you don't need the checkpoint?
#🏞｜general-with-images
tame spear
u can use lora for every other checkpoint 
sly patrol
u can use lora for every other checkpoint <:Shruge:1010913657456164885>
so my character will look like her
but I don't want the character
I want the art style
tame spear
ohh so u need checkpoint not lora
sly patrol
ohh so u need checkpoint not lora
yeah
tame spear
yea
bc amd and drivers 💀
ofc would be best if it work fully well i would jump into 7900 xtx
tropic shell
As a quick note, please make sure all images you post adhere to #✍🏼｜rules-and-tos regardless if you are asking for help or not 🙂
sly patrol
As a quick note, please make sure all images you post adhere to <#10402999656474...
damn
dense tapir
ofc would be best if it work fully well i would jump into 7900 xtx
Well, 7900XTX too but not yet for rocm, and how much faster is it from a 6700xt for this I wonder?
6700xt is about 2-3s per it for 512x512 euler a 20 steps
my 1060 is 1-2it/s so tons faster, but raw rasterization speed it would smoke me.
tame spear
i hate when u try new checkpoints but it needs other sampling method or it end looking weird
lapis quartz
guys can you help me with my prompts
Joe Biden as a model in Balenciaga fashion show, stern face expression, hyperrealistic, high resolution, realistic lighting, fashion show model, sharp focus, highly fashionable, pretentios clothing, dim color scheme, upper body shot
Negative propts: smiling face expression, office clothing
i cant manage to take off his goddamyn suit
glossy herald
it's a problem on some things. Like yoda not green too for example : bias in the model because all pics of them are like this
tame spear
anyone tried generate like biden kissing anime girl ? 
lapis quartz
we can try rn
lapis quartz
anyone tried generate like biden kissing anime girl ? <:emoji_kek:83703923188065...
tho it is dreamshaper
glossy herald
anime drawing of a Biden in a grass field, trees, masterpiece, intricate artwork
tame spear
hmm interesting
lapis quartz
hmm interesting
lapis quartz
what prompts, what checkpoint
glossy herald
what prompts, what checkpoint
anime artwork, digital painting of a Biden wearing a white dress, masterpiece, intricate artwork
the model is the one I'm still training, I'm testing the checkpoints currently
glossy herald
the shoes ?
tawdry iron
just did this...
quick current
gray monolith
Sabaton's Bismark, same pipeline
SD is BAD at ships
MF DOOM - ALL CAPS, again the same pipeline
Karma police as various modern artists
User's workflow looks like this:
You even get director's notes:
fickle gorge
How do you make these animations from images on which there is a small transitions between them, like a landscape with wind blowing on the trees?
Note: not the zoom-in/out or those that change the image radically
dense tapir
fickle gorge
I'm getting some interesting supersonic aircraft, but it doesn't seem easy to mix supersonic aircraft and civilian
subtle storm
Airplanes together
Dog #
Dog #🏁｜airplanes
Dog #1072220168534642768 airplanes crash
#1072220168534642768 airplanes
Dog with bear #
junior gust
ripe cedar
https://toyxyz.gumroad.com/l/ciojz I KNEW this guy was going to update his blender rig quick after he saw the new hand update. Comes with adjustable facial expressions, accurate hands (apparently), and body posing.
My brother is out of quarantine so I can get back to stable diffusion tomorrow on my PC.
gray monolith
I'm getting some interesting supersonic aircraft, but it doesn't seem easy to mi...
Plane and space/ship prompting is quite random. Specific types may help with planes at least https://haveibeentrained.com/?search_text=yak-54 night fighter
wispy nest
for anyone unaware: "sentient" is a great word to turn something more sci-fi or straight up make androids with some models. an example: both of these images use nearly the exact same settings, but the first one is no positive prompt and the second positive prompt is just "sentient". it's a lot of fun at times. give it a shot if that's what you're going for! 🙂
glossy herald
for anyone unaware: "sentient" is a great word to turn something more sci-fi or ...
nice trick, I never thought of that one, thank you for sharing !
ripe cedar
It seems like maybe a lot of words reflect connotations, rather than the word itself.
wispy nest
i've certainly found that to be true
gray monolith
I got quite far with just using fragments of song lyrics, genre and an artist name for making video. It's surprising what SD will do with pretty arbitrary inputs
glossy herald
wispy nest
that was stupid of me. i thought:
i'll just put a space in the Prompt S/R box and that way i'll have a blank positive prompt to replace with 'sentient'
failing to realize it could do what the first image shows. can't complain about all of them though. the second image ended up with that prompt and was still pretty nice imo 🙂
true coyote
Muscle memory hit me and I started prompting google images 😭
smoky oak
Muscle memory hit me and I started prompting google images 😭
That has been happening to me too lmao-
I also find I speak like "hello comma how are you doing today questionmark" sometimes after I have been using voice typing for an extended period of time 😅
more so comma rather than other punctuation
Also, wanted to pop in an apologize for how long my guide is taking. I have been feeling unwell these last couple days, and the process seems to very inconsistent, for whatever reason. To say its extremely frustrating and discouraging is an understatement
I have no current expecations on how long it will take for me to release the guide as of now
Its quite frustrating how the process seems to get worse over time. I follow the same exact steps, yet the quality degrades the more I do it. This is now the third process rework leading to the same problems
glossy herald
photo portrait of a woman with black hair, colorful makeup, delicate details, realistic photo, artistic, symmetric art
Negative prompt: nsfw, bad hands
Steps: 40, Sampler: UniPC, CFG scale: 5, Seed: 3691380467, Face restoration: CodeFormer, Size: 640x640, Model hash: ca3d252be5, Model: Tests_SDArt_mix_v2
sly hawk
dire current
rancid warren
Happy Clatf :)
glossy herald
covert heart
dense tapir
@glossy herald
glossy herald
<@223150786929098755>
wow ! That one is delicate ! I like it !
dense tapir
Royalty which all cats are.
dense tapir
I like it
glossy herald
it should render better on 2.1
it was a fast test to see how the caption were working
I did it on 1.5 for speed
just need to find the courage to finish captioning now x)
aka, my second part of the screen
dense tapir
Well, let me know when it rolls out
glossy herald
I'm planning on 🙂 I may open a first "test version" and a collection of around 100 "best of" from people in specific styles, so I'll ask for pics at some point maybe
dense tapir
Oh, nice
I gave up on colab as that was just no fun and more like torture to me.
