#๐๏ฝsd3
1 messages ยท Page 57 of 1
My above example but deleted the CLIP-L and -G
SD3 research paper...
Well, that's a weird monster ๐ ๐
I did not ... I let er rip ๐
but if you read your L and G they pretty much contain similar info to T5 prompt but more summarized
I mean if the machines can't help me, and I have to help [it] what's the point? LOLOL
I'm not seeing where L and G have their own dedicated words , in that prompt example really

bottom right one looks like MJ
Try again 
Always think of it this way... in the end, it is a stream of tokens going into the KSampler...
All you;re doing is scrambling it ...
So, the different encoders SCRAMBLE the bits in very different but CRITICAL ways... It is not TOO random... it is more controlled to get to the trained data to output what you want.
It is not trivial to just leave this or that out... you [can] but you should understand the WHY of it all before casting yopur net into the water ๐
hope that made sense.
Damn that is nearly identical... to my eyes.
right, so the theory is, the information may become too complex/scrambled for L and G to properly understand if its all combined?
its like we tell a group of kids we want an image but one kid is 18 (T5) and the other two are 6 years old ๐
and they all work on the image together, we can't give the 6 year olds any details cause they don't know em ๐
thats kinda the active model my brain has been using unironically for understanding and explaining to the gpt how to prompt for them
one is in charge of the character in the middle and he sketches something, the other sketches the background and the last takes that image idea and draws it from the ground up in detail
can you upload images of the generate it made and ask it analyze them and reiterate the prompt?
yeah ive had some of that as well.
ill give mine a couple iteration passes.
overall it seems to improve the quailty for mine
an interation of your from you gpt
and mine
id say atm i think yours is doing better over all with exact same inputs
mines not bad by any means but it lost its colors
yeah mines like an hour of me explaining to the builder the syntax of how to use a comma and a semicolon
XD
Joker corrupted Batman 
ill have to steal some ideas from your system prompt to enhance mine at somepoint. im heading off to bed
@scenic shadow has anyone told you that your afro is reminiscent of bob ross'?
the quality of the images using this 3-part text system really does show much better it performs
great work! Hope the now available finetuning will show nice results as well and we will have a new great model
Might be down to the A:R, seed, Ksampler settings ๐คทโโ๏ธ
Unless each gen from the GPT is that different each time?
now if only CivitAI could get an answer on the license so they could unban SD3 we'd have some cool finetune models to download on there, SD3 feels kinda stuck in the mud w/o CivitAI's support imo
By all the information we got so far it sounds like something is going on. Also new CEO will either push heavily in the right direction or the opposite. I think there will be clarity very soon!
its the ppl who make the models,before civitai we had hugginface and torrents to share models but a lot of ppl were training them because of 1.5 license
i tried getting on huggingface to find random sd3 finetunes but it's not the same, like civitai's presentation of models with sample images and community feedback just makes it much more pleasent to browse potential downloads
huggingface i just type sd3 in the search and i get a bunch of random search results and there's only text to read nothing visual, its just not the same w/o civit, well need that clarity soon if we want SD3 to really thrive with the community imo
I don't ... It will be a year or more before there is clarity from SAI.
Their track record shows this to be the case.
First off, they have no official P.R. department constantly engaged. This let's the public go wild with ideas that are likely wrong but the ONLY voice. So rumors will become reality ๐คทโโ๏ธ
thats true you can try with shakker.ai you can find SD3 there,but theres not a lot of finetunes because not a lot of ppl are training due to license
Sadly, we re talking in big corpo time scale here. Moreover it would nt help anyone if they released an imprecise and rushed version of their license just to calm the drama. I m pretty sure nobody (including them and some sai devs here already acknowledged it) is happy about the situation. So let s give it time. Note that most of the complains "only" impact professional usage, you can still tinker with sd3 on your local install.
civitai was popping off with loras and finetunes, we had a good 20+ loras and finetunes before civitai brought down the hammer and that was within like thes span of the first 4 or 5 days of it's release
agreed
you can still find those loras at shakker
ive never been to that site ill def check it out thx for the tip, I'm messing with cody's prompt generator, pretty cool stuff
I think a glif using Cody's prompt + chatgpt or claude would be pretty cool, or a Comfy node where you can connect it to chatgpt or claude to fill in the 3 boxes for you would be pretty sweet too
the first two is the same text all in the 3 boxes, the last 2 is the text broken up into it's own box. the ones on the right seem to have a better color range again vs putting it all in the same text looks flatter more muted (fixed seed for all)
@sage burrow SD3 + glif won't see more features
I tried making a custom glif with the triple text encoder input and turns out it's not supported and won't be until SAI figures their stuff out with licensing
Chat gpt is behind and is over filtered
switch over to Claude, seriously.
i call it the brand wars
It's like android vs iphone, like just use what you like, it's not that serious
which one is the android here though
ugly monster
go away
which ever one your parents have even heard of in this case chatgpt
i think we should switch to Mac, i heard its better for digital art
excuse me can you talk louder i cant hear you im using my macbook at my local starbucks and its too noisy

anime art, Grotesque Pikachu creature, taking a selfie, burning football stadium, sky filled with alien ships, eerie green lighting, chaotic crowd fleeing, otherworldly creatures descending, intense smoke everywhere, neon electric bolts, unsettling facial expression
once you use claude for a bit, you might change your mind ...
he needs to be wearing glasses,a fedora or a trilby,a peruvian hand made bag,a scarf and browse reddit to be considered a proper macbook user
Are there any finetunes / liras that acualy do something for sd3?
SDXL vs SD3 base models... Same seed, same KSampler settings. Only difference are the source empty latent (from SDXL) for both models and the model shift and negative conditioners for default SD3 workflow.
Interestingly, the SD3 Empty Latent is not compatible with SDXL but the SDXL emty latent is compatible with SD3 ...
It is fitting for sure ๐ itโs like a car and u lift up the hood has all the same parts
Default workflows from SAI? Comfy/?
share your workflow here
All the bots do the same thing just some are more updated than others
idk wht that is i use automatic1111 version 1.9.4 (i'm kinda new to all the AI world so i don't have the vocabulary that should be used)
the main advantage of switching away from GPT is sampling
Gotcha... OK, A1111 should work too by now.
