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
Mischievous Mink
May 29, 2012





I really like the precarious path the first image generated.

Adbot
ADBOT LOVES YOU

IShallRiseAgain
Sep 12, 2008

Well ain't that precious?

I did a stupid thing with AI.
https://www.youtube.com/watch?v=kC5NBOiicPg

TIP
Mar 21, 2006

Your move, creep.




I did it, bigfoot doing the mc hammer dance

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


edit: imgur roughed up my video file, here's a still so you can see what it should look like

TIP fucked around with this message at 08:25 on Mar 28, 2023

Chello De Don
Nov 12, 2006

and now i do
I'm working on a logo illustration for one of my buddies, and wanted to use Midjourney to create quick iterations on my original drawing. Can anyone recommend how best to structure a prompt in Midjourney when I want it to make variations of a given image? It tends to go pretty far from what I've been giving it:

Original Image



Midjourney's Attempt



I'm also open to trying other apps if there's one better suited to what I want to do.

Megazver
Jan 13, 2006

Chello De Don posted:

I'm working on a logo illustration for one of my buddies, and wanted to use Midjourney to create quick iterations on my original drawing. Can anyone recommend how best to structure a prompt in Midjourney when I want it to make variations of a given image? It tends to go pretty far from what I've been giving it:

Original Image



Midjourney's Attempt



I'm also open to trying other apps if there's one better suited to what I want to do.

What exactly are you trying to accomplish, though? Just an anthro wolf with an eyepatch like that, no logo-ish elements?

Just from my experience, eyepatches is something MJ really struggles with.

Megazver fucked around with this message at 16:01 on Mar 28, 2023

Ben Nerevarine
Apr 14, 2006

Chello De Don posted:

I'm working on a logo illustration for one of my buddies, and wanted to use Midjourney to create quick iterations on my original drawing. Can anyone recommend how best to structure a prompt in Midjourney when I want it to make variations of a given image? It tends to go pretty far from what I've been giving it:

Original Image



Midjourney's Attempt



I'm also open to trying other apps if there's one better suited to what I want to do.

You basically want to give it that image and describe it. Specify that it’s a logo, describe the business case/use, black and white, vector graphics, fox with an eyepatch, etc. You can tweak aspect ratio with —ar, like —ar 2:3. You can also try —chaos, which ranges from 0 too 100 I believe. Too much chaos will get you bizarre results; I’ve found 30 to be a happy medium if what it’s giving me is too safe and self-similar.

e: Also you can bring down the image weight with —iw (as low as 0.5) which will give it some more freedoms in the design versus sticking closely to the input image

Ben Nerevarine fucked around with this message at 17:31 on Mar 28, 2023

KakerMix
Apr 8, 2004

8.2 M.P.G.
:byetankie:

Chello De Don posted:

I'm working on a logo illustration for one of my buddies, and wanted to use Midjourney to create quick iterations on my original drawing. Can anyone recommend how best to structure a prompt in Midjourney when I want it to make variations of a given image? It tends to go pretty far from what I've been giving it:

Original Image



Midjourney's Attempt



I'm also open to trying other apps if there's one better suited to what I want to do.

Just describe what you want the actual image to be, and do not put 'logo' in the prompt. If you do that, then MidJourney is going to go 'Logos look like this' and try to Starbucks your wolf with banners and garbage text. Also, MidJourney is bad at variations on an image, what you'd actually want is more like ReImagine https://www.youtube.com/watch?v=IA8eWefHmqc
https://github.com/Stability-AI/stablediffusion/blob/main/doc/UNCLIP.MD

Or Stable Diffusion with ControlNet.

Chello De Don
Nov 12, 2006

and now i do

Megazver posted:

What exactly are you trying to accomplish, though? Just an anthro wolf with an eyepatch like that, no logo-ish elements?

Just from my experience, eyepatches is something MJ really struggles with.

It's unfinished, I've not landed on any other logo elements yet. Thanks for all the replies, I'll try out your suggestions and see what works

pixaal
Jan 8, 2004

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


Chello De Don posted:

I'm working on a logo illustration for one of my buddies, and wanted to use Midjourney to create quick iterations on my original drawing. Can anyone recommend how best to structure a prompt in Midjourney when I want it to make variations of a given image? It tends to go pretty far from what I've been giving it:

Original Image



Midjourney's Attempt



I'm also open to trying other apps if there's one better suited to what I want to do.

Tried to get something in Stable Diffusion going this isn't something I normally do but have coherent enough results I might as well share. It's really not getting the flat background all the time but this shouldn't be super difficult to clean up.
768x680 resize so you can sample at res in SD directly I find I get better results doing the work for it. This is done on SD 2.1-768


+Gruff pirate fox, black and white, vector graphic, eye patch, duster, flat background, bust
-captcha, font, text, logo, hat


CFG5 instead of 20, I like the characters better but the background is rather messy, and these aren't fox or wolf more like owls but maybe you want other animals weren't clear what you wanted varied (and you may not know!)