2.1 didn't help either
This sort of cracks me up for some reason
glossy herald
my cats can turn easily into girls. especially when going intricate
dense tapir
I dunno that first one that one there looks shifty
glossy herald
it's the hair ^^
not sure how it happened
I haven't prompted for it
but that hair is sus
dense tapir
yeah
glossy herald
dense tapir
btw, you notice the first one has too small of a head and the second too large? lol
dense tapir
Damn, even came with a huge ass goiter.
glossy herald
since it's trained on the different PoW submissions (with usable tokens this time, this is "cat", not "sks" 😉 )
it's quite good at the themes we did in PoW
like Cats, or futuristic clothing
but it lacks in other subjects
dense tapir
Strange when it does that
glossy herald
it's 14 different PoW datasets
it's also why I want to collect a little more from the best on different subject where it's still lacking
dense tapir
Dark Trump cometh for you.
glossy herald
I lost my "obama jedi", I had a good series x)
glossy herald
I had to format my computer and I went a little rough on it, did a clean reformat without taking my files, when I got hacked
dense tapir
I just installed dual boot Linux for testing and spent all day yesterday doing it several times, etc... I pucker when I do that because in 2016 I lost everything when Linux decided to zap all my drives when I was installing. Ryzen was new so Linux didn't know and ZAAAAAAP
stoic pivot
"No mr Bond, I expect you to dye".
glossy herald
nothing like a good pun
fierce crest
stoic pivot
When I said I want to take you on a date, that was not a question honey puff, how do you want to get there, over the shoulder or piggyback?
earnest solar
smoky oak
Alright, I hate to say it, but I may have to pull the plug on my guide.
I cannot for the life of me figure out why sequential generations degrade so massively
and thats not me being overdramatic. I can quite literally generate the same image 2 times with the exact same seeds and stuff, but the more I do it, the worse the results get, and I do not understand why
If anybody know what would lead to degrading results, please let me know. This is now my third process rework to get around this, yet it is also the third process that started good and ended up losing all of its original quality to blur and artifacts
@glossy herald @dense tapir @static tusk If any of you guys have any idea as to why this keeps happening, please let me know. It is genuinely discouraging me from pursuing this, even with 100's of hours dedicated to it
static tusk
<@223150786929098755> <@315547788211388417> <@824919846570819584> If any of you ...
Im happy to help you out but without knowing what your are processing its impossible to know. Perhaps provide a draft of your guide so a few of us can perhaps test. Not sure here
smoky oak
I would provide a rough outline, except its not behaving at all like it was just a week ago. This is genuinely pissing me off, and I am not sure why i keep having to refine my process to achieve results temporarily
for example:
First time using this new process
as you can see, there is an excellent amount of detail and realism
(please, feel free to pixel peep)
I mean, its not hard to see the difference
the process clearly works extremely well there
This is the second time I used it. Again, insanely good results from it
Once again, the process works great
now fast forward to how its working now, with the exact same process
What is even happening here. I don't understand how the process that worked so good before is now not working at all
what even is this
static tusk
maybe the process was great for certain subjects but not others? Do you have the same bad result with the same images earlier?
smoky oak
There is no way for me to replicate them perfectly now, but I can give it a shot
This is now the third process that has had this problem
started off amazing, now I can't get anything good out of it
static tusk
its this reason why beta testing is important. Its not easy doing everything on your own
smoky oak
I just tested it
the result is considerably worse
something is actually happening here
Original
New
as you can see, its more blurry almost
its not as bad
You can see instead of detailed hairs
its scratchy noise
I will be right back
static tusk
as you can see, its more blurry almost
yep not as good. just go through each step till you narrow it down
dense tapir
smoky oak
yep not as good. just go through each step till you narrow it down
The problem with that is that I can't even get it to recreate a single high quality image, so I can't reverse engineer from exact settings I already have. And also, it's an extremely extensive process with multiple variables that you are supposed to change in order to get better or worse results
It's extremely frustrating to say the least. I've spent well over 400 hours working on this process at this point, and I've nailed it down three separate times, and then while I'm reporting on it for the guide, it inevitably degrades severely inequality
Antenna answer anybody who is asking, no I have not updated any aspect of stable diffusion other than my extensions, but that happened immediately after the first image, where I then produced the image of the deer, so it didn't degrade there
At this point I might just redo my entire VENV
static tusk
it does sort of sound like a problem on your PC. everything should be repeatable with the same settings etc
static tusk
At this point I might just redo my entire VENV
are you overclocking at all?
smoky oak
Yeah, I'm not quite sure what's happening here.
I have examples of the process producing absolutely stellar results, and I have every single step of that process written down as a followable guide, but I just cannot seem to get it to recreate the quality results from initial process refinement
smoky oak
are you overclocking at all?
I overclock my GPU
But I've never had any problems with it, I validated it stability a long time ago
static tusk
or maybe its un underlying issue with the process you have discovered but have yet to understand the reason why. Again Im just making guesses here
in any case you can use a program like HWINFO to check for windows hardware errors. its a field displayed in the GPU section
typically its related to overheating or degraded vram etc
I need to step out for a bit so will be afk
smoky oak
I'm quite busy as well as my mom got home. I'll take a look into all of that stuff
split rover
Ooo thats a cool one 😄 Curious what the prompt was for it
split rover
haha no worries, I have thousands of prompts signifgantly more embarrassing 
Next version of XL should do far better at the darks for ya (little preview with your prompt ;D)
static tusk
If you want a darker image try using "dark atmosphere"
split rover
😄 Excited for people to make stuff with it here in a bit. Still got a ways to go though on training (hence the whole, XL beta haha)
wispy nest
IS THAT AN EGG
dense tapir
yep
wispy nest
I have the perfect idea
dense tapir
a space seed so to speak
wispy nest
a space seed so to speak
alien cum
dense tapir
well, the size was YUGE
wispy nest
LMFAO
dense tapir
wispy nest
MORE BALLS
wispy nest
wispy nest
when you just gotta take a moment and admire the destruction
arctic needle
during the generation, i happened to like the preview one step before the actual final image. is it possible to generate from this?
hasty nova
when re-captioning a dataset of 100 images... it looks like this (the drop shadow on windows stacks)
yes im using notepad, the dataset tag editor extension wasn't working
smoky oak
when re-captioning a dataset of 100 images... it looks like this (the drop shado...
lol
I do all of my tagging in notepad too lol
hasty nova
Its a pretty boring process but it has to be done.
smoky oak
Yeah, I feel you there haha
hasty nova
I actually have captioned this exact dataset but at some point the text files got replaced with captions generated by blip. Dunno how, dunno when, but it means I had to reception, which means it is hopefully improved.
hasty nova
Thank you! It's training right now. I'm trying Lora for the first time
It's for concept art images geard towards game developers/asset modelers, or environment/character design.
