Anonymous at Tue, 11 Jan 2022 11:47:58 UTC No. 875189

>>875184
image morphing algorithms work by creating a mesh that has a series of points you then line up to match to each image you wanna blend.
It can be anything not just faces or front faces. You make a custom mask for whatever you wanna warp and you line up your points and
morph the mesh halfway between either target while blending halfway between either texture.

This is very old tech that has been around a long time, Michael Jacksons video for 'Black or White' used it back in 91.
we replicated the effect using 3D studio max in the schools computerlab back in 99.

They're easy to do and you can trivially create one in whatever 3D program you run.

Anonymous at Tue, 11 Jan 2022 11:53:38 UTC No. 875192

>>875184
too bad you only get only one angle. But you can still do some projection mapping with something like Keen tool Face builder. Or just morph a base head by hand and then just project on the face for the texture. But this is only viable for oneoffs, not mass production.

Anonymous at Tue, 11 Jan 2022 12:16:59 UTC No. 875197

>>875189

Anonymous at Tue, 11 Jan 2022 13:11:30 UTC No. 875207

>>875197
Just do like this Nathan. You can morph anything the same way, just make a custom mask that outlines and marks the features on any similar two pictures you wanna combine. Aeroplanes, horses, motorcycles or penuses etc, no plugins required and works for anything.

Anonymous at Tue, 11 Jan 2022 13:50:06 UTC No. 875212

>>875184
>like this
get a big pile of 3d faces and train a neural network on it

Anonymous at Tue, 11 Jan 2022 14:49:49 UTC No. 875224

Train a GAN on an absurd number of 3D scans
Then train a GAN to retopo scanned faces

Anonymous at Wed, 12 Jan 2022 03:00:24 UTC No. 875374

>>875224
who even owns the millions of high quality face scans and the computing power required?
should be a well funded company with a risky business plan, or a huge corporation with money to burn on R&D

Anonymous at Wed, 12 Jan 2022 14:17:53 UTC No. 875438

>>875374
Nvidia most likely. EG3D is the closest thing as far as I am aware

Anonymous at Wed, 12 Jan 2022 14:46:35 UTC No. 875445

>>875184
Just make a face-gen with blendshapes or custom bones like they've been doing in videogames for years

Anonymous at Wed, 12 Jan 2022 17:47:45 UTC No. 875465

>>875184
> Its just a random value script on metahuman, right?

... works, but you will get all kinds of shit ugly faces because your script needs some kind of "awareness" how a population looks like - not all are shit ugly or extremely beautifull. your script will just randomly shit out freaks of nature, which will be absolute useless, stuff like the elder scrolls character editor produces ...

the other point is, which style! real faces, or stylized? when you can model by yourself and your creations doesnt look shit ugly, modell a male/female shit ugly, medicore and beautifull faces, then you can create >5++ different faces each hour, you just need to rigg your faces good enough that it can be morphed into all kinds of variations, the pluspoint is, they dont like look souless trash ...

Anonymous at Wed, 12 Jan 2022 21:39:01 UTC No. 875504

>>875207
Always suspected some manual work was in play. In the 90s when I was a kid I got Photoshop and Camedia but was never able to replicate this simply using transparency or fading. Cheap euro TV stations sure did tho

Anonymous at Wed, 12 Jan 2022 23:08:56 UTC No. 875517

>>875438
No.

Epic Games (who owns Unreal, estimated at 28 billion $) couldn't even get more than 50 facescans for their MetaHuman project.

For photogrammetry you need to pay a person to get inside a special room then spend hours if not days to clean the data for each scan. For 2D pictures of people all you need is a bunch of idiots posting their faces on "free" social media platforms which make money by selling said data.

The reason why we're seeing all this 3D scanning tech in smartphones is because companies are hoping people will start freely uploading their 3D scanned faces the same way they do with pictures. We're probably gonna start seeing governments around the world implement a Digital Identity thingy which will work exactly like an identification card except they're going to have a 3D scan of your face instead of a picture. The purpose of this is of course not the Digital Identity itself, but the data collected through the production of a DI.

Anonymous at Thu, 13 Jan 2022 00:59:36 UTC No. 875540

>>875517
meds, in case this isn't bait

Anonymous at Thu, 13 Jan 2022 01:22:09 UTC No. 875543

>>875540
Data (and data-gathering) is one of the most lucrative businesses on the planet for the last decade. This is not a secret. Even boomers know this. Every site or app that is "free" makes money by selling its data. Facebook, YouTube, even 4chan are just data gathering factories. Just google it, normie.