OpenAI are very behind on samplers at this point
look in techsupport channel i'm sharing my struggling in there
why you telling them our secret
the foreground lighting is good
a little cooked, wasnt cooperating
100 steps, CFG 5
150 steps, CFG 2.0
The prompt if anyone wants to play:
A mesmerizing Gothic fantasy portrait rendered in digital painting, showcasing a young woman adorned with a meticulously detailed metallic mask. The mask, with its intricate and ornate designs, envelops her face in cool hues of blue-gray and metallic silver. Her long, wavy hair, cascading in silver-gray waves, adds to her ethereal presence. The backdrop, softly illuminated by warm candlelight, provides a stark contrast, highlighting the focal point: her intense, enigmatic eyes. The composition is masterfully balanced, with a shallow depth of field that draws the viewer's attention to the elaborate details of the mask and the soft, reflective textures of the metal.
That's an overkill ammount of steps but cool.
I completely agree โฆ Gemini and meta are up to date with correct prompting and not filtered out โฆ. Iโm seeing more of the same art from chatgpt no matter how different the prompts are
Funny to think that 150 steps used to be the standard, now 50 is considered overkill
150 were stupid even during the v1-4 days. I'm pretty sure the "20 Steps Euler A" days reigned early and for a long time.
well i never used 20 Steps in both 1.4 and 1.5, all i used was 150 and it was decently fast
hey y'all want to contribute to list of best practices for how to use SD3?
DM me
The first best practice should be "DONT PANIC"
first best practice should be reading license 
Just finetune with sdxl and don't tell anyone
and then, right after - don't panic, yea
try to add some levity an it's all so cereal still
the first page is always don't panic. i guess the reference is too old to get
That's really unfortunate ๐ฆ
How about huggingface?
not sure i just got on the glif discord and asked the devs and they said nope SD3 is staying as-is until they can get clarification on the license stuff
The silence from the new leadership at stabilty is deafening. Don't think the community is going to like the big reveal tbh
Better than Gemini advanced?
best to just consider SD3 medium as non-commercial since it's still in beta, isn't finished, and the actual commercial release is only part of the API right now
we could be having a cool v7 pony model, and the ability to browse and follow trending new models based on SD3 on civitai and novel new ways to create images on glif, and all we need is for SAI to just say something, just officially say something
in my opinion. your milage may very
if you're operating within the $20 self serve license, sd3 is super viable for commercial use
individual creators rather than widely deployed services
glif is an interesting site
You guys hear about Walmart's new approach to security?
Really bummed out that it's Mid monday after the new CEO and investors acquired the product, and they're still radio silent. This doesn't bode well for community prospects. I was expecting something to be said today.
I expect their legal team are swamped
Better than Gemini advanced?
IDK how to prompt Gemini, it doesn't follow me
Radio silence isn't good from leadership. The new CEO isn't the legal team and can make all the time he needs since he's the CEO
do you know how long it takes to get lawyers to sign off on the wording of a contract that's been called into question?
corporate law takes ages
corporate time is different from regular time, haven't you heard of valve time?
probably a good idea to consult the legal team before making decisions that could cause trouble
Where did i say anything about the license?
Radio silence is radio silence. Didn't even say hello to the community.
He's not here to make friends
TBF, Valve isn't exactly a standard corporation. When the founders die though it could go that way ๐ฆ
Leadership changes are meaningful
i'ts only monday. for crying out loud. He just started a new job, doesn't even know his staff, and you want him to put out press releases. you are way too impatient
Gemini advanced is amazing for code help, everyday life help, etc. But SUCKS at image rendering. I mean MJ 2022 level of images
I personally trust Gaben in particular is smarter than to not plan ahead. I do hear that he hand picks board members (iirc) and plans for his son to continue the company after him(?)
TIL the opposite of radio silence is many press releases....
You've gotten worse since i rebuked you in PM. Please stop engaging with me @craggy crest and i'lll try to do the same with you.
Y'all reminding me of a customer I had once (not ai related) who go angry bc I didn't reply within 4hrs on Christmas day ๐คฃ
i have hope too. he's built quite a legacy. it would suck to see it dissolve.
i haven't changed at all. and you ignored the apology you demanded and I tried to provide. i'll stop engaging when you stop acting so negative all the time
first bets practice is to wait for civitai to unban it ๐
oh wait
Daily popcorn time? What ar eu up to now wizard
๐
no popcorn. he tried to gossip in private with me and i had feelings about that and his past treatment of me. ignored him after that because he blamed his actions on everyone else's emotions
puts away the popcorn
pretty boring narcisist shit
Gemini advanced is amazing for code help, everyday life help, etc. But SUCKS at image rendering. I mean MJ 2022 level of images
ah okay I use GPT 4o for that
although GPT 4o hallucinations can be a bit tricky
with Gemini advanced it just needs a slightly different prompting style I guess but I haven't got the hang of it yet
@kindred mica I still don't understand this ban thing everyone speaks of?
Hello ๐๐ป any new advancements or developments?
SDN is out
mebbie some stuff slipped thru the cracks but as i understand sd3 and sd3 related stuff is banned for now
which sux i bet some finetunes would be up by now and my harddrive would start to fill up
i know there ar eother sites but im used to civitai and i trust it
SDN?
Huggingface is pretty awesome ๐
I may have had to scroll through way too many worgliws to find an actual lora lol
Use eulercfg++, that improves SD3 by a long shot
it's just grandstanding by civit. i don't think stability is going to do business with them in the future and they are protecting themselves by dumping them first kind of thing.
you could try telling the truth. i told you we'd figured out what the issue is. that was a technical comment which i figured you might find useful since you were having problems. you then screamed at me, demanded an apology, and ignored the one i provided.
the truth is just boring. like i said. i just don't like you and see no need for private conversation with you.
let it go dude. move on. stop engaging me.

my popcorn sense is tingling
Is this two friends fighting, backstabbing drama ๐ญ ๐๐ผ
i've never known them in any context other than how they interact with me on this server
Where'd all the art go on this channel? ๐ฆ
i ship it
oh look
the guy who blocked me for no reason is fighting someone
๐
144mb model?
Some suspect this doctor might be on the take
its a lora
Where are the 4GB of Vram here? ๐ฎ
That is pretty large for a lora!
girl hanging upside down from rope
beautiful

are you running SD3 on 4gb?
I didn't even try on my 1050 ti lol
I can't even run comfyui
wot
mine is AMD
ooooh I see
I am just using the API
how much memory does the API use up? hopefully you have some left over
4GB AND AMD instead on Nvidia... :/
AND a card from 2018
gives you 24GB of Vram Nvidia

downloadmorevram . com or something
im ready
gimme some too lol
hell yeah free vrams for everyone
Got 2 rtx 3090's, bought them for 3D work, using them for AI, world's too fast man
@bitter hearth this your card?
tell me about it, the community just embraced SDXL when i upgraded and now it sbecoming oboslete
Not obsolete yet, 2 more weeks. after euro 2024 they have time for big announcement
Just waiting for the license to be updated. Any minute now
when hardware will become more powerful we will probably move more towards training personalized checkpoints rather than downloading eveyrthing.