smoky oak
Best of luck, it can be a bit of a pain to get the captionign right
hasty nova
Well I think sample images don't have a prompt.. I was probably supposed to set that up
hasty nova
Oo fun
It appears I still do not have a prompt on my sample images, but I think it's still training so it's fine. I'll leave it overnight and check it in the morning.
smoky oak
I have been working on LoRA's for abstract animals
or I guess exotic would be the better term here
I made one for Opossums and ferrets
stoic marlin
wispy nest
glossy herald
let's test the v3 of the PoW megamix
Ok, I'll run the 615 prompts that I have in the dataset, let's test for real
this feels good
(I skip mermaids share, Clyde isn't having it)
little fox
little fox
glossy herald
it feels a little too trained
I have an over epoch earlier, I'll compare
I did also notice quite some typos in my caption :/ doens't seem too bad in the results but still, it's fixed for the next version
frosty wave
it feels a little too trained
It looks very cool, but yes I can definitely recognise a lot of the submissions. (btw how is that, I thought you did not include user_tokens?)
glossy herald
yes, but I'm doing the testing on long prompts here, and the exact same long prompts I used in the training, like "a photo portrait of a man with a smoking face, and clocks on his suit"
the nice thing now is, you can replace any of those tokens and get something coherent
but if using the full prompt like it was trained, I get too close of a result
I'll use the 100 epochs for tests next
this is all on 120
frosty wave
the nice thing now is, you can replace any of those tokens and get something coh...
ah nice cool :) that's awesome
glossy herald
it was quite some work, making a good caption for 615 pics lol
test 3 🙂 not good enough yet, but soon
frosty wave
I feel like you might have rather high standards regarding what's "good enought" 😂 most of these look very cool :)
glossy herald
I do yeah, but so does the community with their nice pictures, so I want to make it worth it
epoch 100 seems better in terms of diversity
same prompt for example
split rover
I definitely relate to this
glossy herald
I love the little cute dudes it can do in the underwater world
lol
OK I'll finish a full round of prompt, but I'll put this test checkpoint online later
glossy herald
here are the main trained tokens that I'll test
Type of art
3D
abstract
paint
painting
render
rendering
anime
artistic
artwork
digital
drawing
photo
picture
portrait
Alterations
asymmetrical
colorful
vivid colors
cute
dark
detailed
cyberpunk
futuristic
elemental
glowing
half X half Y
intricate
made of
masterpiece
fractal
rainbowshift
realistic
strange
Subjects
alice
cat
city
clouds
cow
creature
dress
field
fire
flower
forest
giant
hat
landscape
mermaid
miku
mist
monster
moon
music
neon
night
portal
room
sea
shadow
sky
space
street
tentacles
trees
underwater
water
world
smoky oak
@glossy herald The guide is officially on Hiatus. I am getting to stressed about it, and its not fun anymore. I am going back to LoRA training for now until I feel its right for me to resume the guide
glossy herald
glossy herald
<@223150786929098755> The guide is officially on Hiatus. I am getting to stresse...
I saw. it's one of the reasons I gave you the advice to release a first version, and improve on it : to not burn out on it
You're creating pressure for yourself by sharing intermediary steps, teasers, and getting feedbacks mid way. You're doing yourself more harm than good spending so long in this status, yes.
I hope you manage to salvage something out of all the work you put into it, but for sure, take the break you need
smoky oak
The problem is I can't release a first version as it doesn't work. Thats the problem I am running into
all 3 revisions worked for about 2 dozen gens, then just completely stopped working
glossy herald
you had a first version of the guide working at one point though. you wanted to add more tricks to it back then before releasing anything
smoky oak
I was consulting pure fire about it, and I even did side by side comparisons where you can see an unexplained drop in quality
smoky oak
you had a first version of the guide working at one point though. you wanted to ...
The first version started making lack luster results, which lead to version 2, which then did the same, and finally I got to version 3 where I do the downsampling, and while it worked damn good for the first few gens, its been about 6 dozen gens in a row with no proper output and I just cannot for the life of me figure it out.
The more I use the process, the worse it gets and I have no idea why
this is now the 3rd time I have reworked the whole process only for it to slide back into blurry/artifact filled generations for no clear reason. Its genuinely frustrating me
this was when I first made 3.0
and these are the exact same steps now
It just makes things crusty, and grainy, and messy. The only reason that looks even slightly decent is cause I used the base image from when the process was still working
Now its either consistently producing crusty and artifact filled gens like this
or soft/blurry results
glossy herald
hum... not sure why yeah. I have the same types of artefacts when I upscale too much on comfy too though
smoky oak
I have every step and value written done for it, but it just doesn't work anymore, and I have no idea why
Its the whole reason I had to make a 2.0 and 3.0, cause the previous before did the same thing where it just drifted towards blurring things over time the more I used it
For now I am pausing to work on a long list of LoRA's I have had in mind, but when I feel up to it, I am gonna try to rebuild my venv and see if its maybe an extension I have that is somehow causing these effects. I genuinely have no other guesses at this moment
I haven't been able to make it past the sample images for any of the 3 processes, as they justs tart falling off into lower and lower quality results regardless of the process being set in stone. So bizarre
wooden silo
How to get rid of such noise ?
lime storm
привет
graceful basin
Superman is very fat with a dark forest background and rough face, photo size 4.3 and quality 2
crystal deltaBOT
@graceful basin
plucky pike
Create Wolf wild realistic
crystal deltaBOT
velvet wren
graceful basin
Zombie-shaped mermaid in the dark sea with red eyes and image size 4 and 3
crystal deltaBOT
@graceful basin
empty canopy
Made this from a section of another image i made, https://cdn.discordapp.com/attachments/434541578678435844/1097874215295782972/00223-898021256.png
glossy herald
randomizing on the tokens in that model I trained this morning
crimson dawn
people make things up about things they don't understand. often times projection fills in the blanks. maybe they're just insecure about their line game
haha they're stressing so hard about it
look at all those squiggles. they can't do it!! hahahahahahahahahahaha roflcopter ahahaha
gaunt heron
glossy herald
lime stag
test
gaunt heron
You did it!
dense tapir
Please, explain to me, given those numbers, why is AMD bad for SD?
gaunt heron
I don't know what those numbers are for.. I think AMD is just more difficult to use with SD, but can be used 🤷♂️
dense tapir
Naw, it is pretty bad. I mean a 6700xt is as fast at gen as my 1060 6gb. Not joking
Until I saw these numbers I just don't get it.