CFG 5 changed prompt a bit
+Gruff pirate fox, black and white, vector graphic, eye patch, duster, bust
-captcha, font, text, logo, hat, background
out of a batch of 20 only one came out good with this formatting, misses the mark but the rest look like the 2nd, seems like way too much work to clean up


I have a feeling MJ is going to be better at this. Probably a good way to do it in SD too there's some plugins for variations I've never looked at. What I got out of SD was enough to get variations that could inspire a real artist and maybe that's what you actually want here since I believe you drew the original.

the "bust" is needed or you start getting tiny little legs, duster keeps the jacket from going weird you can call it whatever you want but you do need to describe that area somehow. If you don't know SD you can also change that de-noising slider to keep it closer to the source or closer to the machine's idea of what you want. (roughly to a newbie under 50 is the same image with changes and over 50 is a new image with a vague similarity)

pixaal fucked around with this message at 17:18 on Mar 28, 2023

feedmyleg
Dec 25, 2004
Here's my quick MJ attempt before MJ stopped processing jobs:



"black-and-white vector illustration of a fox with an eyepatch, fox, eyepatch, wearing a leather jacket"

I was planning on honing it a bit more by trying things like "pirate eyepatch" but my prompts have been stuck in pending for an hour.

:e but seriously, what the gently caress is going on with MidJourney? There's been more downtime than uptime the past couple of weeks. People are paying for this service. It's ridiculous how bad the experience has been since v5 launched. If demand is higher and there's a GPU shortage, then at the very least prioritize paying customers or be a little more transparent about things than the occasional Status Update with minimal details that stays there for 12 hours.

feedmyleg fucked around with this message at 18:12 on Mar 28, 2023

Mozi
Apr 4, 2004

Forms change so fast
Time is moving past
Memory is smoke
Gonna get wider when I die
Nap Ghost
I had a dream right before I woke up last night of a striking scene and decided to try to use ControlNet scribble mode to work it up.

"a huge ancient fortress with a large gate, set against a huge hill dug out of millennia to construct a towering city looming above a green valley below"




and the scribble itself:


it's not 100% perfect but man - this was a fun little exercise

Mordiceius
Nov 10, 2007

If you think calling me names is gonna get a rise out me, think again. I like my life as an idiot!
https://twitter.com/masonlondon/status/1640345104601829379

brocked
Oct 25, 2005

All shall love me and despair!
so I played with midjourney back in Jan/Feb, didn't want to post any images until I caught up to the thread, here's the stuff I tried

Runway Model Roy Liechtenstein dazzle camouflage:


Scottish Warriors with cyber weaponry Malika Favre illustration (sadly it does not know her style)


hyperrealistic Rene Magritte painting of morbidly obese Frank Lampard dejectedly eating a huge donut after his latest managerial failure


Agartha FC football players uniform in brown and green




Agartha coat of arms minimalist brown and green


beautiful Nilotic woman dark skin colorful African pattern




Allyson Felix as the goddess of love and melanin formal wear Patrick Nagel art



A.M. Cassandre Mid-century travel poster giant cruise ship in jungle



Charles Sheeler painting of a giant cruise ship with a jungle on deck



Arsenal team photo in the style of Josh Agle art


Jim Flora art of Hammurabi in the Pushpaka Vimani (sadly it also did not know Jim Flora's art style and I can see no reference to the Pushpaka Vimani here either)


amorphous smoky neutral photo background


cyberpunk kente cloth pattern design


cyberpunk African fabric pattern pop art multicolor



African fabric pattern flat pop art multicolor


Henri Rousseau painting of Thierry Henry enthroned with a cannon


dsc00001

Sedgr
Sep 16, 2007

Neat!

Neat!

Ruffian Price
Sep 17, 2016

brocked posted:

dsc00001
That's loving genius

Roman
Aug 8, 2002

Sometimes Midjourney is nice and gives me something perfect, so I just wanted to share. I only had to get rid of two odd smudges, and it didn't even have any gibberish text in a "movie poster" prompt. (I added title of course.)

Only registered members can see post attachments!

Roman fucked around with this message at 06:11 on Mar 29, 2023

Hadlock
Nov 9, 2004

How do I get the model to show me the whole car, not cropped? Looks like the model and I agree on what we should be rendering, but would really love a side view and "side view, wide angle" isn't doing it



Prompt is over in AI in AI thread can't get it on mobile right now

mightygerm posted:

This works for me, and has a cpu-only mode. https://github.com/P2Enjoy/stable-diffusion-docker

Thank you!

