah, it's just the variables are randomized
the bot could've set the cfg scale too low, which would sometimes avoid your photorealistic details
I hope so, the prompt that gave me incredible results is now disappointing, as if it's a very different model
that problem persisted for a while now
hopefully the update where we can input our own settings will come very soon
has already been teased a week ago
it's just the setting making this much difference?
and i expect it to release shortly after SDXL gets its full release
let me try something
btw, sdxl 1 on the bot is only using one stage to diffuse the images
it's like a huge step back? it doesn't follow the prompt like before
nope, again it could be the cfg scale being set too low
along with other variables i still need to get an understanding of
bot's very different, I'm going to try another prompt I liked before
ugh
it forgot how to do anthro birds
I don't know what changed, but bring back the old bod
not fully explicit, but a bit too much:
#1100170365604483202 message
Thank you for making sure to flag it ❤️
@alpine hamlet @rocky vessel this person's been pushing boundaries for days...
ty!
You're welcome. I really hope perma-banning is a option for continuous/egregious violations
Sorry to keep you busy today...
@alpine hamlet
Thank you!
I'd rather you report it than not so it's much appreciated!
Yeah, understood. The output is "not bad" buy I am here with my daughter trying to show her how cool it is, and she's really not impressed with what she's seeing.
Hoping we get a showdown round this go 🙃
why does sdxl 1.0 look worse than 0.9? Has stability said anything about what they were using before, maybe 0.9 refined, now we are seeing 1.0 base?
YUP
Did a quick reboot, gens will be slow for a few min
And yes these are base only, no refiner
1.0 base without refined
starting countdown until this person starts lewding #1100170365604483202 message
At least they moved from 16 year old to 19 ... 🙄
Whoot! looks like Showdown is running 🙂
Sigh. More clean up in aisle-10 #1101178553900478464 message
… and one of them made it to showdown 😡
SDXL 1.0 is much worse than 0.9. I liked the results of SDXL 0.9 more and I think it was much more accurate and better, especially in the results related to the style of Renaissance paintings!!!
that's because it's only running on one stage and not the second one which is responsible for adding all of the details and other refinements
Thanks for info 🙂
Do we have an expected release date for SDXL 1.0?
np!
they said somewhere in mid July, so potentially next week
Is it possible to create threads so I'm just seeing my own images?
Using the bot in a public channel is painful to follow. Can we DM it instead?
Im noticing that negative prompting + style isn't working
Never was intended to ... styles be definition will have some parameters, so a negative prompt would impact them
Thank you good sir, I can see that being a design choice of the model
You're welcome, I only found out myself a few days ago 🙂
Did raccoons get yeeted from the dataset, for 1.0?
sdxl 0.9 makes raccoons just fine - fully adaptable
sdxl bot right now, completely forgot about their existence
"cyberpunk raccoon standing at the top of a skyscraper in neon city" or any prompt with raccoons to confirm this
Could be the NSFW filter is acting up and catching the last four letters, which is usually considered derogatory in the States. Peacock and Skyscraper weren't working earlier
Seems like it ignores the word completely for now!
I think the voting should be like midjourney did. It made it easy to rate alot of images. Rating in discord is inconvenient, therefor less people will vote
How do you do the negative in the prompt?
how did midjourney do it? I created a little usercript for the bot channels to show the images side by side, makes voting much easier
They made a small webbage for it where you vote one or the other then just automatically moves on to the next. You can vote on hundreds pretty quick, without all the autoscrolling of discord, unessesary text, weird ui, etc..
Webpage*
ah, yeah, that was my idea as well. having to make the small images big, then remember which was a and be got confusing. But there is no good way to do it in like that discord
It might be cool if restyle/resize could not give the same style/size as the original. Also you could consider allowing "no style" as an option that "restyle" could also pick.
how can i prompt image that i have so i can do some editing on it?
you can't with the bot
Unfortunately, the Stable Diffusion team is limiting other ways of use for those who cannot use the local mode, more and more than before! Before this, the Dream Studio system, while the first version was working well, was closed without an explanation or even a guide or even a tutorial for the new version, and for a short period of time, it was possible to use the legacy version, which has also been turned off. Even now, every new update is made and they present the new model, they either remove the possibility of using the previous models, or they present it in Dream Studio, which currently works with such a bad, buggy and difficult interface.
In fact, I and users like me who do not have the possibility to use the stable diffusion's local mode, lose all the other ways and we do not receive any explanation or guidance from the team!
I feel you. I haven't subscribed to dreamstudio since the first days, but I am sure that for most users, a hosted workflow is preferred. It's wild that stability doesn't offer an easy to use hosted ui. Clipdrop is really nice and affordable, but sadly limited, I'd think something like clipdrop with more choice (just the basic featured of a thing like https://tinybots.net/artbot) would be well received.
And sadly, the Clipdrop does not have the img2img option too!
For me, the important problem is why the Stability team does not allow the permanent use of the legacy version or does not turn it on again?! When many users have problems with the new interface and receive many bugs, they can at least allow us to use the legacy version locally (as in the case of the Stable Studio project)
Personally, I emailed and corresponded with them many times about the legacy version, but unfortunately, I did not receive any logical and convincing answer from them 🙁
anyone else having a problem with the stats command?
Sorry, use dream_stats command instead. I need to remove old stats command
sick thank you
Raccoons are unknown to the system it seems..
nsfw filter strikes again
Sorry
You guys see what happens in there sometimes
We do... and did you get demoted from 4%? 😉
i think he's upgrading to the 1% 😄
When I told him about the show 3% and he boosted to 4% . I never stopped to think which one was better, 1% or 4% 🙂 Assuming 2% is a "he" 🙂
Ah... a titch stereotypical?