gaunt heron
yikes, yeah, if diffusing is important, it's not the best choice of card
dense tapir
that doesn't explain why though since those numbers blow away the other
It must be in the shitty implementation at the software driver level cause that hardware is way faster.
the last two are 6700xt vs my 1060
it should be ~3 times, if not more, faster fp16 vs fp16
the more comes from faster clock speeds
gaunt heron
no idea.. glad I don't have to deal with that haha
dense tapir
Well, I do because Nvidia is Satan on roids right now so I am heavily leaning towards team red even if slower.
gaunt heron
Maybe some answers in #1011228477954998273
dense tapir
maybe
swift horizon
empty canopy
Only times i've had issues with lines in general are when they're obscured for a while, they tend to not line up well.
image unrelated
something that came out at lunch
fading mango
A sweet-looking short-haired blue cat sister with a crown on her head, wearing a princess dress and smiling.
dense tapir
Thing is even without those numbers do not lie it blows my card into the garbage can (which it should). Has to be a software driver issue and let's not forget rocm is still waiting for 5.5.0 that will be for the current gen cards but for now 5.4.3 does not work with them. 3-4 months late with it looking like another 3-4 months before release.
The first set of numbers is the 7900xtx vs a 3090
It is useful if I am looking for a card with more memory and faster for SD
Well, I really do not wish to pay 1700-1800 USD (I just don't have it) for a card when I can get half the cost for not half the inference, or training, speed. I am brand agnostic.
of course, though I think you are missing my point.
My point is I was looking at a card that I can train with and I ran across a lot of vids with 6700XT/6800XT and I used my current T4 (colab) and my 1060 (local) numbers I know. I was underwhelmed because even though my card (next gen got the tensor cores) lacks so do those AMD cards. raw fp16 numbers they should not be the same speed as mine when they are TFLOPS vs GFLOPS.
Yet, as you said, my card lacks the same shit yet if I considered the same gen AMD card as my gen Nvidia card my card would not be the same speed it would simply devastate the AMD. Go forward 2 gens in AMD and they are only the same speed. I am not even attempting to compare current with last gen though I wish I could see vids of it.
a current 7900XT vs a 6700XT, for inference, are they too 2s/it for 512x512 euler_a 20 steps?
btw, on my card 512x512 is 1.x it/s not until I do 768x768 do I take a massive hit
I dislike shark because I am limited to its models
Has anyone used the xtx to train with not simply inferred with it? I am mainly wondering how fast, or slow, it is doing that for lora/locon/dreambooth.
please do, really, because I am so tempted. I had two brands I always bought from. Nvidia EVGA, and AMD Sapphire. EVGA is now gone and I am despising the current Nvidia with a passion. I will take slower over the avenue they are going. Been buying Nvidia for 20+ years and the loss of EVGA before the 4k series came out already made me sick.
My last AMD card was the 7870 and it burnt up due to a fan that died (it didn't have throttling back then) so I needed a card fast and grabbed this 1060.
the 24gb is too juicy for training
Do you have a link to those benchmarks as I would like to see them for myself?
btw, I already had to buy more ram, a new cpu, a new case, and a new psu (eventually) then the Nvidia card. This is beyond ludicrous.
This is the T4 training a locon a few weeks ago at 512x512 (for speed purposes)
Well, all of this was just to use a Nvidia AIB card. Not cool
empty canopy
My plans dreams are to build a compute server, 1 GPU card though
dense tapir
My case is now obsolete because of these behemoth long ass cards
they are now 1-1.5 inches too long
one last thing
docker rocm in windows is slower than rocm in linux everyone says that it is about 50-75% faster.
docker slows it down or maybe it is windows
Yes
DirectML is garbage the AMD users have all said
works but super slow
something is troubling me
@smoky oak Uses a 3060ti and their it is 11it/s
dense tapir
I can't wait to hear from you and your results because I am HEAVILY leaning towards it.
ping me when you test it
smoky oak
What did I miss?
dense tapir
sleepy time, and thank you
smoky oak
whats the information
dense tapir
your it/s vs the 7900xtx
early stages right now and not too bad. when fully implemented should be a lot better too
I'll take 15it/s and 24gb
dense tapir
Yes, I agree. look above at the numbers
smoky oak
The 7900XT is such a good card
insanely powerful, WAY smaller than NVIDIA, much less power
dense tapir
XTX is what I am after
smoky oak
XTX, yeah
dense tapir
I just do not like anything Nvidia is doing or the vibes I am getting from Jensen
He hates us
smoky oak
AMD's cards also absolutely rip in gaming right now as well
dense tapir
yes, even last gen raw rasterization
dense tapir
yes
dense tapir
this is why I am going to drop the bomb on it
smoky oak
The fact that it is even working
dense tapir
Dude, you totally rock. I really appreciate all the time you obviously spent on this. Thank you for sharing, worked the first try. 7900xtx SD1.5
For the above image
smoky oak
but also, I would say its a shame cause it doesn't have xformers, but it has so much VRAM its not that big of a deal for gens. Especially if you follow my upscale or similar upscales that never go above 1024
dense tapir
xformers is such a mess
smoky oak
Its nothing but pure gain on my card, but I do know some cards have issues because of it
dry crow
So thats with the directml version or not ?
dense tapir
So thats with the directml version or not ?
ROCm
dry crow
Ahh on windows ?
dense tapir
Not yet
Still great news and 5.6 (the version after this beta one) will get full on RDNA3 support so about 75-100% faster
dry crow
Wow thats gonna be game chaning
smoky oak
That sounds amazing
smoky oak
AMD has had the hardware for years now, their hardware is genuinely in my opinion above NVIDIA in so many ways (efficiency, VRAM, form factor, power draw)
Only problem has been the fact that they're useless outside of gaming
dense tapir
That is when the game flips
Agreed
It always made me giggle that team green is far from being green YET team red is
smoky oak
AMD entering into the AI game is the leap we the consumers need, and the slap in the face Nvidia needs
dry crow
Team red is Linux friendly too
dry crow
Thats why i prefer amd and the good amount of vram and performance for a decent price
dense tapir
We have Battlemage, Celetial, and Druid coming from Intel and Celestial, and especially Druid, don't have the stink of their previous CEO on it. Let us not forget that CDEO is who made Vega cards for AMD that were utter trash
A lot of changes are coming and Nvidia needs to be knee capped for the way they are treating us right now. I mean I only sense pure hatred from Jensen for us.
dry crow
Okay so what does that means for Intel ? I heard they wanna improve performance with their new driver again
dense tapir
As if we peed in his morning coffee or something.