From what I understand if you are not using the commercial license you can do whatever you want...
let me speed up the wait for you
I dream of them everytime I'm refreshing civit for sd3 checkpoints
They wear Emad badges 
Paint is superior
Exactly why I do not use it to generate images so behind on images ๐ but the prompts are up to date and advancedโฆ they work greaat for SD3 model and horror art
What about "A black line drawing of a hand with five elongated fingers. The lines are thin and unshaded. The background is entirely white. The drawing appears to be minimalistic and lacks detailed features or textures. The lines are slightly irregular giving a hand-drawn feel"?
if theres a list f most frequently requested images- cats and boobs ar eup there
Show off, I still like my hand ๐ฆ
that finger is more scary than the monster
wheeeeze
That's not bad actually.
#AwesomeSauce
Wonder why they chose this title image on civitai
Cats w boobs?
Sorry, too much time on DA again lolololol
So what's today's theme? ๐
Anyone pls try generating mermaid in SD3, All my attempts was a failure
what part of the mermaid fails top or bottom?
It just gives blurred images
Whenever there's a word "mermaid"
Couldn't even generate a single pic
enchanting Disney mermaid, vibrant tail scales, intricate seashell patterns, flowing hair with seaweed strands, deep oceanic eyes, sparkling underwater ambiance, soft ambient light rays piercing through the water'n surface, lush coral reef background, detailed texture on skin and fins, expressive face capturing curiosity and wonder, subtle interaction with marine life, cinematic depth of field for focus emphasis, realistic scale and proportion in relation to surrounding sea creatures, dynamic underwater lighting effects mimicking natural aquatic environment,
Wow
Idk why it's blocking for me. Lemme try ur prompt. Mby I should have used word disney
Cinematic lighting, soft glow, underwater ambiance, mermaid tail scales shimmer, vibrant sea colors, intricate patterns, iridescent, human-like features, friendly smile, casual waitress attire blended with marine elements, white apron with seashell buttons, pearl necklace, red hair tied in a loose braid, serving food on antique wooden table, retro diner style, checkered floor, patrons dressed in 1950s fashion, large framed glasses, classic soda fountain machine, realistic textures, high resolution, depth of field, selective focus technique, I guess she's a mermaid we just can't see her tail below the counter
Still gives me blurred pics ๐ญ
The little mermaid fine detail, hdr, ::lora_detail0.1:: in the style of Tim burton Disney
Iโm curious why itโs doing that for u :*(
Mermaid Monday \m/
Ehm?? Tim Burton Disney? Shouldn't it look like this?
LOL i think it's getting it's diseny movies mixed up this is like Pirates of the Mermaids: Jaquelyn Sparrow's Revenge
What you also can do is just to go for a upper body portrait ๐
I am too lazy to go that extra mile to replicate James and the giant peach ๐ but it did get the colors and Disney Pixar style / cgi
๐ jack is a sexy Eric mermaid ๐
eerie atmosphere, Halloween vibe, Tim Burton's signature gothic flair, whimsical yet dark fantasy setting, intricate character design blending Pokรฉmon with Nightmare Before Christmas elements, highly detailed creature features, Victorian architecture, oversized furniture and props for scale, deep shadows contrasting sharp light sources to enhance realism, rich autumnal colors palette, 4K ultra HD resolution, dramatic depth of field emphasizing subject focus
where my pokemon
Not a mermaid...
Gothic Monday Tim burton style ๐
now we cookin
Gotta catch them all! Except this, don't catch this...
not sure if I should poke it or squeeze it
beauty and the beast?
Yes
2beauties1beast
Is SD3 medium fixed yet
Wrong party ๐ฆ
no way these are sd3 2b, how did you even get that image?
why not
Took this image, sent to paint, add birthday party in background. Back to chatgpt and asked for a prompt
looks like we have a non-believer in 2b
seems to be a diffusers issue then, give me a good prompt for sd3 2b
default image
i got this lol
welcome to 2b
thats what you got??
the only one supported in diffusers is euler so far btw
ethereal glow, atmospheric perspective, deep shadows, subtle textures, soft lighting transitions, layered depth, rich color palette, organic gradients, intricate details, high dynamic range, film grain, bokeh effect, natural environment integration,
A young man, about 20 years old, with short chestnut hair and green eyes, takes a selfie in the autumn park. In the background you can see trees with bright yellow and orange leaves. He smiles, wearing a light gray sweater and a blue scarf. There are fallen leaves on the ground in the park, and the soft autumn sun is shining."
2B
i suppose i needed to add default image in the beginning bc without i just got grass
its alright but not impressive imo
just buy one of these:
RTX 5090
looks like it runs hot
nope colder than ever ๐
not bad, what prompt?
Rite I was thinking the same thing
Very detailed portrait in anime style of an adult woman wearing rocker clothes, leather jacket black jeans, extremely detailed face, (very detailed hair), creepy dungeon environment, (mushoku tensei merged with dungeons and dragons), JRPG, art by MSchiffer, pixiv, konachan, high contrast,
It looks cheap
I can't get more than 5 fans on it LOL
I would not waste money on that
amazing prompt 
I rather update the graphs card first
world's best
ok decent i guess
sd3 is first one
1 step hyper sdxl is 2nd one
Upscale now
now it's chopping off good hands
whats the best upscale? the model itself or something else?
wait till you see the 5090Ti, has it's own wheels
lmaoo
Has anyone else played around with the SD3 Large model through the API? I found it to be restrictively censored, far beyond what Midjourney restricts. The prompt wasnโt even close to being nsfwโeven pg13.