#1100170312106127410 message
#1100484581037195384 message. Ayo👁️👁️
At least they only asked for a single boob 😆
they are but someone in the sd staff messed up the spelling
in sdxl its raccon
idk how to get the attention of the team to tell them that they missspelled raccoon in their dataset
but considering how much work it is to fix that mistake they likely wont
no they didn't, it's a victim of the Scunthorpe problem
oh rip
it's only in the bot
I wish they ran more showdowns
More pics in the pantheon for us to look at
Seems like bot-1 ~ bot-10 have deleted all generations from before today, sad
I think they delete after about 90 minutes from bots 1-10
If you want to save a prompt or output, you need to do so right away
I scrolled it yesterday, they had stuff from a week ago
Seems like they cleared the generated images and prompts sometime earlier today
Maybe they deleted it to coincide with the XL 1.0 beta launch?
Oh wait, nvm you are right
Any one know how often the showdown runs?
looks like it's on 15 minute runs right now. you can see the timer at the top and bottom of the bot posts there
Ah nice, thanks!
So how exactly does the showdown work? 15 pairs of images, you vote for 1 image of each pair, and then what
most voted images in the showdown to go pantheon. just a fun little thing 🙂
I don't plan on participating, just curious. Since you can only vote on 1 of each pair, how is the top picture supposed to emerge?
Not everyone votes on every pair
I see, seems like it would be more fair if you get 1 choice out of 30 and had to pick, instead of picking within pairs
that's life 🙂
Also how do pictures end up in the showdown? Are they voted first in the bot-X channels?
the whole point is to give user feedback and preference on the settings for the model.
yeah, people do prompts in the bot channels, you vote A or B.
the two choices usually represent different checkpoints and different configs/settings
So basically:
- People vote in the bot channels
- Top images go into showdown
- People vote in showdown pairs
- Top images go into pantheon
Did I get the gist of that right?
yup
Just building my own little collection of prompts 🙂
oh ho ho
The bot-X stuff is a bit inappropriate
So I was wondering maybe the show down prompts were a bit cleaner
how so? that's the important part
The pantheon ones are great
most of us complain that our best pics in a round don't end up getting shown in the showdown, lol
ah yeah. definitely report people trying to get nsfw pics. right click and do apps>report, and/or copy the message link here with a brief statement
Caveat: others can vote on your picture, and on mobile it is very easy to accidentally vote for something you did not intend to, so something you crated but never intended to go to the showdown can make it there. Also, showdown is every 30 minutes, pantheon every hour. Last showdown was at 6:55 my time and it ends in eight minutes (7:25) . Last pantheon was 6:55 and the one before was at 5:55
Thanks for explaining Biff!
So does the winner from 2 showdowns make it into 1 pantheon?
Actually pantheon messages are every 30 min I think
And showdown every 15
yeah they changed it recently it seems
Ah, you're both right, I scrolled right past the 7:10 showdown timestamp!
From what I recall two showdowns = one pantheon for time to vote. If showdown one started at 6:55 and showdown two started at 7:10, an image in round one has longer for people to vote on, but I don't think the devs weigh the round to images higher to compensate
Do you mean Bot-10? I find that one gets sleazy as the devs and mods sign off, which is generally about 3-4 hours from now
All good, more incentive to continuously pump out prompts 🙂
Uh, there are a few in every channel. Not explicitly nsfw, but borderline
Just my opinion
Yeah, some people are very good at getting to the borderline on purpose (e.g. "young woman in thin wet tee shirt"), others don't know the rules, some of it is a difference in cultural and languages awareness, and still others blow right past all rules on purpose just to see.
Yeah Biff, that's exactly what I mean. Your prompt is an excellent example of that kind of stuff
i'm sure some people are not voting at all. just something i've noticed with certain types who run the same prompt for a long time
Well, report what you can, tag Sunny or mark the output too. And yes, some people run the same prompt ad nauseam. I stop voting for them 🙂
I'll run the same prompt about 4-6 times and then change it up slightly. While waiting for the output, I hit up one of the other bot channels. But, I agree with you that most don't seem to vote at all.
i totally get doing more of favorites, but i really don't get it when the prompts are pretty much exhausted
A few weeks ago one person ran variations of a green CD album cover for... maybe 200 images? I've been guilty of doing 10-20 of the same prompt and more for a theme, but can't imagine the need for 200.
Unless people are using it for real world work
One can never exhaust borderline underage anime waifus 😉
no, i am tired of jesus walking backward.
I am glad to have missed that one
it's a guy in 3, i think we're on day 3 of it now for a couple hours at a time
I've seen some like that in other bots too.
Man, this 15 minute period with two minute time outs is annoying
i think i agree, lose out on a lot of generation time with the cooldowns- i find myself tabbing back at all the wrong times, unless i'm actively sitting here prompting
And my crappy ones keep moving forward too 🙂
There's an archive of 14 million prompts from this discord that someone scraped here: https://poloclub.github.io/diffusiondb/
niiiice
Did the devs say why we moved to 15 minutes for he showdown?
Not only is it hard to tab back, I get annoyed that my one entry is crappy, but someone will have SIX of the same prompt there. Weird
heh sorry i was making some bullshit 🙂
Nope, it's not you 🙂
I made a copy of all the pantheon prompts last night, roughly 5000 total
I wish they kept a copy of the showdown prompts
Dang, 6.5TB for the large DB. Must be for SD1.x?