For intel it is getting an AI chip
Battlemage is 2024, Celestial is 26, and Druid is 2028
Battlemage still has too much of the former CEO but Celestial barely any and Druid none
My hope is they will pull it off. I mean if they make the AI chips for autonomous cars why not for GPUs?
dense tapir
My way of thinking about this is how much am I spending for 24gb and how fast can I train in comparison to a server based 3 gens ago card or a modern 4090. 4090 at almost 1800 is too much
smoky oak
The big problem is that NVIDIA doesn't give a damn about the consumer anymore, their income comes almost strictly from industry now.
they can afford to lose a few billion dollars on the consumers when they're making hundreds of billions of dollars off of the industry
dense tapir
I know they are talking about a price reduction coming from Nvidia once the last gen stock is dried up so by summer
dense tapir
30k sold at about those prices is not uncommon for them
what do they care if we buy their limited stock of consumer "gaming" cards? they don't
AI is making them billions
btw, A100 destroys H100
Sooooooo, 45k for H100 A100 must be even more
this is why they are treating us worse than dog shit right now.
Well, I asked Google which brand is the best for AMD and found this (echoes how I purchased precisely)
dry crow
AMD, Saphire and Asus making the best amd cards imo
smoky oak
Power Color makes kick ass AMD cards as well
dense tapir
I was alwas EVGA for Nvidia for 20+ years then they left I knew this was going to be bad
smoky oak
My friend has a 5700XT Red Devil and that card could kill somebody
dense tapir
dry crow
Power Color makes kick ass AMD cards as well
Yea i had a good R9 280X from them
smoky oak
Her card is the nicest card I have ever seen from AMD
dense tapir
when 7870 was the thing they were rubbish
smoky oak
this is her card
dense tapir
lots of recalls and returns now I guess they changed
smoky oak
Look at this monstrosity
dense tapir
Devil card is a good one, yes
smoky oak
gorgeous GPU, and she got it for cheap at the time
it was newest gen, about 9 months old in the GPU crisis and she got a new one for like $310
Its so damn heavy it should be classified as a blunt force weapon
3.9 pound GPU
dense tapir
Well, I am waiting on ceres to ping me with results when they get it all up and running but I have been leaning 7900XTX for a while now. It is not just the price it is the vibes from Nvidia and the direction that company is going I don't wish to go.
smoky oak
Yeah, for sure
I am super conflicted about what I am gonna have to do with the whole laptop situation
dense tapir
I don't need to be treated like dirt and expect to take it. I will gladly lose it/s than to take what Nvidia is serving.
dry crow
Im need to go off for sleep now cya, but thats a nice topic ^^
dense tapir
Take care. I look at it like this I can buy this gen and next gen and be about the same cost as just a current gen 4090. That is pretty damn sick. 😦
dry crow
Well, I am waiting on ceres to ping me with results when they get it all up and ...
Ping me to if you got more Information on that pls
dense tapir
Ping me to if you got more Information on that pls
Will do.
RDNA4 is a big big game changer but right now the reason the drivers are so late is the new RDNA3 tech using chiplets.
all of the card from AMD have dropped in price except for the XTX 😦
shadow rivet
hey
uh
I get this thing
I have it running though
do you happen to know what the issue is?
dense tapir
gradio looks to have crashed
shadow rivet
oof
smoky oak
Thats a new one. Never seen that before
smoky oak
thats the proper way
dense tapir
wth is hiccup?
smoky oak
grand bone
I keep having a problem where my cartoon style images end up blurry. Like the preview looks like it's going to be amazing and then the final result is extremely low-res and blurry, like this. Does anyone know why this is? more steps and different sampling methods don't seem to fix it. But I have had good experiences with this model too, it seems random.
smoky oak
this is what he look like in the movie
dense tapir
Ahhh, yeah
smoky oak
I keep having a problem where my cartoon style images end up blurry. Like the pr...
Are you using high res fix?
grand bone
No, mainly because it usually dies when I try, but plenty of times it works fine without it
dense tapir
I never watched 2 but loved 1
smoky oak
1 was amazing, 2 was great, 3 was ok
however side note rant
actually, let me just say it
Watch Puss in Boots the last wish if you haven't. Genuinely a phenomenal movie, and I know that sounds absurd lol
smoky oak
the original was meh at best IMO, but the new one is actually insanely good
so good that many people are saying its one of the best dreamworks movies all time, and I agree
smoky oak
It is a master class in character design, story telling, hero design, and animation execution, genuinely
dense tapir
Antonio Banderas still?
smoky oak
Puss in boots the last wish stomps every disney/pixar movie from the last 5 years
smoky oak
Antonio Banderas still?
yup
dense tapir
oh, sweet
smoky oak
excuse the IMDB, the movie got hate bombed by critics
dense tapir
I still remember Puss being played by Christopher Walken
smoky oak
Puss in Boots the last wish actually won the box office over Avatar 2 and wakanda forever in the span it was in theaters lol
dense tapir
Damn
smoky oak
I cannot stress it enough dude, you should really watch it, Its extremely good
it deals with HEAVY ass topics
dense tapir
you know it is the studio's fault we never got 2 and 3 when they were supposed to come out and only after the contract ended did it go back to Cameron and he finally made them a decade later
shadow rivet
I fixed it by restarting my PC
smoky oak
life, death, hatred towards yourself for your past failures, accepting your mortality, it has actual deaths and a bit of blood, the story is rich, the visuals are phenomenal, the animation is as good as it gets
and its also a low budget movie by todays standards
only 80m. the original puss in boots back in like 20010 had a 130m budget
dense tapir
Just wait when this AI really takes off and budgets are 10k for Hollywood blockbuster budget level.
smoky oak
that would be amazing
dense tapir
Why Hollywood will get involved as I can sit in my bedroom and churn out a blockbuster
dense tapir
Personally, I can't, but so much talent out there thatcan and will
smoky oak
its just the opening 10 minutes of the movie
shadow rivet
guys, can you give me some suggestions on models?
