Loading video...

Video Failed to Load

Go Home

vibe coding metahuman with threejs and codex: first demo I mean it's still a long way to go, but i'm quite surprised myself, turned out way better than i expected and the character isn't too far from AAA? all textures generated by gpt image 2, the facial expressions and...

11,043 views • 4 months ago •via X (Twitter)

69 Comments

robot 2.0's profile picture
robot 2.04 months ago

this was my day one rant, enjoy the crashout:

Astrid Wilde 🌞's profile picture
Astrid Wilde 🌞4 months ago

the most complex part of this work is getting FACS properly setup (source: i made a similar tool for EA in 2023) i bet you could get that up and running in under a week now just vibecoding though. brave new world

robot 2.0's profile picture
robot 2.04 months ago

oh here we gooo thank you for the next rabbit hole!

Astrid Wilde 🌞's profile picture
Astrid Wilde 🌞4 months ago

the way every AAA studio does it is with a map of 2d sliders like this: enjoy the rabbit chase!

robot 2.0's profile picture
robot 2.04 months ago

do i understand this correctly that i have to build a library of base expressions and use these as morph targets with sliders? sigh, isn't there like a fully programmatic way

Astrid Wilde 🌞's profile picture
Astrid Wilde 🌞4 months ago

you are understanding this correctly there is a fully programmatic way but it's a 70D math problem and an optimizer is going to get you a poor result

robot 2.0's profile picture
robot 2.04 months ago

goddamn, alright pulling up the sleeves and getting to work

King Stanky MBA's profile picture
King Stanky MBA4 months ago

You could singlehandedly replace the entire staff at bethesda

robot 2.0's profile picture
robot 2.04 months ago

lmao i'm not sure about that, i really enjoyed bethesda games

robot 2.0's profile picture
robot 2.04 months ago

mr mogman i can't be mogging when i'm token poor wen free pro sub??

LEYVERSE's profile picture
LEYVERSE4 months ago

For lipsync and expressions u can check this:

robot 2.0's profile picture
robot 2.04 months ago

here we goooo

al's profile picture
al4 months ago

he's more charismatic than 99% of X 🤪

robot 2.0's profile picture
robot 2.04 months ago

i literally just prompted in chatgpt "given this UV map, pls generate a generic human skin"

JERIC陳傑瑞's profile picture
JERIC陳傑瑞4 months ago

Wow this is really awesome and I can see many use cases - please keep cooking and pushing this

robot 2.0's profile picture
robot 2.04 months ago

will do :)

Reigen's profile picture
Reigen4 months ago

are you aware of the open source anny project and makehuman blender extension?

robot 2.0's profile picture
robot 2.04 months ago

i saw it, but the fidelity of the humans was meh?

Rawlala's profile picture
Rawlala4 months ago

already on par with the next james bond lol

robot 2.0's profile picture
robot 2.04 months ago

don't mock me 😭😭😭

Rawlala's profile picture
Rawlala4 months ago

lmaooo don't be rude they have a whole ass team on this 😭🤣

robot 2.0's profile picture
robot 2.04 months ago

of course the actual meta part is completely missing for now! But i wanted to share some progress, been deep into 3D the past couple days 😭

loground's profile picture
loground4 months ago

damn, that's a fire result

᲼ٴ`᲼'s profile picture
᲼ٴ`᲼4 months ago

awww

robot 2.0's profile picture
robot 2.04 months ago

i've seen worse tbh, can't say it's slop!