I believe itโs a personal preference some love giga pixel topaz, krea , photoshop
What time will you be home your GPU delivery needs to be scheduled
They said itโs for photography landscape thus why itโs like the way it is
I have no issues with it
But others do again personal preference
See thatโs not cheap thatโs ๐ฅ
I tend to make images with human subjects, so it wasnโt really workable for me. That said, SD3 medium has mostly been a ton of fun.
just put clothes on em
gonna need a big pc case for that :3
I tried dressing them like it was winter in Minnesota and it still blurred it!
environmentally friendly 5060 with 4GB vram
share your prompt might be some word you used
Might be the words u are using
quad gardenhose gauge power connectors should be a cinch to install
Iโm fairly sure this is the prompt I usedโฆ
Photo of a man getting tattooed by a female tattoo artist in the arctic during a blizzard, shot in the style of National Geographic. A woman sits in profile on a metal stool in front of a man as she tattoos his forearm with a tattoo gun. She holds the tattoo gun like a pen and wears purple latex gloves. She has a hipster aesthetic and has an intent expression as she focuses on the tattoo. She has long black hair, high cheekbones, tan skin, and light freckles. She wears a white ribbed tank top, loose jean shorts, and purple latex gloves. The man sits on a metal stool in profile and has a buzz cut and wears jeans and timberland boots. They are centered in the frame in the midground of the composition, with the austere frozen tundra in the foreground and background. They are partially obscured by the blizzard, with eddies of snow swirling around the landscape. The horizon stretches into the distance. Hectic yet austere mood, muted colors with a cooler palette.
Granted, perhaps not actually suitably dressed for a Minnesotan winter, but certainly not nsfw
We can so help u
Brb
are you using Cliptextencodesd3 along with eulercfg++?
Iโm using the api through fireworks.ai
Oh sorry, though you were running local with comfyui
I have used the API before a lot, it's pretty restrictive, not much you can do about that
Title: Tattooing in the Arctic Blizzard
Description: A stunning photograph shot in the style of National Geographic, capturing a moment of artistic creation amidst the harsh Arctic environment. A female tattoo artist with a hipster aesthetic tattoos a man's forearm as a blizzard rages around them.
Visual Elements:
- Subjects:
- Tattoo Artist: Female, hipster aesthetic, long black hair, high cheekbones, tan skin, light freckles, intent expression
- Man: Buzz cut, jeans, Timberland boots
- Setting: Arctic tundra, blizzard, snow-covered landscape, austere and frozen environment
- Composition:
- Midground: Tattoo artist and man centered in the frame
- Foreground: Austere frozen tundra
- Background: Blizzard and snow-covered landscape stretching into the distance
- Mood and Atmosphere: Hectic yet austere, muted colors with a cooler palette
Technical Details:
- Camera:
- Model: Nikon D850
- Lens: AF-S NIKKOR 24-70mm f/2.8E ED VR
- Aperture: f/5.6
- Shutter Speed: 1/125s
- ISO: 100
- Lighting: Natural light, overcast and dim due to the blizzard
- Color Profile: Muted, cooler palette with shades of blue and gray
- Rendering: Highly detailed, with a focus on textures and atmospheric effects
Creative Notes:
- Inspiration: National Geographic, Arctic landscapes, tattoo art
- Purpose: To capture a unique moment of artistic creation in an extreme environment
- Mood and Atmosphere: To convey the harsh yet beautiful conditions of the Arctic landscape
If only. Iโm still new and working my way up to that. Iโm not resentful at the restrictions, more just surprised at just how restrictive.
That prompting style won't work for SD3, even for the api version
Itโs not ment to be a prompt but itโs just a key to see what itโs missing to make it a prompt โฆ
Here is a draft of ur vision
Very impressive! Is that with sd3 medium?
These are
No that was meta itโs very up to date with prompt drafts
Last 2 is sd3
Better prompt adherence with meta
I am currently trying Samby's prompt on the SD3 api and all i am getting is blurred images
i am going to experiment a bit with it, to see what causing the filter to go off
National Geographic style ๐ธ, a female tattoo artist with a hipster aesthetic, long black hair, high cheekbones, tan skin, light freckles, intent expression, tattoos a manโs forearm ๐ฉโ๐ค. The man has a buzz cut, wearing jeans and Timberland boots ๐จโ๐ฆฒ๐๐ข. Setting: Arctic tundra, blizzard, snow-covered landscape ๐จ๏ธโ๏ธ. Composition: Tattoo artist and man centered in the frame ๐ฏ, with austere frozen tundra in the foreground ๐๏ธ, and a blizzard and snow-covered landscape stretching into the distance ๐ฌ๏ธโ๏ธ. Natural light, overcast and dim due to the blizzard ๐ฅ๏ธ. Mood: Hectic yet austere, muted colors with a cooler palette ๐จ. โar 16:9 โs 0 โv 6.0 โstyle raw
Try that prompt
Itโs better than chat gpt โฆ
Chat is behind
Itโs too filtered โฆ
So does Gemini and meta and Leonardo ai
this sounds like you decided on something and then started finding evidence for it
archer gave up and picked up a sword, warrior has a shield and a broken arrow, the mage is in a very interesting pose lmao
I brought it to the attention to the owners of chat gpt
gfy
I cancelled my script too with them no more paying for there services when itโs free
They were over charging for api rent and I was forever getting in trouble with them

I was getting data I was not suppose to and got put in time outs there bot is too easy to jail break ๐
API sd3
Meta is weird, it won't depict famous real people at all, but on the other hand it doesn't care about copyrighted fictional characters, and is also very very lenient WRT "suggestive" content
I donโt think these bots are ment for photos but for other useful toolsโฆ
Itโs funny when it does generate images
Meta has an Imagine page that's just the image generator. Like it doesn't talk back to you or have opinions at all outside of a few canned responses.
it can also animate images but that kinda sucks, imo even svd is better, and luma is way way way better
cat fractal 
Yeah the animation isn't great. It has really good prompt adherence for the images for being free and not rate limited though imo. And it happily does "hard realism", unlike anything based on DALLE
thefuck
What you've been sending is "fractal" then
I was confused
Yeah thatโs true
A see through apple on a gold pillow... technically not wrong.
can't be more accurate
I was almost 100% certain it was a see-through strawberry before I checked your prompt
Really it could be a very small John Cena
Lying perfectly with his cute head on the pillow...see-through
Been a little hesitant to try people. But that's not bad at all. Rock Hudson as Superman.
safest model in existence
**Hey all. I'm done with SD3 for now. Good stuff from most of you guys though. Love it.
If you want to keep up with my posts just look in SDXL every once in a while. I'll be posting Cascade images in there.
I'll just be waiting on the model maker community to make a better 2b.
See ya**
ok

๐
upside down cat
hmmm looks like there wasn't any upside down cat heads in the dataset
Probably the model just didn't train enough to get the concept of it even
Quick question, I doubt so but can the t5 model be replaced with another model with the same embedding size?