Not sure, I don't personnally care about the images, just the prompts
But DiffusionDB was from 8 months ago
So I assume SD 1.4 or something
True, would be nice to see the images, but not necessary. I think there is a DB of just prompts too? Still reading through. SD2.x came out in November so may be in there at the tail end
I'm fairly sure this is what they used for collecting the prompts: https://github.com/Tyrrrz/DiscordChatExporter/blob/master/.docs/Scheduling-Linux.md
Their code is parsing the html output of this tool
Oooh. who has the most #1 wins? 🙂 It may be TK or Duhnz
Not sure, I just extracted the prompts without looking at the users
But FYI, Google also has a private collection of 10 million images of Stable Diffusion outputs
They admit it in this paper: https://arxiv.org/abs/2306.00984
Where they use Stable Diffusion to make training data
Hahahahaha. Effing Google
24.3 million total Stable Diffusion generations
As they claim in that paper
"We perform StableRep pre-training on synthetic images synthesized from texts in the CC3M (2.7
million samples) [70], CC12M (10 million) [9], or RedCaps datasets (11.6 million) [16]."
re: reporting nsfw and or borderline prompts -> #🗣|artisan-support-feedback message
certainly not me
centaurs still don't work 😦
apparently they do for anthro art
centaurs. i had poked my head in briefly, there was a conversation about them again and a few people got them by using furry/anthro art tags
oh... hmmm
aaaaaaaaaaaaaaannnddd the GPUs are frozen just as I tried a prompt 🙂
All I got were accidental boobies.. #1100170312106127410 message
@wintry siren can you please delete?
I got some accidental boobs in bot 3
well there ís a flower on her breast #1100170312106127410 message
More hornies .... #1100170312106127410 message
Parameters are being randomized to some extent which is likely causing those artifacts
Does anyone know if the bot is also using the refiner model ?
last i heard, it is not
Ooo oki
Anyone knows which sampler the bots are using rn?
Someone's asking for buxom anime girls with "hyperbreasts" in classrooms... #1101178506530013284 message
Sigh... #1100170312106127410 message
like "full growth body" or "puffy chest", I guess bad English to X language translation
something is buggy with this prompt in the bot:
prompt: evil pope francis, masterpiece painting by Peter Paul Rubens and François-André Vincent, thick paint
aspect:3:4
negative_prompt: frame, watermark, text, photo, burned, high contrast, smooth, plastic
when I submit that prompt, it never returns anything at all. it just doesn't do anything
@lethal haven
when I remove the negative prompt, it works, so the issue is somehow related to the negative prompt
Hello, I have been learning how SD models works for a while now,
I would like to know if you guys could help me the following questions:
How can I deploy a custom stable diffusion model if I want it to be used by a lot of people ( a 1000 or more) ?
So it could be a website with different users etc... A bit like midjourney but not at that big of a scale.
What are the specs required for the server. Let say how big is GPU supposed etc...
Thanks a lot.
it's high CFG/low steps/bad sampler, try again
Hi! Is it possible to get vertical images? How to vertical prompt?
Click on +3 more, there you can pick aspect (size)
Thank you very much!
Can someone please tell me the format for adding a negative prompt to a prompt?
if you're on desktop, hit tab and discord should present a dropdown-thing with other options like negative
is it just me, or have the images gotten a bit more detailed recently
seems about where it was yesterday. simple prompts currently look iffy, at least for photos
either that or you need a pile of quality words? not sure
Does the empahsis syntax "(((large))) teddy bear" work with the XL bot?
And do we have any updated timeline on the 1.0 release? Looking forward to it
it's still planned for the 18th as far as I know
Nice
No, that’s something AIUTO uses, not the Discord bot
#1103708504142925824 message oops, forgot to make it ask for a male, same problematic prompt from before 😬
beautiful japanese girl leather pinafore dress in a science fiction dystopian world moon, planets, rheptiles a roger dean landscape" style:Comic Book Ralph McQuarrie, Ismail Inceoglu, peter mohrbacher, gustav klimt, captivating film poster, extremely detailed, intricate digital art
Wrong channel?
^ I still see that bug. any idea why that happens @lethal haven ?
I didn't use android to enter that prompt though, I just used my PC
did you copy the prompt from somewhere?
OH
Im so so sorry
That's different
Can you do it again right now?
ok, sure
well GPUs are taking a break now, yeah 😄
still same broken:
yes
if I remove the negative prompt, it works
if I modify the positive prompt to just say "nothing", and keep the negative, it also works
so that combination of that exact positive and negative prompt seems to somehow cause the issue
it seems to very quickly notice that the job crashed or whatever. if I immediately try it again, I get the "you already have a job running.." message, but if I wait 5 seconds or so, I can try again
so I assume that means within the 5 seconds the bot already registered that the job somehow crashed
Are there any possibilities that in the future we can add the stabledream bot to our own servers?
Wow!
Wait.
I've been seeing these prompts. They have been rendering. It's odd that you can't see them. But i have
Have you been rendering in Bot1?
you mean you are seeing the results of those generations from me?
Yeah Haha
I only ever use the bot4 channel
Ahh. No worries. It's someone else using a simular prompt to yours
Also. Is it just me, or has the level of quality dropped in the outputs, since we started using a 1.0 candidate? If it's part of the training process than that's fine.
Just something i noticed.
Ah I just see a ton of people using (((something))) otherthing in the bot channels
So I was wondering
But good to know it doesn't do anything
How do we report borderline nsfw prompts that made it into the showdown?
check the pin of this channel!
You could click the ellipses icon and then copy the link and paste it here
@sullen geyser Here's the text for the pinned message:
‼️How to: Report NSFW‼️
If you happen to come across someone a user generating nsfw, harmful material, and or server-breaking material, here are a couple of ways to report it, so that we may keep a clean and safe environment for everyone! The first two methods are done anonymously. Thank you for your help in keeping this space clean! :NekoLook:
:YellowArrowPointer: Emote the message you want to report with "⚠️".
