Early R&D using our SP-6M dataset. Exploring image-to-3D reconstruction from single images, including heavily modified inputs (lighting, hair, etc). Still a work in progress.

Oh my god. That… is… insanity. Currently all the ai image to 3d human services fail horribly 😂 but this could be the answer!

Yep I know :) data and diversity are king when it comes to this kind of thing. We have over 7600 individual scans with 11 expressions each so around 80k scans across a hugely diverse group of people.

Very cool, can't wait, this is actually an amazing use of all your scanned data

Cheers Vadim!

holy cow. Is this real?

Yep! It's all we've been doing for the past 18 months. :)

Fucking fantastic dude. I dont even want to know what your compute bill looks like 😅

Ha ha wait till I show you what we're using to train. We will show that soon.

It is too good to be true.

Ha ha 🤣

This is fantastic! Are you training your own model? What about topology?

Yep we're training our own model from scratch with all of our own data. Topology is production ready consistant morphable geometry with 8k textures. This is the dataset we created for it.

Instant sub. Way to go!!👍

where can I buy the alpha?

It's way to early for this I'm afraid.

This is cool are you using flame smplx models with your textures overlayed? That’s pretty sweet!

Thank you! no this isn't flame this is our own topology / model.

well this is

Woah 😭

Is the image-to-3d construction available to try yet?🤔

No I'm afraid not. We have a lot of work still to do :)

any timeline on date of release?

No I'm afraid not. It's very early in development.

Amazing!

Thank you!

How ! 🔥🔥🔥😭

A lot of work :)

My first question - when is the tool coming out and what would it be priced at? (genuinely want to know on behalf of 3d artists around the globe)

Im afraid thats not something were announcing at the moment.

Can you train more AI based on images of people/AI who are smiling, showing their teeth shape and dimples? Thank you. @Ten_24

Do these work with MetaHuman?

It will do yes but thats not its intended use :) more on that later in development.

RIP Metahuman

ha ha not quite :) but thanks for the confidence boost :)

Does it produce a mesh?

yes a fully retopologised production ready mesh

Yes, more pls. 🔥🔥🔥

Wow!

Cheers!

So impressive! Does this build it with the standard topology you've established for all your scans?

Thank you and yes indeed it does :)

That's great, really looking forward to trying it out!

🔥🔥🔥🔥🔥🔥🔥

Thank you! 🤩

What????!!!

amazing, great job bro

Thank you!

WIll there be a ten 24 to Blender pipeline ?

wtf this is crazy

Yeah it's been an insane project. Just scanning this many people has been a full time job for 20 people for the last 18 months.

I know you said elsewhere this is still early stages, but are you able to feed it any image and have it create a model? Or only specifically curated ones? I’m dying to see what it would do with a photo of me hahaha

Yes the goal is any image :)

Making good progress! Still some issues with noses, and some random tattoos (LOL), but good first steps

Thank you! Yes we're learning as we go. It would be a lot easier if we built on top of another model but we're committed to using all our own data. Totally clean, not scraped, and GDPR compliant. There's a lot still to do and we have so much more to add to this. We expect to see a 50 - 100 fold increase in quality over the next 12 months :)

@100PercentRobot Where can I invest in the company and the tech? This is definitely going to explode for sure.

@100PercentRobot Cheers Roberick! I'm afraid there are no investment opportunities at the moment.