I was able to fix it using a multi-million dollar R&D program called MS Paint but I would think the AI could've gotten it from the start
Paint is too powerful
seem all of our resident experts are on break at the moment, but I don't think you can easily replace the text encoders after training of the model
show me your belly
I Just discovered strange text in every SDXL image metadata, is that some masonic prayer or SMTH? What language is this?
I could replace it with flan t5 xxl and it worked properly or it seems so
I'm confuse you did it then asked if it was possible?
dalle3 has very little censorship if you access it via a corporate API
No bc, thatโs a fine tune of original t5 xxl, I am saying a different model with same embedding size.
Like ciciai.com/chat does. You can get full nudity and other stuff
i dunno
can't tell if this is a wall or the floor
No worries Mack u and I still can ai battle with Cascade thou โฆ I love Cascade more thou \m/
true true
Rip, youre now part of the cult

Ah, yes. Good old Lorem Ipsum.
Iโll follow u in SDXl
nice
Must be "handy" to have extra hands in the right length and position I guess... ๐
nice
Sense u and I are one of the Cascade originals
The only reason that I keep jumping from base model to base model is becasue the hype that surrounds it before release. That hype normally comes from the company too.
You have a Skillz issue ๐
mine's broken
Hey I was honest about it betta testing before it came outโฆ hands were going to be a issue and I got backlash for it ๐ but turned out I was right but before they destroyed the Cascade server they said sd3 was gonna be for photography model not for anything else and they were right safe model against humans
I had no clue of that
hear me out though, what if you don't want humans
So cuteeeee ๐ฅบ ๐ค๐ค
Then it's the best model available 
perhaps there is a compromise sorta half human half not
Okey bro, I loved most of your images, I hope community finetunes are on the way ๐ฅบ ๐
same
Yeah i loved the Cascade model and server but i never understood the drama within itโฆ and why they destroyed itโฆ but yeah i did ppl were like giving into the hype
A big meow
Yess so much funn
some times it messes up though
What I saw were the cherry picks by stable to promote it.
Im blind af, do anyone got the cascade channel from this discord?
you got a ghost problem
It's just gone I think
๐ฏ percent
Thatโs my watermark ๐
Why ๐ญ there were a lot of amazing gens there
Was it active?
hot as fuck
Itโs gone :*( I have some ss
@odd basalt and I think some others did hear that there was drama from within the channel.
omg now I see why they don't want pony on this I'm getting nipples
Not more active than sd3 but there were a lot of images, I didnยดt post there but just watched other peopleยดs posts ๐ฟ
Iโve made so many and share on deviant art thatโs why I watermark hereโs hott โค๏ธ
It would be so funny to see a mangled watermark that resembles yours when someone uses that feature on the site.
Yeah I did save some of others art and ss of important information that helped improve my art but sucks they destroyed a good server

my condolences to your destroyed server
I would die laughing
oh so you cheated
Beautiful and thanx
FUCKING CHEATER! FUCK THEM!!! FUCKING BITCH!!! lul kittens
If it's inside comfy, isn't cheat 
I don't make the rules, i just follow them
2b is trying to furrytize me
There were other channels that were deleted but they weren't active for a long time, weird that cascade got hit
it's like a funkopop if hot topic made them
It was taken down as api SD3 came out
fucking losers really
yeah
are you saying they did it because the API was coming out?
maybe we ARE in a cult
No. I'm saying that it seemed fucked up. As if they wanted the spotlight as soon as possible.
yeah
Funny because my inspiration is from Y2K 2008 when leopard print was very popular โฆ I am also such a girl and my gothic scene girl fashion shows
how would getting rid of the channels spotlight the API though if anything it would detract from it cause of the controversy
Exactly why they did it cascade barely came out and they wanted SD3 to be their main priority
SAI never really promoted cascade much
I think
I dunno why their eyes so weird cx
They wanted SD3 to be their main goal and cascade everyone quit after they made the free model right when stable xl came out too
They made a large celebration blog post about cascade. It's not even their model actually.
they really don't promote much of anything only when asked it seems
Correct
cat koolaid does that to their ๐๏ธ
I lost count how many ppl came in asking for bot images. ๐ and we troll
I miss SD bots too
I take their prompt and put Shrek in it
I would like to speak to your cult leader please
Only big guys allowed

LOL... the size of the heads... ๐
It's a small car 
But... but... those heads LOLOLOL
base on my research I have concluded the dataset was comprised of 98.6% black cat pictures and 5% other stuff
is that a blank prompt?
Don't forget to MIND YOUR SKILLZ, sir! ๐
how do I do that I can't even remember to type "surfboard"
procutto cheese cake ๐งโ๐ณ
my mouth is wattering just thinking about it
Kittens Log: 3349A::We've arrived at the CatNip nebula, but just as we began our scans I noticed most of the crew acting strange...
๐ ๐ค
these look good.
did a second iteration on my gpt incorporating what you posted
"I need a prompt for a elvish ranger in a magical forest, dont have them have any weapons" too all of the gpts. One shot and seed. Yours, my old, my new.
litterally just copy pasted both of the instructions and pasted them into a new builder and said combine em XD
it looks like yours is with the new system then assuming you pushed the update in the past like 15 mins
it was the instructions from your old set up
but with my added prompting syntax
ive got it still, im still building mine to be tune to my own workflows. most of the time im prefering my out put vs your new one, but thats more because I like full shots instead of close up when making characters
of course, mostly for me its a preference of how to write the G prompt. I still prefer old school way of making sure its separated out with commas and semicolons to get more distinct subjects
Only thing I lift are my fingers after typing these sickass prompts
God do I know that XD
eh thats the builder is for
iterate with it in a chat format
has anyone figured out how to use eulercfg++ sampler with ultimate sd upscaler?
as the sampler that upscaler uses?
correct.
that works as a separate node which you can feed into custom sampler, but ultimate sd upscaler doesn't take input like that.
so i'm doing stage 1 with eulercfg++, but then upscaling with something that's of lesser capability.... so i'm going backwards.
they probably need to update the node so it's an option
yeah i ran into that problem when setting up my first time for my normal sampler. not sure what the solution is to plug it into a node that doesnt natively have it as an option
yeah....
what i've done in the past is upscale the latent then resample over and over at like .33 upsizing increments before ultimate sd upscale came out
yeah unfortunately sd3 needs tiled upscaling or it just bombs
looks like it's pulling the sampler choices from here
and comfy only has eulercfg++ inside nodes_advanced_samplers right now
I've loaded quite a few of images with my past comfyui workflows embedded into them. Suddenly what was producing art masterpieces, yesterday even, is now producing crap like this. I haven't changed anything.... well except some ComfyUI SD3 additions. But they should be irrelevant to 1.5, SDXL and Pony (tried all 3) Has anyone else had this happen?