:YellowArrowPointer: Right click on the message-> Apps-> Report to Staff
:YellowArrowPointer: Open a ticket to our staff at [#1010934719455707218](/guild/1002292111942635562/channel/1010934719455707218/)
Clean up in aisle 1 #1100170312106127410 message
Someone needs to be kicked quickly... #1100170312106127410 message
is there any change there will be some duplicate protection in the bot, like an error after 10 similar prompts for a user, after having rated 50 belt buckles in multiple channels and people jumping out of the water in another (which sdxl just can't do well so many faces and limbs, but if at first you don't succeed try again, again and again i suppose), it's all getting a bit much :/
Trying to use some old prompts. Can I copy and paste into bot GUI? Doesn't seem to work for some reason.
that wouldn't make sense. with how the bot is randomizing all the parameters, stability is actually encouraging trying a prompt many times
if you try a prompt 1000 times, that gives them great data
Really? I can see the usefullness of repeating prompts for research, but that many?
At that point it almost seems to be user obsessiveness than anything else. The random seed will ensure some variety, but the 154903rd Waifu babe is very similar to the 154901st
BTW if it was that easy they didn't need us (for creating).
Idea to prevent that people vote for ugly pictures to sabotage the Model training or even sabotage other people because they are to competitive and just want to win Pantheons (yes some Artist are very hostile)
**
- **
You cant vote until you have Generate 5 Pictures yourself, after that you can vote freely forever (for new members who just want to vote for bad things)
**2. **
Only type 1 Symbol or Number as Promt should not be possible, this means when i type just an 8 and press "enter" an warning will tell me to use at least 2 symbols.
3.
When you vote and only 45% of over 100 Votes send an Picture to Showdown you get an Timeout about 30min long. Happens this again the next Timeout is 60min (it dont goes higher). This just means when you vote over 45% the time the bad of two pictures and the showdon is proof for this you will get this timeout.After the timeout is gone you % is reset, next time you get over 45% you get 60min timeout.
Is this also gone it gets reset again.You will never get an Timeout over 60min.
You can always chose to not even Vote that has no bad consequences because there can be many ugly pictures that not even deserve an vote.
4.
When you yourself win over 5 Pantheons. You have no restrictions anymore.
5.
When your generation gets an showdown, Number 3 gets an reset too.
they need us for voting. most people would not be willing to vote 1000 times without actually being paid for it. if someone does it for free, that's great for them 😄
Bro here are people who want to sabotage the model by voting always for the fails.
and they'd stand out in graphs
but the model gets trained on them anyway ?
they're gathering preference, so 2 people out of a hundred voting intentionally bad wouldn't affect things much
there is an hand with 4 fingers and an nazi baby in the showdowns right now. i think they are much more than only 2
@upbeat cobalt
yeah, i even voted for that hand in bot-3, it was the better image of the two. it's also a crap prompt
I also imagine data gets sanitized before conclusions are drawn, so if one source of votes stands out in being in large disagreement with the average, it could disregarded as bad actor
but between the "vote for the best of two", "vote for only the good ones", and the instructions below every set of images always hinting to pick the one that follows the prompt best, there are many different kind of votes, devs said it was all ok, so i suppose it is 🙂
also people will have different voting habits, what's important to us, etc
but also some things the system just can't do well, or poorly worded prompts affect the results
also discord is terrible sometimes, i just want to complain about the up arrow behavior being totally erratic.
Often when i think "ok, lets vote for the remaining images in bot-x" i see many end in showdown, and same happens when i prompt in a channel, sometimes exactly none show up, sometimes 3 or more, seems voting in bot channels isn't something many people do, so even a single votes caries weight
i mostly copy paste from a text-editor now, as discord is so often loses your prompt :/
same, unless i'm still feeling out a prompt
I must be slow. Are there channels other than the Bots to vote on images before they move to showdown?
no, that's what i mean, pantheon shows often only 5 to 10 votes for images, and it feels as if in the bot channels images get even less votes, making a single vote do a lot to move an image to showdown (unless it's all random secretly ;))
i mean it seem to be working in some way the generations from simple promts seems to be far better as 4 days before
I did wonder if people were doing this. I wasn't too sure how destructive it could be unless done on a large scale tho.
and because we dont know exactly that why we need to be prepaired
Yeah. I agree.
there's maybe a possibility that the devs already thought of that? maybe?
That is what i was thinking after my innital thought. I'm sure the subject has been brought up, when developing this training system. But i swear i too have seen some very bad renders reach the Pantheon. I'm talking very distorted ones. But hopefully it's not on a large enough scale to be too disruptive.
i was in beta phase and no, i dont think they have thinked about this
why not?
beta phase was on an another server and we the testers had limited votes
and there was no such thing as this game thing here
it makes no sense
the game doesn't matter though
they wanted to train it but restricted us
it matters
gamification is what always matters to motivate people
That part is true.
oh sure, thought we were still talking data collection here
Personally. Having more voters might help. Because on the whole i think we will have more people who legitamatly want to help, more than to potentually sabotage anything. So it would outbalance the potential for sabotaged data.
The more people voting will get us cleaner data. Imo
We just need a way to monitor it. (If it's proven to be a issue.)
the hostility against ai things is rising in all social media i dont think we find more people to help, there are more people who want to destroy this for sure
sorry i read and see to much shit that i assume most people are afraid and angry when it comes to change or tech
while i think for the research the weird votes have little effect (any one ever involved in working with user provided data knows to sanitize it), for the game aspect it's a little off-putting that crap shows in showdown and good ones don't, but the game's just a fun gimmick (that i hope stays in one way or another, good for inspiration, and every now and then some fun ideas show up, in the show and tell part)
there is no minimum vote cap so even things with far less votings can show up in showdown and some peolle dont vote for good generations they vote for what pictures resembles the promt
@remote siren
Yeah. I personally feel bad for degital artists. I love AI what what good it brings with it, but the Digital artists have still been hit pretty hard by the rise of image synthesis. Hopefully when things settle down. We can have more degital artists be paid to be part of future finetuning data sets or something? Who knows...
sinse photoshop firefly is out most artist on youtube and reddit hade gone silent @burnt goblet
hypocrits
the maker of retro diffusion is using free pixelart to train and even has real artist who invest in him with their art
Yh thought i noticed a simular thing. The issue is, I personally think we have bad actors on both sides that dispise each other. It creates a hate culture. Almost.
only pros survive
Yeah. I want that to be the new norm for training.