Autark's profile picture
Autark4 months ago

Uh, Blender has an MCP plugin dude

robot 2.0's profile picture
robot 2.04 months ago

man that's absolutely not the same

Autark's profile picture
Autark4 months ago

You want AI assist in Blender, that's your path of least resistance

robot 2.0's profile picture
robot 2.04 months ago

the previous community MCP was absolute dogshit and the anthropic one was just the start with very light tools

Autark's profile picture
Autark4 months ago

If only there were a way to build one, maybe one that runs arbitrary Blender Python code...

robot 2.0's profile picture
robot 2.04 months ago

again, you are looking at raw features retopology work can't be done by just writing a python script

Ryan Fitzpatrick's profile picture
Ryan Fitzpatrick4 months ago

could make a bioware style story driven game with tools like these. Unlocking new possibilities every week!

robot 2.0's profile picture
robot 2.04 months ago

can't stop won't stop

ChineseTechBro's profile picture
ChineseTechBro4 months ago

i used unreal metahuman for a long time, the potential i see here is crazy considering it's a full web stack

robot 2.0's profile picture
robot 2.04 months ago

robot delivers 🤝

Atelier Elysara's profile picture
Atelier Elysara4 months ago

it's still a million miles away from any studio work

robot 2.0's profile picture
robot 2.04 months ago

true, but already miles ahead the average indie graphics, so it's a step forward

Atelier Elysara's profile picture
Atelier Elysara4 months ago

dunno what indies you're looking at but not any of the ones i see look like the characters got left out in the sun for a few weeks with no sunscreen

robot 2.0's profile picture
robot 2.04 months ago

that's the hdr preset from drei

Atelier Elysara's profile picture
Atelier Elysara4 months ago

it's not just the hdr, the textures bug-eyes and mouth hanging open are certainly not helping it look good

robot 2.0's profile picture
robot 2.04 months ago

alright lfg i guess

mongrlz's profile picture
mongrlz4 months ago

What are using for this?

robot 2.0's profile picture
robot 2.04 months ago

what exactly do you want to know?

mongrlz's profile picture
mongrlz4 months ago

Oh I meant what are you using to build this, copilot like the old streams?

robot 2.0's profile picture
robot 2.04 months ago

noo copilot rugpulled, codex app

mongrlz's profile picture
mongrlz4 months ago

Upgrades, love it. Hope you remember me when ur up

robot 2.0's profile picture
robot 2.04 months ago

i have no idea who you are but remembering i was streaming on copilot makes me think you've been around for a while lol

mongrlz's profile picture
mongrlz4 months ago

lol i have been, just under a new name, dont say it out loud tho

second_order's profile picture
second_order4 months ago

the uncanny valley

robot 2.0's profile picture
robot 2.04 months ago

i'm nots sure its the uncanny valley or just the fact that his eyeballs are about to pop into your hands while watching

AINZX's profile picture
AINZX4 months ago

can u guide i cant belive it to can u make a video regarding it

robot 2.0's profile picture
robot 2.04 months ago

@SurajChawh79862 what? there is. a demo link

AINZX's profile picture
AINZX4 months ago

but codex cannot lipsync add visemse for a mesh build in tripo 3d

CoffeeShop's profile picture
CoffeeShop4 months ago

You call this garbage future????? LMAOOO the bar is really low nowadays as it seems; you will never get near the world class while you still keep trying to do everything in the most lazy way possible

robot 2.0's profile picture
robot 2.04 months ago

hey coffeee where have you been man? you only show up when i get a lot of views :(

CoffeeShop's profile picture
CoffeeShop4 months ago

Sorry, I was kind of busy lately, so I haven't logged in

robot 2.0's profile picture
robot 2.04 months ago

you can't just roast me on my highs, you gotta show up on my lows too man

Abe's profile picture
Abe4 months ago

Blender is already AI boosted 🤔🤷‍♂️

kvick's profile picture
kvick4 months ago

we're at garry's mod level 😂

robot 2.0's profile picture
robot 2.04 months ago

THAT bad 😭😭😭

kvick's profile picture
kvick4 months ago

hey man, garry's mod WAS AAA in 2006, only 20 years ago! 20 years in AI time is maybe just 1-2 years?

robot 2.0's profile picture
robot 2.04 months ago

u bitch

paddy's profile picture
paddy4 months ago

yooo nice, already looks solid af

robot 2.0's profile picture
robot 2.04 months ago

thanks paddy!