I am currently using v1-5-pruned.ckpt
smoky oak
ah, the worst model possible lol
shadow rivet
oh no
smoky oak
1.5 is made to be trained, not used
shadow rivet
d'oh
shadow rivet
which model should I use? Can you perhaps link a few?
dense tapir
Pretty good. I watched half of that but don't want to ruin it
smoky oak
Pretty good. I watched half of that but don't want to ruin it
all good, its literally the first 10 minutes of the movie 😅
dense tapir
I watched a bit over 5m
smoky oak
ah, so you got to see the incredible fight choreography a bit
Just please, when you get a chance, watch it
you will not be disappointed
long trench
trying to install a lora in google colab, Help
dense tapir
It would appear people are catching on as the xtx is now over a grand.
smoky oak
trying to install a lora in google colab, Help
noisy
dense tapir
Sapphire makes two a Nitro/Nitro+ and a Pulse
smoky oak
The 7900XT is also good
if you could get one second hand, that could be dope
I know you don't do that
dense tapir
I need that 24gb of goodness
smoky oak
just saying in general
dense tapir
Some reason they spiked today
long trench
i have a lora model, im just trying to install it lmafo
dense tapir
Ahhh, pulse is their more affordable version so just go with Nitro+
more affordable means 80 USD difference
long trench
huh?
smoky oak
Sorry, I have never used google collab, otherwise I would help
long trench
steals
long trench
Bro's a savage
dense tapir
Well, this sucks that the price of AMD is going up for their xtx.
Oh, @smoky oak I read a few days ago that AMD has superior outputs because they have 2 HDMI and everything is to the current specs while Nvidia is using last gen specs as the new specs had not came out yet when the cards were sent to manufacturing.
smoky oak
@little fox
This si them stating a straight up generated image cannot be copyrighted.
The "work" was copyrighted, but not the images
dense tapir
I still have to buy a new case with it because I can safely do 304mm long cards while this is 320.
smoky oak
The images cannot be copyrighted on their own unless there has been human work done to them
little fox
And what do you consider creative contribution from a human actor. In the case in question it says "based on the application's representations" the examiner found that the work contained no human authorship. And there was no creative contribution from a human actor.
And still, nowhere does it say that you explicitly cannot. It was decided that in that person's work/case it didn't have creative contribution from a human. It does not say anywhere in this documentation that "Any and all generated images produced by AI cannot be registered for copyright."
dense tapir
Though I think I might can squeeze in 330 it will be interesting
little fox
It says that in that case, the person that submitted it was denied.
It could've literally been just someone pressing generate on a prompt simply written as "One girl at a beach."
That's not creative contribution from a human actor.
The criteria it specifically mentioned is "For example, when an AI technology receives solely a prompt [27] from a human and produces complex written, visual, or musical works in response, the “traditional elements of authorship” are determined and executed by the technology—not the human user."
For example, if a user instructs a text-generating technology to “write a poem about copyright law in the style of William Shakespeare,” she can expect the system to generate text that is recognizable as a poem, mentions copyright, and resembles Shakespeare's style.[29]
But the technology will decide the rhyming pattern, the words in each line, and the structure of the text.[30]
When an AI technology determines the expressive elements of its output, the generated material is not the product of human authorship.[31]
As a result, that material is not protected by copyright and must be disclaimed in a registration application.
So if a human decides on the expressive elements. The rhyming pattern, structure of the text, or for art... what elements are in the picture, where, what colors, etc.
Then the example they used isn't valid. Because you can decide on those things with prompts as a human.
dense tapir
@smoky oak https://www.cnn.com/2023/04/18/tech/universal-music-group-artificial-intelligence/index.html Next up will be Hollywood studios because they know their end is nigh when the bedroom gang hits.
smoky oak
<@296720689468145664> https://www.cnn.com/2023/04/18/tech/universal-music-group-...
Not surprising, universal music group is one of the biggest and most corrupt music groups in the entire world
sand copper
The "work" was copyrighted, but not the images
so does work mean the prompt? I don't know what else it could mean in this context
hearty turret
guys, why does my output look like it's slightly less saturated than the preview
it's like it loses it's punch somehow
dry crow
Update your Python Version to 3.10.9
calm canopy
😱 i've been running on 3.10.6
dry crow
I keep having a problem where my cartoon style images end up blurry. Like the pr...
You can share the settings used. Maybe something triggers it
dry crow
😱 i've been running on 3.10.6
Thats not a problem but as time moves and sd and the extensions get updated you can upgrade your Python too.
SD should work up to 3.10.10 for sure, didnt tried 3.10.11 yet
calm canopy
i think you just solved this issue that ive had and been talking to a dev about on a SAM extension for a1111
im noticing that me and others are having issues and we're all on 3.10.6 hmmmm
dry crow
Yea maybe thats the cause.
Some people get a message to install 3.10.9 when using xformers or dreambooth or kohya_ss
So that would be the Version you should try
I never used 3.10.6 with sd and started with 3.10.8 xD
Now using 3.10.10 without problems so far
calm bay
hi does anyone exist here
dry crow
hi does anyone exist here
Hey
wispy nest
Do you mean non-base models? custom ones or what?
calm bay
non base ones
wispy nest
yes?
dry crow
non base ones
You can find a lot of models on Civitai.com
torpid rune
Man it looks like i can learn new things here!!! Im so happy🥲
Can I ask you guys some questions? How to make this awesome ai? And how to make multiple poses of it? Im sorry for my question but im just a newbie...I've watched this on youtube and i guess its pretty cool to do it as well
dry crow
Can I ask you guys some questions? How to make this awesome ai? And how to make ...
Hey, you would need Stable Diffusion for it. You can run it local on your PC, in the cloud with google collab or by using Dreamstudio. There are also some discord bots and StableHorde to generate images.
Check #1072220168534642768 and #1080946152318443610
For help on an local installation you can also ask in #🤝｜tech-support
The Image is done by using a custom model mostly from civitiai.com and maybe a lora it seems, multiple poses can be done trough prompting like standing, sitting, lying, or by using an extension called ControlNet
dry cosmos
torpid rune
Hey, you would need Stable Diffusion for it. You can run it local on your PC, in...
Thats what i want to learn the prompt commands.. maybe I'll just search here if there's any.. thanks man!!!
dry crow
Thats what i want to learn the prompt commands.. maybe I'll just search here if ...
ah okay, for that there is the #📝｜prompting-help channel 🙂
i tried and got this image
or this rn
wispy nest
i just accidentally made a very metal album cover by putting vaguely metal words as the prompt. i don't think i can post here due to the screaming skullface and the blood...but jeez i wasn't expecting it to go quite so hard.
worthy otter
/generate Classmate Virtual Collaboration App
the app should allow students to type in their school,class, and section for the class (what time the class is at) and if there isn't already a chat for the class, students, will have the option to creat one. if there is one created, they can join immediately. this takes out the stress of trying to find somone in your class in order to join the group chat.
the pupose of this app is to allow students to find and join group chats for college classes and then communicate freely in those chats in order to work toghether and help their performance in those classes.
crystal deltaBOT
@worthy otter
glossy herald
hey there 🙂
smoky oak
Figured I'd ping you for a non problem thing to give you a little break lol
glossy herald
quite all right, moderation, models... the usual :p
smoky oak
understandable heh
glossy herald
it's calm right now, just some support going around
glossy herald
but I'm preparing the next phase for the model I released earlier yesterday
smoky oak
Oh very nice!