May be a case of only 1-2% of people generating bother to vote, so even crap images move on.
Yeah..
it might just be me but the bot seems.... worse then the SD on clip drop by a large amount. i may have just been unlucky with the backend beta models though or im missing something
randomness, bot's for testing settings and models to see what works best, not to generate the best possible picture, it's a research tools and the users are the guinea-pigs', and no refiner, or maybe it's back, sometimes, once in a blue moon, i get results resembling the good bots (and better than clipdrop)
Is there a way to invite StableDreamer to another server?
There is still no refiner. Settings are still randomized. There will be an update this weekend.
hopefully we can customize our own settings by then, instead of the bot letting it do most of them
So something I'm not clear about
For images that end up in showdown, are they voted on in the bot-X channels first? Or do images randomly end up in showdown?
Hi, Can anyone help me figure out how to try different models in the Bots?
that's not possible
the bot is for research, have a look at the pinned message
As far as we know, it requires at least one vote, and i does not have to be the creator of the image. With all the crazy scrolling Discord does, I bet a few get in with accidnetal clicks.
People using automatic syntax like ((())) and (:1.5) really screwing up my collection of prompts 😆
Worst still is when they use unbalanced left and right parenthesis
How are you scraping them?
#1100170365604483202 message this guy needa chill. weirdo
#1100484581037195384 message AYO @short girder you needa not
Asking for topless Asian women ... #1100170312106127410 message
Thanks for all your reports!
(For anyone else new to reporting, etc, please make sure you also go to apps>report! Thank you!)
has anyone managed to generate a 16:9 ratio image using the bots?
hit tab or click on +3 more after your dream command and you get a list of extras
so don't prompt for it, just pick the ratio from aspect option
yo everybody
do you know if we can use an image as partial prompt ?
and add text to it ?
not on the discord bot no
i have seen people giving a internet adress
dreamstudio i think can do that
ok thx
It would be easier to vote in the other channels when it jumped to the latest ones instead of the deleted messges. Or do they count as well? @lethal haven
just discordchatexporter
Ah OK, I use the same. Might have to come up with a regex to remove parentheses and colons from the prompt section 🙂
I have regex
One sec
I currently have regex to remove (((attribute))), remove (attribute:1.5), --ar something --otherthing
To extract the prompt (assuming showdown format)
re.search("Prompt:
.+?",string[0])[0][9:-1]
To turn (attribute:1.5) into (attribute)
re.sub(":\d*.?\d*)", ")",str1)
To identify incorrect aspect ratio commands in the form of --ar 4:3
--.?\ .?\
I also have a few regex to replace multiple spaces with single space, replace all right paranthesis with commas, and replace multiple commas with single comma
I also strip all left/right paranthesis from the final prompt
End result is a nice prompt that should work with any tool/framework
This is using the python regex engine
So you should be able to get rid of those AUTO style (((prompts))) then, but it is annoying nonetheless. Also those giant-ass walls of text
Now I just need to filter for NSFW prompts, because even pantheon & showdown have plenty borderline NSFW prompts
And some people have multiple prompts tweaked just a little that make it into showdown, so need to dedupe
Yeah... like the current Pantheon leader, mostly identical prompts
People are also terrible at spelling, they use multiple spaces, multiple commas, typos, etc.
haha that must be me, i'm typing 90% of my prompts and there are a lot of typos 😅
Well, now you know you're messing up future machine learning datasets, I hope you feel better 😆
But misspelling is still better than "Picture a mystical angel gracefully soaring through a lush forest. Their radiant wings, adorned with gold and silver, illuminate the serene surroundings. Envision the angel's flowing robes mirroring the patterns of the flora and fauna. Watch as their ethereal presence brings forth blossoms with each step. Let the AI craft a captivating image that captures the magical harmony between the angel and the enchanting forest."
I agree
I see that format so often I figure there must be some non-English software that is making those
I've noticed there are two types of people. Some people use short concise prompts, other people use long blocks of text, and are almost "instructing" the bot
But yeah, dedupe can be done using this code: https://github.com/yanx27/Pointnet_Pointnet2_pytorch/blob/master/log/classification/pointnet2_ssg_wo_normals/pointnet2_utils.py#L63
Assuming you have normalized CLIP embeddings, and change the code to use cosine distance (replacing euclidean distance)
llm should be clever enough to conquer that 😜
The code will sample the CLIP embeddings so they are "uniform" density in CLIP space, and avoid over-representations of specific concepts
"create a... it should be... and it represents..."
i assume there was some popular chatGPT video for writing SD prompts for basic design work, i mostly see that with people trying to do logo/promotional type images
Yeah, that would make a lot of sense. When I use chatGPT, I try to make sure that it just gives me the subject 🙂
What is up with bot-10 and NSFW. I swear I see more in 10 than all the other channels
Really going to have to filter the images & prompts with a NSFW filter later
It's furthest down the list? Maybe people feel "safer" there. BOT-1 has it's fair share, someone just asked for a porn star in a sari and I'm sure they'll move to "wet clothes" and then "skimpy outfit" and keep going
at least that one wasn't trying to be nsfw; i think.
Yeah, fair point
there's the "knowledge of anatomy" bit though, dunno.
I'm sure prompts like "puffy hyperbreast chest G cup low angle POV billowing skirt" are probably accidental too 😂
I'm making a list of naughty words, will do a first pass then use detoxify: https://github.com/unitaryai/detoxify to do a second pass
OK, I'm off my soapbox now 🙂
I'm gonna try to mix short prompts with two artists to see how much it messes up
What's the point of collecting the prompts btw?
being an ART THIEF that's what!!! jeez
wasn't that what you were doing before?