SD3 is great and all but it won't do what I need it to do for my art sales, so I need to use 1.5 and SDXL still as well. Time to start over and reinstall everything maybe?
Did you update Comfyui?
I loaded a bunch of custom nodes, (SD3) so it may have updated in general as well?
It's the comfyui general update that was my problem
Perhaps I'll read up on how to turm comfy back to the last version....
Are they censored by any chance? That is another oddity suddenly lol
Lol. Nah, but something looks like it's not working correctly if you update comfyui...at least for me
Do you have the pcm lora models?
Models and checkpoints galore lol
Though, tbh, I've installed so many comfy custom nodes ever since SD3, that something was bound to break one day...
Maybe you can test if you have better result on SD3 medium model without updating comfyui and then one time after update?
You can try just to copy over custom nodes folder and models folder. worked for me...but I'm afraid to update comfyui now, so I can't use that new euler++ node ๐ฆ
Just in case comfy update wasn't the cause, I'm now nervious about future comfy updates lolol
If we could blend our styles maybe ๐
if you add a bit of 3D it might work, but it will probably add too much detail of it
don't complain 
try play-doh style {prompt} . sculpture, clay art, centered composition, Claymation

cute!
Hey this is pretty sweet, I could never get SDXL to do this decently
I had to make a LORA for SDXL to do electronics like that
Practicing my inpaint skills in Stable Swarm.
Gen
Just practicing adding things to the scene with inpainting, and what settings affect it
Cool
Using SwarmUI. SD3 Prompt = streamline moderne, grand guignol, aztec, venice masked ball, van gogh, braque, fibonacci, fineliner wimmelbilder
Looks good, went to work.
kind of..
glass giving me some troubles
so far the pen, the bird, the logo on the bottle, the spider, the water glass, and the iPad added with inpainting
You doing inpainting with sd3 or a SDXL model?
I'm using Stable Swarm UI with SD3 and doing the inpainting there. Trying out the inpainting settings to see how I need to adjust things to get it to look correct without blur or seams.
not using comfy nodes for it. Just using Stable Swarms image edit stuff
Might be worth it to used a good fine tuned SDXL model for inpainting. I know that the base 3 model can be kinda jank.
Mind you it looks decent so far
original image
Oh I totally get that. I usually use Fooocus for inpainting because it's godlike for that
It not doing bad so far though, like you said main thing that sticks out is that shot glass. iPad a bit as well.
ya I think it just has a hard time getting the angle I wanted on the ipad
main thing is, I was able to prevent a "blurry halo" and seams
Looks good so far, I think swarm sets a Gaussian noise blur affecting radius of the inpainting area to help with that
by learning how to adjust the mask blur and the "mask shrink /grow" settings, i found you need to play with the shrink/grow settings to get the new object to have a similar quality
Yeah thatโd be the setting, what setting have you found that have been working well?
and it looks like if you make the mask blur large enough (along with using "differential" mode) it will correctly sample the original area and do more of a "content fill" to blend the pictures.
well, surprisingly, by setting values much higher than I would have thought
example:
Mask Shrink/Grow of 75. Mask blur of 30. Seems to make sure it samples enough data around the mask so it can generate a proper match. The vertical window line for example, comes out better if you use a higher mask blur when I was adding the spider. Differential beleive means it samples the orginal image when adding the inpainted object.
You tried using the the other samplers and schedulers
no I hadn't, I use uni_pc right now always since one of the tutorials I watched on Youtube mentioned it. I used to sometimes use heunpp2 but its way slower
Was it a SDXL tut or one specific for sd3?
a SD3 with Stable Swarm specific one. 90 minutes. Actually was good: https://www.youtube.com/watch?v=HKX8_F1Er_w&ab_channel=SECourses
Do not skip any part of this tutorial to master how to use Stable Diffusion 3 (SD3) with the most advanced generative AI open source APP SwarmUI. Automatic1111 SD Web UI or Fooocus are not supporting the #SD3 yet. Therefore, I am starting to make tutorials for SwarmUI as well. #StableSwarmUI is officially developed by the StabilityAI and your m...
SwarmUI - SD3 Prompt = streamline moderne, grand guignol, aztec, venice masked ball, dali, richard dadd, egon schiele, fibonacci, fineliner wimmelbilder, highly detailed eyes
Iโll have to give it a look sometime.
Might try to play with the new sampler for cfg++ to see if it handles inpainting well, will prob have to do it manually
Yeah Iโm at work so feel yah
nice! Have a good day
are there any loras or finetuned models available for sd3?
I just remembered that several days ago we were supposed to get a SAI statement... and it feels like #2Weeks all over again.
The tools for basic finetunes from what I remember just started rolling out. They still need time to train.
Heard thereโs a few Loraโs out and about. Check huggingface.
Looking for a new wallpaper with a bug with transluscent skin that glows in the dark, some wires and stuff. Black background. Anybody has a nice prompt or picture laying around?
they're just onboarding a new ceo right now so things might take a while to get sign offs from upper managment
Well... let's see how that goes...
The CEO is literally [upperest] manager ๐คญ
the tippy top yus
tf
@fossil pagoda
your cracked
@muted cargo
@kindred mica
for god sake.
Lucky you! ๐
Is none of the mods active or what?
Nope
they didnt respond
Sleeping
we have found the window they are not around
none of the devs in the online list are seen talking here, boto either doesnt check
Someone posted cheese pizza here once, and still went unoticed by the mods
i wanna to forget about that incident smh
yeah it took the mods like too much time to ban him
he should've banned instantly, but i guess the mods are sleeping
sure but still you posted porn on this sfw channel
i mean we rarely got those who destroy the rule lol
They really should have mods that live in europe or any other place that's awake when Americans are sleeping
wtf lol
?
it has been 1 year since that guy talked so i wonder if hes still on sai
GG
I think palp just go touch grass anyway

unlike mcmonkey I guess he did not actually have time wasting on Discord SD tech supports or discussion something
I live in europe, but you can not expect people to be around 24/7 unfortunately
Needs more mods ๐คท๐ปโโ๏ธ
Took me around 10 mins to act and you complain 
since when the last time we got mod application
wait that was 10 mins
Not for me to decide
Yellow role people arent mods, Sunny and myself are... I know its a short list
2 mods and a few community guides that can act as well if needed
it is a real miracle if you able to managed 20k pops discord server with 2 mods only without any assisstance ( bots )
A better class of clientele ๐
ohh i thought they are because i saw the purple wizard guy timeout someone
They have access to do so, but usually its left to us
We have automod and a few other things that catch most stuff, but sometimes it slips through
But yeah thanks, were doing our best
yeah its insane thats sum dedication
Definitely. This sub is a cesspool at times.
oh man Simo Ryu the Cat is cooking real hard
I was litterally in a meeting :p. But yeah it s fixed now, thanks.