glossy herald
smoky oak
let me know how it goes
I am currently dealing with the hell that is my main drive having 0 bytes of free space
glossy herald
next phase is collecting community's best of, I'll open a community project soon on it
yeah I saw that :p
the TB is full :p
I'll need to do a cleanup too one day, but I got some fringes before it gets to be a problem
smoky oak
I have 4.5TB of ultra fast storage, but man I need a big sata SSD or something
I can't go back to a HDD
glossy herald
my big disk is a HDD. I do have a 2TB SSD (F:) and it's enough for the trainings and image generations
then I move to the big HDD when I make space
smoky oak
I have 2x 1TB NVMe drives, and 1x 2TB NVMe
so peak I can reach about 12GBP/s if I run them in raid lol
but I need like a 4TB standard sata SSD
glossy herald
yeah... I prefer some loading time here and there, it's ok to me, but I got quite the VRAM to keep things in memory so it's less a problem in SD
still, my LLM can take more than a minute to load the model
smoky oak
bro, my download stopped at 99.79%
and its refusing to continue
Guess I have to redownload the entire 11GB file
glossy herald
😢
smoky oak
yeah... I prefer some loading time here and there, it's ok to me, but I got quit...
For me, i handle a ton of high res audio files and waiting for my samples to load is a genuinely infuriating experience, so I had to go NVME cause even my old SATA SSD was getting bogged down
I GOT IT TO CONTINUE WOOOOOO
woven dawn
For me, i handle a ton of high res audio files and waiting for my samples to loa...
This. I use massive sample libraries. I've got em installed on raided nvme ssds. Zippity fast.
smoky oak
peeng
woven dawn
smoky oak
This. I use massive sample libraries. I've got em installed on raided nvme ssds....
Yeah, i do sound design and high res recordings, and my god clicking through my audio slideshow before was hell
bought an NVME drive that same day and the 4000/4000 I get on it never leaves me waiting
so now I have 3 lol
smoky oak
lol
just remembering how much carnage this single sound rained down upon my CPU
That single sound in live rendering brings my CPU to its KNEES
4 layers, 2 sounds, 3 notes, 41 effects, 14 automations
Ended up making its way into this basic demo
Audio synesthesia tinglies
Chromesthesia is the right word, I always forget it
glossy herald
rancid quiver
I just release Perpetual Diffusion: https://itsperpetual.art
I hope you'll like it
More examples here: https://twitter.com/thibaudz/status/1648662092055027714
rancid quiver
I just release Perpetual Diffusion: https://itsperpetual.art
I hope you'll like...
Large or Tall image with important ratio works (less duplicate).
crystal bloom
SD + Kaiber AI
hasty nova
Hey evryone! I am working on this lora (my first one) and I am thinking the 4k steps one is the best. Anything past 10k was overtrained.
(click to open full size)
Even though it got it wrong here, imma go with 4k
imma try it in conjunction with another lora, same prompt
probably not the best prompt for it but it worked
rancid quiver
ripe cedar
I tried to use Multi-Control-Net but the images turn out like this ie a person and then good hands but the hands themselves are seperate
These are the ControlNet inputs
wispy nest
dry cosmos
I tried to use Multi-Control-Net but the images turn out like this ie a person a...
You have to use it separately, with the right weights and in different steps.
trim phoenix
what process are folks using to change the color of a object? I'll use this simple one as an example -- I want the sandals from black -> red but keep everything else the same; I'm using controlnet/canny and getting mixed results
ripe cedar
You have to use it separately, with the right weights and in different steps.
I believe ControlNet models can be combined so they don't need to be used separately?
golden sorrel
real
near fjord
iamgine Forbidden book
crystal deltaBOT
@near fjord
rain gazelle
rancid quiver
(perpetual diffusion)
void rampart
Hi guys, I need a bit of help to achieve something I have been trying for some days without luck. The concept is simple, I have created this simple image. I need to separate the front trees from the castle in the background and the mist. I have tried several approaches, for example tried to use SAM (Separate all model) to try to select the trees but although it does a nice job, it is far from correct and this is because this picture with the mist and the thin branches are really dificult to separate. So my question is. Is there any effective way to separate the front trees? One solution could be to generate the trees in an image, the castle with the mist in another and then combine both (Something like photoshop does with layers) but don't know if this is possible with any plugin, controlnet, etc. Could anybody please help?
royal thicket
lol I found a website selling text prompts
dry cosmos
I believe ControlNet models can be combined so they don't need to be used separa...
The combination is not to create something final, it is for one model to cover the weakness of another.
ripe cedar
The combination is not to create something final, it is for one model to cover t...
Yes. So they can overlap ie canny hands+depth hands for example
I changed my Stable Diffusion shortcut icon to a SD card
dense tapir
Not surprising, universal music group is one of the biggest and most corrupt mus...
UMG says it is not against the technology itself, but rather AI that is so advanced it can recreate melodies and even musicians’ voices in seconds. That could possibly threaten UMG’s deep library of music and artists that generate billions of dollars in revenue
Well.... at least they're being honest :/
smoky oak
PEENG
smoky oak
UMG says it is not against the technology itself, but rather AI that is so advan...
I can at least respect that aspect
dense tapir
Aye
smoky oak
but then again its only a matter of time before it happens, whether they like it or not
and as a musician, I am excited for the prospect
dense tapir
I agree, but get ready for the dogpiling because none of these will just lie down and lose everything.
all about the money
dense tapir
lol I found a website selling text prompts
Why I left lexica's discord almost a year ago.
Demand me to give them my prompt. Why? So we can sell it. piss off and I left finally as they were something else.
snow wyvern
UMG says it is not against the technology itself, but rather AI that is so advan...
are you talking about Sovits?
dense tapir
I am not talking about anything just posting an article of what UMG is trying to do.
snow wyvern
I see
I just happen to know an AI tool called Sovits that can do just the same, it is available on github
dense tapir
There are a lot of them with better and better coming. Eventually a few seconds will learn and sing/talk/voice over like the original
snow wyvern
it is nice to have that kind of tech, but at the same time a bit concerning
dense tapir
Oh, anything good can, and will, be used for evil. It is the yin/yang of the cosmos so I am not going to stop something due to fear of the evil that will use it.
snow wyvern
Just last month, two families in Canada were scammed by fraudsters using this AI voice mimicry tool
dense tapir
Cool, it isn't the tools fault.
snow wyvern
probably not the same AI tool, but serves the same purpose
that's true
not AI's fault
dense tapir
I am just saying if people are stupid then so be it. Trust nothing, or anyone, and always be alert/on your toes. How I live and if more did less would be conned.
