Register a SA Forums Account here!
JOINING THE SA FORUMS WILL REMOVE THIS BIG AD, THE ANNOYING UNDERLINED ADS, AND STUPID INTERSTITIAL ADS!!!

You can: log in, read the tech support FAQ, or request your lost password. This dumb message (and those ads) will appear on every screen until you register! Get rid of this crap by registering your own SA Forums Account and joining roughly 150,000 Goons, for the one-time price of $9.95! We charge money because it costs us money per month for bills, and since we don't believe in showing ads to our users, we try to make the money back through forum registrations.
 
  • Post
  • Reply
pixaal
Jan 8, 2004

All ice cream is now for all beings, no matter how many legs.


RIP Syndrome posted:

*yanks politely*

https://i.imgur.com/Cw633aI.mp4

Adbot
ADBOT LOVES YOU

RIP Syndrome
Feb 24, 2016

Nice. How's it work, can you tell it to make the eyes light up too?

feedmyleg
Dec 25, 2004
Nah. If you want it to resemble the input image you can't add a prompt. If you add a prompt it creates something that looks almost entirely, but not completely, different than the input image.

Swagman
Jun 10, 2003

Yes...all was once again peaceful in River City.
love to whisper to my cuda cores and gaze into the mouth of madness



















pixaal
Jan 8, 2004

All ice cream is now for all beings, no matter how many legs.


RIP Syndrome posted:

Nice. How's it work, can you tell it to make the eyes light up too?

That's just feeding the image without a prompt to Gen 2 Runway (you can tell from the morphing icon in the bottom right) I still have some free credits from free signup, no idea if they ever refresh
http://app.runwayml.com/

Make sure to use gen 2, there's not really a way to directly link to it that I can find it shoves your team name in there and if someone isn't part of your team the link kicks them to login.

I'm sure better video stuff is coming but right now it's a massive crapshot.

pixaal fucked around with this message at 18:27 on Aug 8, 2023

RIP Syndrome
Feb 24, 2016


Love/hate these a lot. Good prompting and a long-ish selection process?



Anyway it's back2work for me

hydroceramics
Jan 8, 2014
I've had a little luck recently with some mythological beasts. Midjourney struggles with them quite a bit, but I was able to get some fairly close results with prompts like "hybrid animal half horse half eagle." Historically, without "hybrid animal" half eagle half horse would veer into kronnenberg territory. No style prompts were given:











The results are good enough that you don't even have to guess what the animal combinations are.



hydroceramics fucked around with this message at 19:23 on Aug 8, 2023

XYZAB
Jun 29, 2003

HNNNNNGG!!

RIP Syndrome posted:

Great. I'll be posting a lot less soon (work), making you the designated spam-slack-picker-upper.

Promotion accepted.

Now, who wants breakfast?



Also, I think I just uncovered a new origin myth. Am I witnessing the birth of a new religion???

XYZAB fucked around with this message at 19:25 on Aug 8, 2023

hydroceramics
Jan 8, 2014

XYZAB posted:

Promotion accepted.

Now, who wants breakfast?



Also, I think I just uncovered a new origin myth. Am I witnessing the birth of a new religion???



A breakfast food known as a Roman Glacier;

XYZAB
Jun 29, 2003

HNNNNNGG!!

hydroceramics posted:

A breakfast food known as a Roman Glacier;


It's like they always say: if it isn't a locally grown citrus monstrosity handcrafted by an Austrian leatherworker at the highest peak of the Swiss Alps, you can't call it a true Roman Glacier.

XYZAB
Jun 29, 2003

HNNNNNGG!!
Where my crystal gazing horse fanciers at?



I added ((((((realistic horse penis)))))) to the same prompt that gave me all of those breakfast crystal things and it's giving wild horse girl energy, but I haven't seen a single realistically depicted horse penis yet. Or any horse penis, for that matter. What are the rules for brackets that I can more clearly tell it I want less emphasis on "realistic horse" and more emphasis on "horse penis"?