Just a nutting dick pic, multiple postings of the same.
which one's better? I tried combining the 2-pass technique I did with my workflow with CFG++
what is the prompt for that style kinda
same comparison
I think 2, the one with one waterfall only 
Both good but maybe that's more correct, terrain looks a bit nicer to me in that
The water looks a little more natural in the 2nd one. Looks fake in the 1st one. Same with terrain also
I feel like it's just slightly better
old masters painting style, intricate details, full body view, famous artwork inspired by (james cameron's avatar:1.2) and (otto marseus van schrieck:1.3) and (makoto shinkai:1.3), detailed expressive eyes, fantasy style, (fractal art:0.8), beware this otherworldly glowing vertebrate abomination aptly named "quaint rockdonker" with its countless tentacles, this exotic and deadly poisonous creature roams the bioluminescent shnorbsabblershroom forests and foggy keydering jungles of its alien world "blunderward ii", where gigantic tangflurb and towering trees gorble the ground, lora:lyco-style-envydramaticlightingxl01_digital_art-xl:0.72 lora:midjourney_mimic-xl-by_andrexsel:0.4 lora:more_artful-xl:0.4 lora:extremely_detailed-detail_slider-v2-xl-by_ntc:1.0 lora:envybetterhiresfixxl01-xl:0.0:hr=1.0
it does seem to work nicely with the 2-pass workflow I made though
loras not actually included but i still pasted them
Hola
yeah, I see more of what you mean now. the aesthetic choices CFG++ causes are nearly identical, but it seems to produce better 'shapes' in the images
a lot like what better prompting tends to do sometimes
Sd3/pixart A young adventurer wearing a headband and boots holding glowing blue spiked fruits illuminated by the warm light of a setting sun vibrant, high above jagged rocky cliffs fierce dragon-like creatures with sharp teeth and colorful scales snarling, broken shards of rock and fruit suspended in the air, dramatic clouds swirling in the sky a sense of tension and exhilaration.
Shoreline on the second one is ALOT better in my opinion
I've been testing EulerCFG++, as it seems to work quite well. However, I've been testing it and it generates the same images with CFG*2 in Euler. That is, the first one is EulerCFG++ at CFG 1.0 and the other image is Euler at CFG 2.0. Not a single pixel changes
Am I not using it correctly?
you doing regular or alt
alternative
think alex and comfy were saying that its pretty much all it really did in one of the channels.
though i thought that it was specifically normal that worked like that, it might be both
yea other samplers often output same thing too
double checked its just the normal was outputing the same as like 4x
I would double check but it might of just been a seed that there wouldn't of been a difference ...
its not magic but its an overall improvement in alot of cases
I am generating both at the same time, with the same seed (only in Euler the CFG is *2), and so far all the results are identical...
I just got home ill test on mine rq
Finally, I have found a difference in this result, the finger... ๐คฃ
yeah its not much, first is ++
small highlight on the chest are, a tree is slightly different on the right and the nub on the tree on the left is a bit more defined
So, it does something, but then the comparisons being made are using the same CFG in both Euler and EulerCFG++, right?
Yeah that lines up with my testing yesterday and like Cody said, sometimes you'll see no difference and that's normal if you're using simple pos/neg prompts
Understood! Thanks, everyone
i know the change in the code WASNT MUCH.
but its just a slightly better version of euler
its just a varient sampler ... its just those small changes MOST of the time will be for the better
I didn't mean to criticize, I just saw the comparisons and couldn't understand such a big difference ๐
its still wild west bro no one knows nothing
I'll double check again just to make sure it hasn't all been a big placebo troll. I recommend using that difference node that's built into one of the common node packs. It blows differences up to make them super visible
Plug both images in and check the output of it
hell its good news for me, means ill just let SDUltUpscale run on euler at double the cfg
Ahh yeah, good point since the sampler isn't in the default list yet(would take like 3 min to manually put it in the list in the code)
wana blind test for me? which do you like more
left painting had a bit more strength put into it
left i prefer but feels a little too noisy
was riding that same fence.
gonna play with some settings
euler_pp and euler_ancestral_pp are already added to the list
they aren't the same, here's an example (had llama3 writing weird fever dream prompts)
the one on the left
here's another. they are closer in this one though, you have to look pretty hard at the difference image
those prompts
shhh i like making goofy shit
ahhhh. Every stray cat's fate
is there any news from Stability AI
If we use The cliptextencodesd3 node and split the prompt into three, then how big is the difference then?
no clue, im too lazy for placebo workflows like that when i can spin another seed or inpaint it with diff-diff later
last comparison for now. but yeah, either way, it looks like eulercfg++(alt) is 1/2 what normal euler is, just some subtle differences in small details and the edges of things mostly(which is what it's meant for anyways if you read the paper about it)
so if you make an image with cfg++, then for now, just use euler+sgm with a cfg of 2x your cfg++(so if cfg++ was 2.5, use cfg 5 in ultimate) for ultimate sd upscaling
it should be close enough
try euler_pp, not euler:
thats the same thing
oh wait nvm, it's a new node, thought that was just from some node pack
just pull current ComfyUI code - I am getting identical results from those:
lemme check the code real quick, they might actually be the same thing
yeah those are them, just added into the regular list of samplers
that standalone eulercfg++ node isn't needed anymore then, sweet
https://github.com/comfyanonymous/ComfyUI/commit/69d710e40f43b87dd7124b9ee1e8982f987827e9
"Add euler_ancestral_pp which is the ancestral version of euler with the
same modification."
to sum shit up, eulerpp=eulercfg++alt and euler_ancestral_pp=eulercfg++ and you no longer need to use a customsampler workflow to use it
too much text, argh!!!