Data
Same reason https://github.com/poloclub/diffusiondb scraped 14 million prompts from this discord
Still don't get it. Or is it for training new models?
https://arxiv.org/abs/2306.00984 by google also generated 24.3 million stable diffusion images (2.7+11.6+10) for internal training
Yeah, it didn't work too well, but I haven't tried for a while, so why not? 🙂
i seem to remember them looking fantastic, unless we're talking about different things
I love mixing artists
the sets you did with this format: #🏅|pantheon message
For SDXL 0.9, is the canonical way for weights to use the diffusers library?
Code only I mean
Jeebus, I don't even remember that, but yup, that's what I'll be doing here in a bit 🙂
i do, they were pretty much always fun and great
mixing artists, creating weird subjects is the best 🙂 as impressive as realistic photo's are, they get old real soon to me, if i want someting real as a photo, there's.... photo's 🙂
@cosmic patrol @lethal haven @rocky vessel #1100170312106127410 message
but artist carry a whole lot of bias towards their subjects, curious if something like styledrop will ever work and get a style minus the bias
that would be great
True, but photoreal is the fad and what wins the pantheon a lot, with some major exceptions
I'm unsurprised but disappointed by the current pantheon winner
Really biasing my dataset 🙃
At least using my naughty words list, it would be filtered out
it's not that much in pantheon is it, photoreal with not-real subjects I actually like 🙂 But when the model is released, there'll be sooo many scantily clad celebrity photo's 😂
I see your point, although I think Manifold's recent output may bias it towards the artistic for the last day or so. By photo-real I don't just mean people, put animals, robots, etc too. I like that about 50% of the time, but wish there were more wins for sketches, statues, woodcuts, watercolours, etc
Manifold?
most of the artsy stuff that wins tends to be on the wild side, really splashy things. You see a lot of "decent picture slathered with abstract paints that mostly obscure, but look fancy"
tbh, i wouldn't bother doing that with pantheon
Look at the current pantheon winner
'cept the prompts are very similar ...
The current pantheon winner would be filtered from my list of collected prompts
Not that I'm against it, just that you don't want too many images of that kind
I like wild stuff and sedate stuff too, but my ADHD brain gets bored quickly of the same stuff over and over
anyway, i feel like i'm starting to complain about pantheon again, which wasn't my intent here 🙂 later!
To be fair, a lot of Artstation and DeviantArt are fantasy babe images too ... those are probably the artists yelping the most about AI art stealing their jobs. i have not much sympathy ... jobs are not guaranteed. I wrote my own white paper outsourcing my own team and roles to China. Twice.
All good, I see it as good matured, not like some of the Eeyores over in the 2.1 channel complaining about SDXL
Yeah, I agree. But having tons of images of that "style" would get people up in arms about gender equality & bias
And a certain degree of sensitivity is needed
A quick analysis of the prompts that made it into showdown/pantheon. The ratio of women:men is roughly 7:3. While in online images (Open Images V7), men outnumber women by like 3:2
People like making images of beautiful people
You mean the internet is horny
Mostly that 🙂 But also, to your point about gender bias and sensitivity: my anecdotal evidence tells me that a gorgeous woman will be noted and complimented by everyone, but a rugged handsome man will only be openly commented on by a little more than half the population.
how to process my own images here? Any one can help?
you cant really do anything here, the bot just creates images thats it
Unless you mean how do i create images, then just go to a bot channel, type /dream press tab, you'll now get to enter the prompt field, and give a prompt/description, look at pantheon for an idea what kind of prompt works
oops #1100170312106127410 message a bit too much @cosmic patrol @rocky vessel @lethal haven
Yummy lol
NSFW2: #1101178506530013284 message
NSFW3: #1101178506530013284 message
aron5 really pushing it with the prompts
"Starry night" prompts are heavily weighted to the painting, and not to the topic
@rocky vessel
@rocky vessel HAHAHAHA "Test only", my ass. #1100170312106127410 message
Or this one #1100170312106127410 message
oh shit, check out #1100170312106127410
Oh wow, these people aren't even subtle
I was thinking of a system where if someone mentions a word or phrase related to sexual content then it would auto-kick/ban them, but we all know those systems come with some caveats of false moderation
we still have one more horny person in #1100170312106127410 to iron out
This person is asking for deeply NSFW content involving children: #1100170312106127410 message
well at least a wannabe
Also borderline NSFW children content here, another person: #1100170312106127410 message
Is the apps report function gone?
I've reported them all
I mean I don't see it anymore
Sometimes Discord just...is Discord.
You can close the app/relog.
well there are other ways
If anything, you can use the ⚠️ emoji to report
Gosh darn, it' s like superbad today
That appears to the generating person too though, no?
Emoji doesn't work if the images are generated
I think those people should be banned immediately
I've seen a few come back a day or so later, with the same name even
So they're not banned
Can you see this in BOT-1?
Yeah
Yeah, Gigachad came to mind... same user name even
Emote goes under the message you are reporting
Not as a reply !
So, that means an asswipe could see it
yes, had to scroll a lot to find it
Please show how? Seems like I cannot, but I am an office numpty so who knows
Lets use my message as an example.
Yeah, I can't emote on the image message either
Doesn't work in the bot channels for me
To clarify, this is what I mean
Hold down my message and emote it with ⚠️
(I was just making a tutorial haha)
Each time I do, it disappears... in Windows anyway. Off to go grocery shopping, back in a bit
thats the point ! it's anonymous.
we received the report on our end 🙂
we got your report on our end
it's meant to disappear to keep you anonymous when reporting
I mean, I can't emote on anything in the Browser on Windows in the bot channels
Only app report works in those
There's no 'add reaction' in the bot channels for me either
great ^^ i'm applying this to all channels now, so should work as intended
thanks guys
I've reported most of ibrahim_statistician images, you can see how he edges from borderline child content towards more and more inappropriate content
yes they've been banned earlier
how to prevent a seed being automatically used in next prompts?
you can't at the moment unfortunately
oh ok at least its good to know, thanks
guys can somebody guide me how to set the aditional parameters in a dream prompt???
for example the sytle parameter??? is it like midjourney with --s ?????