I bet you also got vaccinated.

Anonymous at Thu, 13 Jan 2022 01:50:17 UTC No. 875545

>>875543
>makes money by selling its data

most of these companies value is being a tool for bigger interests.
Companies like Boeing or BP or Lockheed don't play ads on jewgle to get plebs to buy their bombs or crudeoil, it's a gift with strings attached.
manipulating public perception for them
creating artificial trends
creating artificial problems
offering approved solutions
deboosting certain discourse
the data becomes most valuable for studying how to get better at shaping public interests.

It's become for the most part the same thing as offline Media a propaganda dispensary.

Anonymous at Thu, 13 Jan 2022 05:15:16 UTC No. 875574

>>875465
While I agree, we probably can both concur that possibly within a decade we will have something extremely potent able to do that

Anonymous at Thu, 13 Jan 2022 06:10:35 UTC No. 875580

>>875184
I'm very interested in this. It would be pretty nice to generate ultra-realistic crowds without having to pay likeness royalties out the wazoo.

Anonymous at Thu, 13 Jan 2022 15:54:29 UTC No. 875639

>>875184
that tech could be used to create CP, so it obviously should be tightly controlled by the government and kept away from commoner taxpayers

Anonymous at Thu, 13 Jan 2022 18:02:25 UTC No. 875647

>>875639
CP is only when you have real photographs of actual humans. When you do 3d and move values around with scripts on a digital mesh, youre at worst doing lolicon which is 100% legal.

Anonymous at Thu, 13 Jan 2022 19:48:30 UTC No. 875666

>>875639
>kids could be used to create CP, so they obviously should be tightly controlled by the government and kept away from commoner taxpayers
fix'd.

Anonymous at Fri, 14 Jan 2022 12:54:55 UTC No. 875785

>>875184
no. if you did random values you would end up with weird alien people with no gender or race. when an AI comes up with a face, it generates which race, age and gender it should be first and goes from there. there is some randomness involved, but most of it is informed by piecing real faces together.

Anonymous at Fri, 14 Jan 2022 12:58:59 UTC No. 875786

>>875785
> if you did random values you would end up with weird alien people with no gender or race
but the media has been saying for years now that gender isnt real. Also, I would assume that you could put in functions to "train" your output values over time if you so desired

Anonymous at Fri, 14 Jan 2022 14:19:23 UTC No. 875797

>>875786
sure, but how one part of the face looks informs how the rest should. if you have a strong chin, you would also have a stronger nose etc regardless of race. you would eventually get a lot of same face with some slight variation

Anonymous at Wed, 19 Jan 2022 00:34:25 UTC No. 876803

>>875184
>Its just a random value script on metahuman, right?
Yeah but you don't want to randomize everything at once.
You probably want to randomize in stages.
-Random ethnicity (asian)
--Random sub ethnicity (han)
---Randomize skin tone for a range
---Randomize eyelid shape within a range
Some things will be ranges (height on a bell curve)
Some things will be choices (eye color with probabilities for each)
If you want to do mixed races, you would need to set the ranges and choices to be what is available with the two races.

Anonymous at Thu, 20 Jan 2022 07:30:25 UTC No. 877043

All these tools which make random something, the only valuable part about them is always the dataset. If you have the extensive dataset of face photos, you can always use something like Phantamorph to make as many custom faces as you want, and exactly in the way you want.

Anonymous at Thu, 20 Jan 2022 07:34:36 UTC No. 877044

I was superhyped about these random face sites, cause I wanted to generate bunch of unique faces for references. After making about 20 faces, I discovered they have about 2 or 3 templates for ethnicities, so you just get the same face with different texture.

Anonymous at Thu, 20 Jan 2022 18:29:40 UTC No. 877109

>>877044
>the same face with different texture
that's not how deep learning works, retard

Anonymous at Thu, 20 Jan 2022 18:42:50 UTC No. 877111

>>877109
>le deep learning meme

Anonymous at Mon, 31 Jan 2022 05:27:49 UTC No. 879133

>>875184
She cute

Anonymous at Mon, 31 Jan 2022 14:09:16 UTC No. 879176

>>875184
Only in clown world can we fall in love with people who don't even exist.

Anonymous at Mon, 31 Jan 2022 14:11:59 UTC No. 879178

>>879176
stop thinking with your dick.

Anonymous at Thu, 3 Feb 2022 18:18:37 UTC No. 879877

>>875192
Too bad I can't make a fuckable meat golem out of OPs picture and collected human body parts!