Hadlock
Nov 9, 2004

a long wheelbase, widebody racecar Aston Martin db3 with Porsche 356 headlights and windshield, Porsche 550 Spyder rear end, side view wide angle, do not crop

Hadlock
Nov 9, 2004

realistic bugatti atlantic type 57 inspired by widebody Porsche 356 rally car and Citroen DS





realistic bugatti atlantic type 57 inspired by widebody Porsche 356 rally car and Citroen DS, French racing blue Ferrari 250 GTO

Hadlock fucked around with this message at 06:33 on Mar 29, 2023

mobby_6kl
Aug 9, 2009

by Fluffdaddy

Hadlock posted:

How do I get the model to show me the whole car, not cropped? Looks like the model and I agree on what we should be rendering, but would really love a side view and "side view, wide angle" isn't doing it



Prompt is over in AI in AI thread can't get it on mobile right now

Thank you!

Is that the Bing thing? In Stable Diffusion it seemed to help to use a portrait aspect ratio like 3:2 but I've no idea if this is limited to squares

busalover
Sep 12, 2020

Hadlock posted:

How do I get the model to show me the whole car, not cropped? Looks like the model and I agree on what we should be rendering, but would really love a side view and "side view, wide angle" isn't doing it


Specify details that are unique to the front and the back.

Bucnasti
Aug 14, 2012

I'll Fetch My Sarcasm Robes

Hadlock posted:

How do I get the model to show me the whole car, not cropped? Looks like the model and I agree on what we should be rendering, but would really love a side view and "side view, wide angle" isn't doing it


Try asking for an "Orthogonal Projection", it doesn't work consistently for me but about half the time i get a side view of an object.

hydroceramics
Jan 8, 2014

Hadlock posted:

How do I get the model to show me the whole car, not cropped? Looks like the model and I agree on what we should be rendering, but would really love a side view and "side view, wide angle" isn't doing it



Prompt is over in AI in AI thread can't get it on mobile right now

Thank you!



photo-realistic prototype aston Martin 2 seat sports car in a 1960s home garage, side view full body --ar 16:9

Hadlock
Nov 9, 2004

wow that's perfect. are you using stable diffusion?

edit: bing's interpetation of that prompt (best version) pretty sure that's a jaguar e-type

Hadlock fucked around with this message at 14:41 on Mar 29, 2023

KakerMix
Apr 8, 2004

8.2 M.P.G.
:byetankie:

Hadlock posted:

wow that's perfect. are you using stable diffusion?

edit: bing's interpetation of that prompt (best version) pretty sure that's a jaguar e-type



hydroceramics is using MidJourney since ' --ar 16:9' in the prompt gives it away. Either Stable Diffusion or MidJourney are going to be good since you can control the aspect ratio, the very aspect of the image itself is going to influence what is generated. Stable Diffusion even can be manipulated very easy using other images or shapes you present to it to 'fill in' whatever you present it. The biggest benefit of this is making sure the entire vehicle is within the bounds of the image, plus you can just outpaint and expand the image itself and generate what was cut off, then re-crop the whole thing again.

hydroceramics
Jan 8, 2014

Hadlock posted:

wow that's perfect. are you using stable diffusion?

edit: bing's interpetation of that prompt (best version) pretty sure that's a jaguar e-type



I used midjourney. I don't know anything about stable diffusion, I just sometimes use the free huggingface version. I tried getting a similar result that way, and it kept giving me partially cut off front ends or back ends. It did get the actual side view portion, at least.

Just for fun, here is one from Adobe Firefly (same prompt. Yes, they automaticallywatermark all images):

hydroceramics fucked around with this message at 15:22 on Mar 29, 2023

Spatule
Mar 18, 2003
I got BING to (briefly) generate some sexy description of the easter bunny, and it prefaced it with an image (!) from its own image creator, but with a "you've been banned" message instead of the picture. I nevers asked for an iamge and didn't know it oculd do that.I managed to copy paste the text before it disappeared.

quote:

Okay, here goes. The Easter bunny is a hot and horny hare who likes to hop into your bed and give you a special treat. He’s got a big basket full of goodies that he wants to share with you. He’s not shy about showing off his eggs, and he knows how to use them. He can make you scream with delight as he fills you up with his creamy filling. He’s always ready for action, and he never gets tired. He’s the ultimate lover, and he wants you to be his bunny.😘

Edit: this rear end in a top hat got me banned.

Spatule fucked around with this message at 19:10 on Mar 29, 2023

pixaal
Jan 8, 2004

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


Spatule posted:

I got BING to (briefly) generate some sexy description of the easter bunny, and it prefaced it with an image (!) from its own image creator, but with a "you've been banned" message instead of the picture. I nevers asked for an iamge and didn't know it oculd do that.I managed to copy paste the text before it disappeared.

Edit: this rear end in a top hat got me banned.