And check out these two breakfast lovers forever trapped in horse crystal hell. :cool:

Humbug Scoolbus
Apr 25, 2008

The scarlet letter was her passport into regions where other women dared not tread. Shame, Despair, Solitude! These had been her teachers, stern and wild ones, and they had made her strong, but taught her much amiss.
Clapping Larry

Asian female human avatar of the Vermilion Bird of the South.red hair, wings, photorealistic, intricate, detailed, sony a7 III Lightroom presets,Kodachrome



Raccoon in a suit giving a TEd talk


penguin, badz maru,penguin,as a ninja, sanrio (it needed 'penguin' twice for some reason)


Oni-no-Kali.photorealistic,intricate,detailed, jungle,action pose, sony a7 III lightroom presets,kodachrome --ar 2:3


fantasy Baghdad,Persia, day light, photorealistic, detailed , intricate, aerial view --ar 3:2


fantasy Indian City, photorealistic, detailed , intricate, aerial view --ar 3:2


top down fantasy middle east continent map, intricate, detailed

I use Midjourney a lot for tabletop RPGs

pixaal
Jan 8, 2004

All ice cream is now for all beings, no matter how many legs.


RIP Syndrome posted:

Nice. How's it work, can you tell it to make the eyes light up too?

You wanted a light?

XYZAB
Jun 29, 2003

HNNNNNGG!!

XYZAB fucked around with this message at 06:41 on Aug 11, 2023

hydroceramics
Jan 8, 2014

XYZAB posted:

It's like they always say: if it isn't a locally grown citrus monstrosity handcrafted by an Austrian leatherworker at the highest peak of the Swiss Alps, you can't call it a true Roman Glacier.

Its only a Roman Glacier if it comes from the Diabetes region of Italy. Otherwise its just sparkling smorgasbord.

XYZAB
Jun 29, 2003

HNNNNNGG!!


And lastly, the first image out of about 1000 that looks even remotely plausible in the real world:



Countdown until my GPU explodes.

e:



XYZAB fucked around with this message at 21:31 on Aug 8, 2023

pixaal
Jan 8, 2004

All ice cream is now for all beings, no matter how many legs.





(Believe it or not the only difference is 2 CFG and 20 steps before upscale) There's still like 20 minutes of upscaling downscaling loop to make it look nice instead of having disconnected limbs but comfy makes that loop easy

pixaal fucked around with this message at 21:18 on Aug 8, 2023

Mescal
Jul 23, 2005











Some came out more realistic:







This series didn't turn out great, anybody recommend an artist name to make this concept work better?

Mercury_Storm posted:

black and white cartoon with bottom text "Death by cow tools." showing a human farmer surrounded by anthropomorphic cows standing on two legs holding disfigured but sharp tools made of calcified milk in the style of Gary Larson






I was definitely not expecting System Shock 2 or Dead Space enemies from this prompt lol

If you don't read this thread religiously, Mercury, you may have missed my series on Cow Tools. Please forgive the self-quote.

Mescal posted:



Eliasson / Hoefer



Giacometti / Cartier-Bresson



Ono / LaChapelle



Rachel Whiteread / Gregory Crewdson



Sol Lewitt / Hiroshi Sugimoto



Henry Moore / Cecil Beaton



Koons / Photographer unknown



Anthony Gormley / Steve McCurry



Hirst / Andreas Girsky




Doctor Zero
Sep 21, 2002

Would you like a jelly baby?
It's been in my pocket through 4 regenerations,
but it's still good.

XYZAB posted:

I hate to keep spamming this thread but this poo poo is blowing my mind.



:wtc:

pixaal
Jan 8, 2004

All ice cream is now for all beings, no matter how many legs.



Really large collections like this look good as timg

I miss the AI that understood cup sizes and you could say man with DDDDDD tongue and get horrors of a tongue swollen to hell (it looked like a bloated corpse but someone struggling for air at 128x128) SD with the concept of sizing dynamically would be nice.

pixaal fucked around with this message at 22:47 on Aug 8, 2023

Swagman
Jun 10, 2003

Yes...all was once again peaceful in River City.


















XYZAB
Jun 29, 2003

HNNNNNGG!!

pixaal posted:

Really large collections like this look good as timg

I miss the AI that understood cup sizes and you could say man with DDDDDD tongue and get horrors of a tongue swollen to hell (it looked like a bloated corpse but someone struggling for air at 128x128) SD with the concept of sizing dynamically would be nice.

Which AI was that?

I'm a bit disappointed with SD's inability to grapple with direct size prompts as well. All I wanted was larger than average sized hands with extra long fingers on those cat-frogs, or a face that's extra long, but it's like it completely ignores dimension/resize requests. I managed to get some mileage by asking for (((((big red sexual lips))))), but it more often used that prompt to cover everything in red, with a slightly lower possibility that there would even be lips at all, and if there were, it was 50/50 that they'd be prominent or not.

I'm champing at the bit to go full body horror whenever SD can grapple with those requests.

pixaal
Jan 8, 2004

All ice cream is now for all beings, no matter how many legs.