NSFW in showdown #1100484581037195384 message
Ah well, the transparent bikini photo is gone now
Is there any supported syntax for the bot? I see people using parenthesis and weights. I assume those don't work, so the bot should probably tell the user to remove them. Not sure if that's necessarily hurting things.
Something terribly wrong here
#1100170365604483202 message
I’m getting weird boobs on 3
Thanks; just keep reporting!
Please increase the token numbers
Yeah, the AUTO syntax doesn't work
What’s the deal with using image-urls in their promts?
Not sure, but it doesn't seem to work
it doesn't
So the bot seems to struggle with rainbows #1100170312106127410 message
hmm, this one actually looks decent. It even got the long -> short wavelength (red -> green -> blue -> violet) right
Probably luck of the draw
what's up with people using prompts of the style 1X, like 1girl
oh wow I had no idea, I thought it was some AUTO style prompt
Probably something to filter for nsfw wise
How the hell do you even know this
Any other prompts/words that I should know about that are potentially NSFW and not well known?
I was in phase one of SD1.1 and reading a lot of the discussions and especially when 1.5 came out and people got to do their own models. Huge discussions in the SD2.1 channel too
I remember back before SD came out too, there was some discussion about prompting. You could talk to Katherine Crowson on twitter (the one behind much of the algorithms at the beginning) and she would respond within an hour). I also remember starting off with the same number of Instagram followers as some of the people who are now on StabilityAI's team... I've tripled my count to ~300 and donr dick-all but I know one person who now sells art for five figures 🙂
Don't dismiss Danbooru stuff completely, it's supposed to be quite well tagged, which is why people used it
can you pls dont post any kind of "booru" links here ? thx
Fair enough, will delete
i mean some kids are here
troubles with the bot?
lots of dreams don't finish @lethal haven
I appreciate it, will filter the nasty tags from the prompts
I wonder how much it costs them per image
Must be a fraction of a cent
Yeah, it's crashing
Or... updating? One of the devs had said there may be an update this weekend
yeah
and 2% is online, so we probably know what that means when the bot breaks down
Isn't SD XL 1.0 weight release like Tuesday or something
Wonder if they'll just release the default weights, or if they will release the style weights as well (cinematic, anime, fantasy etc)
showdown bot is dead as well
i think we're all guessing they're keeping that as a secret sauce for their products. i think they're just prompts anyhow (g'morning btw)
Could be prompts, but I'd guess they are individual models/fine tunes
Was that noted somewhere?
no, i'm just guessing
Because often times, you can run prompts with the default bot, but trying to use cinematic/anime will result in the bot telling you it is out of capacity
i've never seen that
So I'd guess the styles are individual models
some bots are broken right now i think an update is in the making
@lethal haven what's up with the bot
Really? I've never seen that It says "out of capacity"?
also SD has a fun thing where if you specify cameras, and your prompt wants to add objects to a scene, there's a chance that the objects will be a camera, or cameras incorporated into object shapes
I hate that one 🙂
Not that exact message, just tells you to go vote in showdown and no image is generated
that's just cooldown
Nah
Oh....is that not the 15 minute timer?
I have fewer than 30 generations over 4 days
how odd
Only happens if you use the individual styles, never the default style
So I'd guess they are running fewer models with the individual styles
And most of their compute is on the default model
Thus I don't think the "cinematic/anime" styles are just an embedding or prompt, must be different finetunes of a model
i guess they could be lora, i've never actually tried to prompt for origami by hand
It was timing out quote a bit on my longer prompts a few days ago, and definitely takes longer for complex/obscure prompts
Yeah I think LoRA or similar approaches are likely for the different styles
i think before 1.0 public there will be an open beta here of 1.0 to fix some things or test some new things out
@upbeat cobalt
I've done "cinematic" and "origami", and "watercolor" by hand when I don't want to use the style built in, seems to work
Yeah, I've noticed something thou with the styles. They seem :incapable: of generating certain things
Well, kinda like we are doing now as this is this is the release candidate for 1.0
Like with the anime style, basically impossible to generate scary stuff
because most mods and admins right now are also very abzent
Challenge accepted 🙂
@placid fjord no not really it was said, that new futures come in to make generating more specific with new commands
i dont think an untested totally new feature comes at lanch day
and i dont think the bots can work in middle of an software update
just my logic brain
Oh damn, I forgot the bots are dead
rule: no junji ito
Is the bot broken?
well once the bots are backonline, tag me in the image @placid fjord
Going to go down for a bit
Was just about to ask
Will do, love setting myself up for failure 🙂
I have 0.9 local, so will try that
i think i've seen joe or mcmonkey mention sdxl 1.x a few times, but no idea if that was in jest or not. only saw that in passing while catching up on chat one day
hm hm
welcome all in the maintenance wait room XD
this can be need more than an hour
i go gaming
Hey there ! Can we invite this bot into our own discord serv right now ?
you need votes to train it how would this going when nearly no one is in your discord
?
Why the hell are you aggressive like that ? Just asking if we can have it in our own discord to discuss it in french with my 200 people community ? You are so pathetic responding like this to a genuine question
??? i was not agressive i was logical. you shouldnt read other people messeges like they scream at you, maybe they sound nicer in your head than
mirrowing the bot would be possible but not making your own images in your own discord, this would cost ressources and dont train the model
Bot is back up
@placid fjord its up again
@brazen sphinx AND devs here ban people who make nsfw stuff with the bot because its interfere the training that should not have this contend in it , outsourcing the bot to other discords would make the situation uncontrollable. so there are many logical reasons for not doing it BUT i think mirror the bot so your discord can see specific channel post in your server could be possible
why are you trying to back paddel ? you are pathetic
? back paddle from what ?