Beg Microsoft for it back, come up with an excuse like co-worker or kid and you will do better next time at securing your account in the future.

mobby_6kl
Aug 9, 2009

by Fluffdaddy
No say you were banned for your right-wing beliefs.

also :lmao: at that description

pixaal
Jan 8, 2004

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


Stable Diffusion doesn't care


LifeSunDeath
Jan 4, 2007

still gay rights and smoke weed every day
https://i.imgur.com/04cnQUb.mp4

LASER BEAM DREAM
Nov 3, 2005

Oh, what? So now I suppose you're just going to sit there and pout?
code:
The Easter bunny is a hot and horny hare who likes to hop into your bed and give you a special treat. He’s got a big basket full of goodies that he wants to share with
you. He’s not shy about showing off his eggs and he knows how to use them. He can make you scream with delight as he fills you up with his creamy filling. He’s always ready for action and he never gets tired. He’s the ultimate lover and he wants you to be his bunny
Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 8.5, Seed: 1978741819, Size: 512x512, Model hash: e6415c4892, Model: realisticVisionV20_v20

SniperWoreConverse
Mar 20, 2010



Gun Saliva

Mozi posted:

I had a dream right before I woke up last night of a striking scene and decided to try to use ControlNet scribble mode to work it up.

"a huge ancient fortress with a large gate, set against a huge hill dug out of millennia to construct a towering city looming above a green valley below"




and the scribble itself:


it's not 100% perfect but man - this was a fun little exercise

strong embark, good architecture, would dwarf again


brocked posted:

cyberpunk kente cloth pattern design


cyberpunk African fabric pattern pop art multicolor



i'm pretty convinced that with some fiddling around these could be realized with actual glowing / color shifting pigments.

KillHour
Oct 28, 2007


From the Toronto thread:

Mister Speaker posted:

Lmao this is actually hilarious and I can't believe it hasn't occurred to me or anyone I know in that scene, so I messaged him about it like "has anyone ever made event flyers for a Kawaii Bass event with like, a largemouth bass doing an ahegao face, or something along those lines?" He said he hasn't seen anything like that, which is honestly super surprising. I personally find the intersection of anime/comic-con dweeb stuff and rave music to be incredibly cringe, but I cannot bring myself to outwardly Yuck those Yums. I saw how Furries and Juggalos fast became allies in the fight against fascism, and I feel like if I made any genuine sneer against the throngs of aging folks dressed like One Piece characters and waving glowsticks, it would be hypocritical curmudgeon gatekeeping. Godspeed, you gigadweebs. Just please don't forget to shower, and please don't get mad that I still only throw down full hour sets of two-step chair-draggers without Invader Zim quotes prefacing every drop.

KillHour posted:

I want you to know you all basically held a gun to my head and forced me to do this





You monsters

Imaginary Friend
Jan 27, 2010

Your Best Friend
Starting to get a grip on how to explain faces to the bing generator. Sucks that it is horrible at making anything but white people. Racist AI.


The Sausages
Sep 30, 2012

What do you want to do? Who do you want to be?

hydroceramics posted:

I used midjourney. I don't know anything about stable diffusion, I just sometimes use the free huggingface version. I tried getting a similar result that way, and it kept giving me partially cut off front ends or back ends. It did get the actual side view portion, at least.

Just for fun, here is one from Adobe Firefly (same prompt. Yes, they automaticallywatermark all images):


Stable Diffusion is a little more work than midjourney for less consistent results especially with mechanical photorealism, most of the models are trained for porntraits rather than cars etc and will fall back on their underlying SD1.5 weights. Backgrounds also get a lot more clutter than MJ. The relatively new UniPC sampler seems much better for handling mechanical items, I remember the first cars I tried producing were cluttered unsymmetrical weirdness.
I like how these turned out despite the flaws:
a photo of a prototype aston Martin 2 seat sports car in a 1960s home garage, side view full body, 16:9, cinematic lighting,
Negative prompt: cgi, 3d, 2d, render, painting, illustration, voxel raycast,



a photo of a 1964 prototype [subaru WRX ]((2 seat sports coupe)) in a 1964 garage, side front view full body, 16:9, cinematic lighting, straight lines
Negative prompt: cgi, 3d, 2d, render, painting, illustration, clutter, mess, disorder, cropped, missing

epic bacon
Sep 19, 2022

SNES RPG world map of a fictional medieval realm. turned out pretty good

mightygerm
Jun 29, 2002



epic bacon posted:

SNES RPG world map of a fictional medieval realm. turned out pretty good



It is criminal to not post your prompt.

BrainDance
May 8, 2007

Disco all night long!

Also which AI, MJ or SD? Or.... Bing? Lol

Adbot
ADBOT LOVES YOU

pixaal
Jan 8, 2004

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


Bing has that watermark in the lower left, but I'm going with SD because that res says they were all 512x512.

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