there, happy? guess i shoulda looked at the history on github first lol
need moar images! ๐
well again, if you want to get an almost 1:1 with regular euler or euler_a, use 1/2 the cfg with the pp versions. so 8->4, 4->2, etc. use the comfy essentials enhance image difference node to see where the small changes are happening
the differences with the enhance difference node will look like this if you set the cfgs like i just said, between the regular and pp versions:
more color=more difference, black=no difference
so there, teach a man to fish, yada yada yada, go have fun
interesting
your graph will look like this (above difference image)
Oh damn, the difference gets a lot stronger the more steps you put in. i reran that same prompt with 50 steps
look how much brighter the differences are
take a guess which is which, pretty night and day at 50 steps
And has anyone tested the ipndm sampler? ๐ค
The right image looks more "coherent" with buttons i feel, like the button's are more stable, best way to describe it
yeah there are some pros and cons on both, but that's why i chose a prompt like this, tons of greebles and shit to get wrong
yeah i messed with it a little the other day, there's also that other noise scheduler GITS or w/e it was
iirc, they are geared for lower step counts though
like <20 steps or something
I just started testing it, but it doesn't look bad....
i think they precalculated some manifolds or something like that, i'll have to reread the paper on it
(cfg4 - 28 steps - ipndm)
SD3?
yep fp16
it performs very similarly to euler_pp
You can always use https://www.diffchecker.com/image-compare/ to compare images
like im doing sidebysides of the two, ipndm at cfg4 and euler_pp at 2 (so they match)
i just use a difference node, makes it easier
OH! Good to know ๐
Is there any diffrence between euler_pp and ipndm in terms of text in images generation?
lemme try
#๐๏ฝsd3 message This is an example, same prompt, same seed, the only difference is the CFG, as eulercfg++ is 1/2, to make a fair comparison
ipndm on the left, euler_pp on the right
but afaik, the whole cfg++ thing was more about correcting errors on things in the foreground
like cfg++ falls off on things in the distance, but makes up for it where it matters
in this example I prefer euler_pp
the text is supposed to say "pee is stored in the balls"
but eulerpp fixed her extra arm in this one
but anyways, i'm sure if you ran a thousand double-blind A/B tests between the two, you'd probably find them tied in the long run, in terms of preference
someone should make an imgsys.org but for samplers
wtf
yep
mo'fuggin catdawg on the right
what about other seeds
you dont have the use a custom sampler for it anymore btw, euler_pp=cfg++alt
yeah, they are built into the samper list now, euler_pp and euler_ancestral_pp. whatever euler and euler_a worked with, these will work with
just use 1/2 the normal CFG
10->5, 8->4, 5->2.5, and so on
It wasn't working for me, I have updated Comfy
I found something interesting, on higher model shifts the text is better then low model shifts
using euler_pp
yeah youre just pushing the sigma curve more front or rear loaded
the higher the shift, the longer it spends making large changes, the lower the shift, the more it focuses on smaller details
A close-up of a person's face and upper body, seemingly covered in a textured, muddy substance. The person has intense, hazel eyes and appears to be standing outdoors, with a blurred background of houses and a wooden structure. The mud or substance on the person's face and body has intricate patterns and textures, resembling organic growth or perhaps a form of artistic representation.
sorry for beingdumb but is ipndm is a sampler? or whetever is it how to use it on comfy
oh yeah uts a sampler but i not remember i saw it on comfy
It's ComfyUI's strategy, they want to release 24,000 different samplers this year to drive us crazy
lol
Yep, they work for the government
what the differences? how good will be the input with fp16
ipndm was added like a week ago or something to comfy defaults
oh so i might no updated comfy i guess
euler_pp ??? damn
what means euler_pp? is it special or u take it as a joke
Two k-sampler workflow using euler_pp, same seed, and both of the images are 2 CFG (First pass is left image and second pass is right image)
The shadow and lighting is better in the second image
good prompt
first image is 8 model shift and second is 6 model shift
oops typo
A person wearing an intricate costume resembling a post-apocalyptic warrior. The attire is adorned with metallic elements, sharp spikes, and gears. The individual has tattoos on their face and arms, and their hair is long and dark. They are positioned in a busy urban setting, possibly a street, with various people walking around in the background. There are signs and buildings, suggesting this might be in a commercial or residential area of a city.
i use custom vae for better contrast
i get weird artifacts @ CFG 2 / shift 3 on euler_pp
yeah, well my prompt was very concise ๐ Harry potter is the goat
try using 40 steps plus Sgm_uniform as scheduler
first pass was 43, second was 23. will try to bump 2nd one
I use 40 for the first pass and 30 for the second pass, and for the first pass i use 1.00 denoising and for the second i use 0.75
oh shit, i put the model directly for 2nd pass thats why. (no shift). (I think? will try more)
And try using cliptextencodesd3 and splitting the prompt into three, that gets me the better results the normal positive node
do you add noise between the 2 ksamplers?
this is my workflow basically
ok i pretty much have the same
try: Harry Potter is the goat lying on the grass... it's more fun
euler_pp with 1/2 CFG seems to be better than euler with full CFG:
ittter
There were some I could not post ๐คฃ
A close-up of a person's face, predominantly focusing on their eyes and the area around them. The individual appears to be holding or releasing a burst of fiery energy or light, which envelops their face and surroundings. The fiery particles are scattered throughout, creating a dynamic and intense atmosphere. The person's hair is flowing, possibly due to movement or a breeze, and there's a sense of motion and energy emanating from the image.
The word "or" negatively affects image quality, try not using as much as possible
using the character in the photo, create a new phote where this kid is running in a hot wheels Challenge. your friend called "cup" is a pilot and he's running too.
vroom vroom
๐๏ธ
chatgpt ahh
yo seizure emoji. how do i block emojis ?
SD3 is a clothing designer... new Balenciaga pants
sorry lol I can get rid of it lol
๐คฃ
i think covid made me epileptic. my eyebrow is twinging.
i made it my problem and not yours. i found an option on discord that stops all emojis from animating

Anyone who'd like to post a simple image with workflow (ksampler+mode+triple clip node) for me to test to see if I still have something weird going on after my comfyui update?
Thanks! Looks like I've got a bit different output (almost the same). But at least it's the same quality.
yeah ever so slightly different
Do you use modelsampling with euler_pp or not? I feel like the quality is much better the lower shift value you have
I'm testing autoprompter, and it creates very funny (or absurd) images ๐
leave that generating forever and a bot posting on some website
people will think you're insane 
People already think I'm crazy, or semi-psychopathic... if I do that, they'll imprison me
always love how SD3 does water
nice painterly style
I don't do a lot of painterly I should try to do some
SD3: "I don't know what to do, this prompt is too spicy to generate. I still don't know what to do, I have to censor this somehow. Is this model for male or female? Oh nm, this will do" ๐
it's like a romance novel cover
im happy to see the anatomy is not messed up