There's a good body horror Lora for 1.5 only time until it's in XL

KakerMix
Apr 8, 2004

8.2 M.P.G.
:byetankie:

XYZAB posted:



I'm champing at the bit to go full body horror whenever SD can grapple with those requests.



These systems have a language they speak that isn't like how you and I speak to each other, so asking for 'REALLY BIG HAND' doesn't mean anything to it because it is only going to summon images for tags it has and doesn't understand language well like that if the tags don't match. Asking for something like 'a dude in a blue hat' works because presumably there are lots of images with those tags, but 'big hands man' not so much. To get what you want you'd really want to inpaint to side-step context and just get a big hand. Or, like the model linked above, use user-created models to get whatever you want because boy there is a lot. Note that most trained models right now are SD 1.5, which is not XL which means it's 512x512 and has minor difference in how it prompts.

EDIT

pixaal posted:

There's a good body horror Lora for 1.5 only time until it's in XL

:argh:

Crapple!
Nov 1, 2019


Mescal posted:











Some came out more realistic:







This series didn't turn out great, anybody recommend an artist name to make this concept work better?

If you don't read this thread religiously, Mercury, you may have missed my series on Cow Tools. Please forgive the self-quote.

I would say most of them came out great! Just a couple of slightly wonky faces, but it's a really cool style.

XYZAB
Jun 29, 2003

HNNNNNGG!!

KakerMix posted:



These systems have a language they speak that isn't like how you and I speak to each other, so asking for 'REALLY BIG HAND' doesn't mean anything to it because it is only going to summon images for tags it has and doesn't understand language well like that if the tags don't match. Asking for something like 'a dude in a blue hat' works because presumably there are lots of images with those tags, but 'big hands man' not so much. To get what you want you'd really want to inpaint to side-step context and just get a big hand. Or, like the model linked above, use user-created models to get whatever you want because boy there is a lot. Note that most trained models right now are SD 1.5, which is not XL which means it's 512x512 and has minor difference in how it prompts.

EDIT

:argh:

Interesting. I figured I would have to get around to inpainting soon enough. Thank you both.

pixaal
Jan 8, 2004

All ice cream is now for all beings, no matter how many legs.


Put the flaming shot down Elmo

excuse me did you know you are on fire?



(courtroom sketch XL Lora on Crystal Clear XL)

RIP Syndrome
Feb 24, 2016

whoa

TheGoonspiracist
Jul 24, 2002

The terrible secret of space... :stonk: the Mods, they knew!

Just messing around with Bowie Tarot cards.

Roman
Aug 8, 2002

it's basically if D-ID fixed everything that made it not good.
they apparently only need 2 minutes of training on you to do this.
ngl the possible "virtual actor" applications of this tech interest me more than anything.
https://twitter.com/joshua_xu_/status/1689019874667024384

pixaal
Jan 8, 2004

All ice cream is now for all beings, no matter how many legs.


Roman posted:

it's basically if D-ID fixed everything that made it not good.
they apparently only need 2 minutes of training on you to do this.
ngl the possible "virtual actor" applications of this tech interest me more than anything.
https://twitter.com/joshua_xu_/status/1689019874667024384

I'm going to make Matt Berry and John H Benjamin argue over which super hero would win in a fight while dressed like space pirates.

pixaal
Jan 8, 2004

All ice cream is now for all beings, no matter how many legs.


Touch fuzzy get dizzy



Fixable in inpaining can't get that leg to connect with prompting without loving up the dizzies I need to build an inpainting workflow already for comfy

e: another Yoshi

pixaal fucked around with this message at 21:20 on Aug 9, 2023

Hadlock
Nov 9, 2004

RIP Syndrome posted:

Love/hate these a lot. Good prompting and a long-ish selection process?



Anyway it's back2work for me

What was your prompt for this one

XYZAB
Jun 29, 2003

HNNNNNGG!!
I'm having a problem with ComfyUI again, similar to last time but slightly different, wherein as soon as I open the .bat to get the process up and running, it just starts rendering automatically based on an interrupted prompt that it seems to have gotten stuck on from when I ctrl+C'd out of the cmd window yesterday. The UI is unresponsive to my attempts to shut it down, despite it showing a newly numbered process running in the queue pane every time I click "cancel," which spurs the creation of another newly numbered process running in the pane. I don't even have that little checkmark selected beside the batch count slider, which USED to be the way I was using to assuredly cancel out of a prompt execution.