Being conned is nothing new it just evolves.
Just wait when Holograms arrive
Will not even be able to trust your own eyes
snow wyvern
But if one day you receive a call and the voice on the other end is that of a familiar family member, it's hard for most people to be cautious, isn't it?
dense tapir
No, not really
I know the situation and honestly trust but verify
The problem is too many are trusting and that is where the con artist thrives for centuries
What I hate are the ones who con the elderly. That makes me want to do bodily harm on them.
because the elderly are easy con targets as they are so trusting.
long trench
I've been working on a concept art lora for quite a while... it's getting better...
teach
how do i make ref sheets
hasty nova
it does kinda mess up buildings (right is with my lora, left is without), though this could have just been a bad seed for all I know.
hasty nova
how do i make ref sheets
you could add "character turnaround, multiple angles, concept art"
long trench
oh ):
hasty nova
Try googling it
long trench
I have
hasty nova
oh :(
long trench
and i either get errors
or it just dosen't load the lora
so the pictures look mushy
hasty nova
difference between using lora and not
long trench
right looks banger
hasty nova
I wasn't looking for an interior though
hasty nova
at least 4 but 8 is much better and gives you access to many more tools
and make sure you check vram, not normal ram
long trench
shit, i need to delete so much shit
im just keep serachin
or does anyone else know?
dense tapir
split rover
wispy nest
how do i make ref sheets
character design reference sheet for a game, 2d, - I used this before
well + quality, style and whatever you else use at the start + actual prompt after
ripe cedar
how do i make ref sheets
The new controlnets I've been using now consistently get me good hands, but the other stuff is messed up.
grave pilot
sinful mirage
negative embeds
sinful mirage
wispy nest
warped obsidian
Hello, I would like to ask how I can do generation based on a given image, but with different layout (only as a part of the generated picture)? For example, how I can print a picture onto a t-shirt?
Actually, I trained a lora to generate sticker sheet, and it works well for text prompts:
and now I want to generate new sticker sheet with a given image, like :
glossy herald
hey there
what"s that shark ? x)
yeah explain, but I don't do VC, I got 4 convos going right now
you tried the Visual novel tutoriel and the guides ? what are your questions on it ?
if I get some time I'll try to give you some VC help, but tbh, I need time for myself when I get a break those days. we get quite some people, and it can get exhausting. written help is a lot easier to me, and I'm quite efficient with examples on that format too
it should be possible, I saw a video suite of custom nodes that had a "train UNET" node
but I haven't seen a full training in it yet
turbid crest
i meaaaaaaaan............. with real life models taking high quality pictures from professional camera lenses and so on this doesnt look too far fetched from the real stuff i did zoom into it a bit to hide some weird unrealistic hair but if it was sent to me like this i wouldnt tell if its ai generated tbh @glossy herald its not obvious for me xD
even the hair isnt so obvious that its generated but then again it could be my untrained eye
glossy herald
i meaaaaaaaan............. with real life models taking high quality pictures fr...
it gives a video game kind of feel.
I would add "blender, plastic doll" in negative here.
here is one of my most photo realistic ones
less photorealistic, but cat
dry crow
i meaaaaaaaan............. with real life models taking high quality pictures fr...
the nose shadow would be an indicator
tawdry geode
i meaaaaaaaan............. with real life models taking high quality pictures fr...
The face is too simple, the mouth is too small compared to the nose, the nose is too simple, the ear is weird but you could remove it, the character looks anime, the skin is barbie-esque the hair on the right is too big and also not seeing the forehead can give you a guess to confirm all these imprecision, the neck is bad, the hair follows the jawline. Only the eyes and eyebrows are realistic, and maybe the nose if you would add grain. The placement of the eye nose and mouth are good too, but since I cant see the forehead I cant say it would break with the forehead
what you have is the idea of a face, exactly what stable gives 99% of the time
glossy herald
split rover
pig
glossy herald
pig
really love the 0.2 credits upscale on download feature by the way. very very nice
this helps a ton on XL 🙂
I'm having quite some fun with it !
when XXL though, 15T params
x)
split rover
when XXL though, 15T params
soon ;D
glossy herald
split rover
this helps a ton on XL 🙂
looking forward to getting the larger native res one out soon
glossy herald
I'm sure this won't be a problem for my VRAM 
split rover
glossy herald
wow
that's a nice GPU you got here. RTTTTTTTTTTX 5000
this would deserve to be a gif emote though :p
split rover
have a bird on a 🥔
glossy herald
bird's natural resting place 🙂
glossy herald
have a bird on a 🥔
and have "a bird on the house" :p
I'll send it to the Edit tab
and outpaint this
crystal deltaBOT
glossy herald
sorry Niting, no bot
split rover
@glossy herald I needed to clean off your 🐷
glossy herald
and have "a bird on the house" :p
outpainted
I wish we could change the size of the "dream" window
keeping the same inner generation resolution I mean
like I wanted to work on the bird specifically, and do him higher res=> downsample the inpainted area
well
I'm just loving it like it is too
but yeah, invokeAI is the best comparison I have for this Edit tool
and that's really nice to have there
split rover
haha yea, prob good for #1010312296804917470 I just train the models heh
glossy herald
yep, I know ^^
split rover
I really like invoke, good system
glossy herald
I'm just spitballing ideas
but I'll post in the feedbacks too
invoke is really a nice system yeah 🙂
split rover
glossy herald
wow ! he learned Surf TM !
split rover
though in theory once this is in the api and out it should be able to be placed in whatever front end 😄
Saw some people using the beta one on the api in unreal today
split rover
glossy herald
smoky oak
I think I am ready to resume work on my album >:D
Maybe les intense
more like a...
:>
wispy nest
trying out 1 word prompts sometimes leads to results that make me think SD is trolling me. I present to you "Audacious" (which I suppose it is, but come on) -_-
also: doing 1 word prompts usually has a lot of them that just try to spell the word and make some mishmash of letters...this is the first time i've seen it spell the entire word legibly
torpid rune
or this rn
Man! Thats awesome 😆
torpid rune
or this rn
Do you use those tools or elements you've said to create this?
stuck arrow
wooden silo
d
wispy nest
Cultist 1: "Is the sacrificial cow ready?" Voice from the back of the room: "I am"
dense tapir
fading whale
anyone could help me about this?
rancid quiver
I'm batman - perpetual diffusion
wispy nest
hazy kite
Monke