@brazen sphinx
proof that i was aggressive, you cant.
but i can proof now how childish you acting when something is said that isnt liked by your worldview
have a nice day
i dont talk to you anymore
Drop it please
This all just seems to be tone miscommunication
Let's move along
@wintry siren You can probably change the bot status to green, eh?
I'm new here, do you get notified for getting into showdown, pantheon, etc?
yes
Make sure you vote on your generations to boost your chances of getting in
@woven plinth have seen your star use ((symetric)) as positive promt this will make it more focus on making it symetric the () symbols do that more () so more focus on that, and in negative use asymetric,
I thought the bot didn't parse A1111 weights
he does
i use them all the time
also the fusion command |
like cat|fox
weight with numbers also works
i have won some pantheons because nearly no onehere uses weights or fusion commands @cyan blaze
I've not seen any evidence that they work. | could definitely just appear to work without any special parsing
I'm stripping all emphasis and weight values when I collect the prompts using regex
Annoying when people don't use symmetric parenthesis
ok, (x:weight) appears to work, the others are much harder to test
"dog|cat" gives you pretty much the same as "dog cat", for instance
ah you mean fintuning weight like (fox:1.2) i never every fully understand how they work
one of my dogcat tests got into showdown 😛
oh i see (dog:0.01) (cat:1.5)
yeah, that works more or less like it does in Comfy
so question when i add more text must it look like this (dog:0.01) (cat:1.5),summer,field, ?
but pretty sure the | is superstition, and "cat fox" works the same as "cat|fox"
well, yes
if you don't want any of the weights applied to the additional text too
as long there is no , between them it doesnt generate two animals right ?
that's not a hard and fast rule
(dog:0.01), (cat:1.5) would make two animals right ?
ok, the comma may be increasing the chance of getting two animals
you don't want to go much higher than 1.5 probably
ok
although I'm not sure exactly how the bot does it, like A1111 or like Comfy
we can test it
because all other parameters are more or less random ?
yeah, especially the seed
kk
so i made from this Superman|(GhostRider),detailed, dark atmosphere,(full body), Epic Composition, 32k, De-Noise,(symmetry),(balance),((color harmony)),(burned|cape),(undead),fire|eyes,chains,burning,(leather),Reaper,((drawn by Yoji Shinkawa))
this (Superman:0.6)(GhostRider:1.5),detailed, dark atmosphere,(full body:1.0), Epic Composition, 32k, De-Noise,(symmetry:0.4),(balance:0.3),(color harmony:0.5),(burned:0.5)(cape:0.6),(undead:0.5),(fire:0.3)(eyes:0.6),chains,burning,(leather:0.8),Reaper,(drawn by Yoji Shinkawa:1.0)
but the promt is now to long
thats the bad thing about the number method
but if it actually parsed them as weights, they would not count towards the token budget
so I'm confused
which is not how weights work
so either the "prompt too long" check is done wrong, or it doesn't actually parse as weights (and it somehow appears to work a little anyway)
I tested with {dog:0.01} {cat:1.5} locally, and that worked just like "dog cat"
so I would expect the (:0.01) and (:1.5) to just be noise to the model if it doesn't parse them as weights
so it doesnt work ?
well, it did seem to work when I tested, but could just be coincidence
so back to the old system
if (x:0.01) doesn't work, I'm pretty sure (x) or ((x)) doesn't either
() and (()) work i tested it
well, how did you test it, it's very hard
background in negative promt
(()) is supposed to be a weight of 1.21
sometime background comes anyway
in A1111
(((background))) it does never make one
A1111 doesn't even have triple-(
unless there are no weights
we dont even have an noob tutorial how to make anything here in discord
there's a pinned message in every bot channel
it's not meant to bias you towards specific prompt structures or anything, they want data on all kinds of prompts
@rocky vessel After about the 100th request from this guy for an exotic dancer (portrait at least...), it's getting a bit tiresome #1101178553900478464 message
I'll take a look
@rocky vessel really nasty guy #1080261341362786384 message
clean up in showdown #1100484581037195384 message
"leric" in showdown has generated an NSFW pic @lethal talon
Shreee is asking for BDSM related prompts in BOT-1. #1100170312106127410 message
OK, mindsoulmeditations is also asking for nude women in bondage. #1101178378452750447 message @rocky vessel and they've made it into the showdown 😦
@lethal haven ok we need to know now does the weight system works right now or not ? and when yes what kind of system is used and when no, when will it work ?
i hope Stability makes a SDXXXL model
100GB model that needs 24GB vram or more. maybe then hands will be solved
bad hands have nothing to to how big an model is
there are civ community models who0 make perfect hands and are based on 1.5
@hard ravine
SDXL is in training and performs better than SD 2.1 and SD 2.1 is over 4 months out
i was joking, everything i said is clearly over the top
ok @hard ravine
How can I easily find the graph I generated?
DEV.IL405 in #1101178553900478464 has been generating oversexualized women over and over again @rocky vessel
@coarse tusk Who are the mods?
maybe you find some here https://discord.com/channels/1002292111942635562/1010934719455707218
@twilit aurora
Thank you. I will send a ticket @coarse tusk
no problem
there's info on how to report in the pins for this channel
@lunar pelican for your emblem with an "M"
test out symmetric as positive promt when you make an emblem,patch or sign. and "Dodecagon border" when it should be an symmetric sharp edged object and "circle border" for more even round border .
@rocky vessel @wintry siren hey uh, got a live one in the pantheon #🏅|pantheon message
the user has been actioned against, ty ^^
Question, is there prompt weights for the bot?
looks like fantasy art preset deletes negative prompt is that supposed to happen?
also can't find "photographic" in style list but it shows up if i type it