The hell am I supposed to do now? It's not following any of the nodes either, it's stuck on an 80 step process each time despite changing that number, and the green progress bar that used to show on the ksampler is mysteriously absent.

Is this the rise of the machines that we've been warned about?

RIP Syndrome
Feb 24, 2016

Hadlock posted:

What was your prompt for this one


office cubicle cutaway, floating in the cosmos, seen from the outside, dog face, tilt view, dreamlike, by moebius, intricate, detailed, masterpiece

Steps: 66, Sampler: DPM++ 2M SDE Karras, CFG scale: 5, Seed: 2270973560, Size: 1024x1024, Model hash: 0b76532e03, Model: crystalClearXL_ccxl, Version: v1.5.1


Throwing Moebius in there, or any artist name really, feels like cheating. But I got extremely lucky with the dog (adding steps ruins it).

Roman
Aug 8, 2002

longer RunwayML Gen-2 video generations are coming
https://twitter.com/c_valenzuelab/status/1689380316916367360

pixaal
Jan 8, 2004

All ice cream is now for all beings, no matter how many legs.


XYZAB posted:

I'm having a problem with ComfyUI again, similar to last time but slightly different, wherein as soon as I open the .bat to get the process up and running, it just starts rendering automatically based on an interrupted prompt that it seems to have gotten stuck on from when I ctrl+C'd out of the cmd window yesterday. The UI is unresponsive to my attempts to shut it down, despite it showing a newly numbered process running in the queue pane every time I click "cancel," which spurs the creation of another newly numbered process running in the pane. I don't even have that little checkmark selected beside the batch count slider, which USED to be the way I was using to assuredly cancel out of a prompt execution.

The hell am I supposed to do now? It's not following any of the nodes either, it's stuck on an 80 step process each time despite changing that number, and the green progress bar that used to show on the ksampler is mysteriously absent.

Is this the rise of the machines that we've been warned about?

If I end the batch in CMD with control+C it always recovers to no queue, maybe reboot your computer? Restart, shutdown may cause windows to suspend instead. A1111 worked the same, though A1111 sometimes didn't release VRAM in early builds and needed a reboot or you'd have no VRAM.

Roman posted:

longer RunwayML Gen-2 video generations are coming
https://twitter.com/c_valenzuelab/status/1689380316916367360


neat, but oof it's obvious it's a bunch of nothing at this long. Really needs to accept multiple images so you can give it some idea of how the area around the subject looks. My problem with them is they want money outside of when they decide to give more free credits, and I'm not going to pay to beta test your software.

pixaal fucked around with this message at 23:41 on Aug 9, 2023

Sedgr
Sep 16, 2007

Neat!

top secret, ufo, darpa, classified, photo











Fun mix of photos and documents from that.

Hadlock
Nov 9, 2004

Roman posted:

longer RunwayML Gen-2 video generations are coming
https://twitter.com/c_valenzuelab/status/1689380316916367360

That is super cool. I looked it up, average shot length in 1930 was ~12 seconds, in 2014 an average shot length is 2.5 seconds according to wired

https://www.wired.com/2014/09/cinema-is-evolving/

And then from here, average shot length of a steven spielberg movie is 6.8 seconds

https://vashivisuals.com/category/one-sheets/average-shot-length-one-sheets/

Other statistics

Michael Bay 3.0s
Christopher Nolan 3.1s
Alfred Hitchcock 9.1s
Woodey Allen 17.5s
Sergio Leone 7.0s (the good, the bad, and the ugly)

Movies:
Indian jones raiders of the lost ark 4.4s
minority report 6.0s
boogie nights 11.2s
alien 3 3.6s
jay z - 99 problems (music vodeo) 0.7s
the exorcist 6.6s

anyways TL;DR an 18 second clip like that twitter thing is pretty neat. if you can reliably AI generate images with the same architectural background and "actor" you could string together a feature length movie using this tech right now. probably in 6 months to a year it'll be at the point where the first amateur/art student movies will start being released using this tech. In a year or two I can see this being a collaborative tool used to work with the director of photography on projects. You might even be able to "pre-shoot" an entire film shot for shot before you even get to casting

Adbot
ADBOT LOVES YOU

Hadlock
Nov 9, 2004

Here is a shot list for the matrix, kind of interesting to imagine running each of these frames through that algorithm

quote:

“Vashi Frames” is my project of capturing every individual shot from my favorite shows and films, compiling them into a single high-resolution image for cinematic analysis.

  • 1
  • 2
  • 3
  • 4
  • 5
  • Post
  • Reply