VToonify: Controllable High-Resolution Portrait Video Style Transfer | Hacker News Reader

VToonify: Controllable High-Resolution Portrait Video Style Transfer

mmlab-ntu.com

118 pointslnyan32 comments

As someone who is in touch with a lot of people acting as VTubers in order to keep privacy while being able to express themselves, I'm very happy to see this!

EbSynth gave Joel Haver a trademark style[1], and I feel this might be the next step.

1. https://www.youtube.com/watch?v=tq_KOmXyVDo

Agree in principle, though this likely might be easily reverse engineered to extract the masked face, so might actually be dangerous if someone was depending on it to hide their identity.

I'm terrified of our current and future problems around this and similar tech.

I'm intrigued. Can you provide some examples of current vs future problems?

Seems like a form of virtual eugenics, a sensorship or the body in a way that's oppressively homogenizing. I feel like this is part of why zuck leverages cartoons to market metaverse for corporate settings, or why all the retail-safe anodyne pop muzak is full or autotune and flawless software-based overproduction. Not to be overly dramatic but imagine you are enraged but all anyone can see is a tranquil smile and your words replaced by positive endorsement; you have no mouth and you must scream

On the plus side I can see this being used to self-anonymize by individuals who want to protect themselves from facial recognition or other forms of id from undesirable parties, but that's more of a coping tactic for a society that fails to incorporate any honest ethical QC into product development to begin with.

I don't think this is directly more homogenizing than (say) makeup, which has the added property of being cemented by centuries of public presentation and beauty expectations.

That being said, I think there are probably lesser social ills that come from these technologies, ills that we'll have to adapt to. For example, we know that human interaction and emotional feedback are an important part of child development; it's as of yet unclear what effect this will have.

v-tubers must fill you with existential terror. Like, you look at Korone but all you see is Moloch.

Current example would be the filters that modify appearance that's used a ton on influencer platforms. We've been photo shopping for years but now anyone can create a false identity online. The bad side of this is like catphishing and unrealistic body standards causing non obvious mental health issues across generations.

Future would be continuing to push the boundaries of auto formatting filtering styles based on your preference or the populations preference. Scamming, impersonation, money to be made from gullible people.

This is just the surface. I'm sure there are tons of fictional books exploring this or people with more destructive ideas.

Just because I'm terrified and am listing the bad, that doesn't mean I don't recognize the good that can come from it. I'm not sure if that out weighs the bad yet.

Staring at a filtered, "beautified" version of yourself trying to capture the perfect selfie directly causes facial dysmorphia by making the you in the mirror look ugly compared to the you in the phone display.

Unrealistically large eyes, like in Disney's Frozen.

There is a name for this:

https://en.wikipedia.org/wiki/Supernormal_stimulus

Looks awesome. Will make making toons easier.

Quite excited for the eventual application of AI art like this into game dev and film.

Will be a very interesting time when a solid programmer and artist combo can generate, with AI tools, an entire studio's worth of content (animation, 2D/3D, etc) for their projects.

Make sure to click on the video link to get the full feel of how this performs: https://m.youtube.com/watch?v=0_OmVhDgYuY

Amazing set of projects with open source release. Amazing to see!

agree, is it a university creating all this? from what i can see it looks like it. will have a look at their stuff when i get some more time.

Stays stuck on a loading screen forever on mobile Safari.

The built-in Readability fork in Safari is impressive - even the embedded videos survived.

Can this be turned into a video game shader?

That's pretty cool, thanks for sharing

Fantastic to see more open source ML.

Creepy, offensive, dehumanizing

Magical, limitless, empowering.

I've never been so excited for our future.

We won't have to spend 10,000 hours to get good at something. We'll no longer be required to specialize and be locked into our choices.

I can't wait for this to pan out. We're turning into beings of thought, where expressing ourselves will be as easy as speech.

Fair enough, I envy your optimism!

>I can't wait for this to pan out. We're turning into beings of thought, where expressing ourselves will be as easy as speech.

But who will listen to us, when AIs can speak better than we can?

Expressing yourself doesn’t require an audience does it?

Were you around when the internet was going to bring people together in a more decentralized, democratic, and free society?

You mean that time when how people appeared wasn't supposed to matter?

Seems like this brings that closer.

> I can't wait for this to pan out.

The virtual world will look better than the real world.

People will look better in the virtual world, and become afraid to show their real image.

People will stay inside, behind their computer screens.

Movies will all become like cartoons (because, in our daily environment people look like cartoons already).

etc. etc.

Says the guy posting on a forum limited solely to text by choice.

It's bizzare to write "this is the end" when almost none of us would be on HN if it so much as required a picture avatar.

People have been saying that all my life (magazines and photoshop give unrealistic beauty standards, etc). The world hasn't ended yet.

Have you got daughters?

Creepy and dehumanising I agree, but only offensive to my eyes.

Reply on news.ycombinator.com