Stavros Nonis's profile picture
Stavros Nonis4 months ago

there is a cool intersection between VRMs, 3d gaussian blendshapes, and metahuman ready to be exploited ;)

robot 2.0's profile picture
robot 2.04 months ago

VRM facial muscles seemed very limited when i looked into it

Stavros Nonis's profile picture
Stavros Nonis4 months ago

they are just blendshapes My theory is that you can take a base model and create multiple blendshapes like from multiple 3d guassian splats and derive a few blendshape presets, you could have metahuman like functionality I will try it whenever am done with my VRM editor

robot 2.0's profile picture
robot 2.04 months ago

i'm using actual bones here though

Stavros Nonis's profile picture
Stavros Nonis4 months ago

yeah, VRMs use bones too to drive skinned meshes I havent seen your code but am sure it's just using a Skeleton and a skinned mesh

Related Videos

vibe coding metahuman with threejs and codex: spitting FACS ok, i'll admit 3D modeling has a lot of depth, going from a cube to a full facial action coding system is.. a lot of work, but really it isn't all that hard to learn like some people want you to believe it took me about 10 evenings to go from scratch (no idea how to even model in blender beyond the famous donut) to a working rigged and shape keyed expression system the actual vibe coding parts are very manageable, i'm not even hitting any session limits the past week since codex is basically just reading the glb file and it writes the drivers to have a full animation system in a single turn after that i usually notice that some things don't look as professional as i like it to be, followed by codex sending my ass back to blender and telling me what i need to do to make it better so here we are, the expression system is still very very simplified, no micro expressions yet and i believe i will need probably 500+ shape keys to actual achieve realism, but why not? Doing the 52 ARKit expressions for this demo wasn't too hard either, just tedious and boring The final driver in the demo uses a mix of bone movements and blendshapes to smoothly animate the face I might mix up some terminology here but i usually do cause it's been like what, 10 days since i started? Guess next up i'll look into face deformation parameters and then we'll move to the fun part of skinning. Once the head/neck area works (hairs are also still missing), full body next and finally LOD systems and we can call it a day Demo (pls use a desktop i'm not optimizing the ui for mobile):

robot

29,357 views • 4 months ago

I got this custom scorpion rigged, animated and running in my game in one full day with AI. GPT-6 built the entire rig. My part was preparing the model and motion references, then giving feedback. After trying a few approaches, I found a workflow that worked: 1. Give it a model that makes sense. I generated the model with Tripo P2.0, then cleaned it up myself. Low-poly, optimized, split into logical parts. I think that preparation mattered a lot. GPT-6 works on the mesh through code. A manageable vertex count and sensible parts give it a clearer starting point for bone placement and skinning. 2. Show it the movement you want. I took a reference image of the character and used Seedance to generate side-view animation videos. I decided which animations the game needed and selected the takes with the movements I liked. Then I gave GPT-6 the model, the videos and that animation list. There was no existing rig to start from. 3. Compare the animation frame by frame. This was the idea I wanted to test: ask GPT-6 to compare its Blender animation against the reference video frame by frame, from the same view. That gave it concrete poses and timing to work toward. I could point to where the feet should be planted or when the tail should strike. 4. Expect to give feedback. A generated run can look like hopping. An attack can start with a jump you never wanted. Choosing the references matters, and a side view still leaves depth and hidden legs to figure out. It took several iterations. I explained what looked wrong and how I wanted it to move, and GPT-6 kept refining the animation. Once the walk worked, I also had it create left and right strafe animations procedurally from that walk. Those didn't need their own reference videos. By the end of the day, the creature was in Unity. I've recorded the full workflow, including the Blender setup and feedback process:

Stefan 3D AI

12,563 views • 6 days ago

I just compared Claude Code vs Codex vs Cursor CLI The task was to build a Next.js app with Tailwind 4 and shadcn components to collect customer feedback and showcase it with a widget. I gave all three the same prompt and let them go for 30 minutes to see what they came up with. Claude Code with Opus 4.1 Even though I told it to set up the app in the existing project folder, it tried to create a directory for it. After I interrupted and told it not to do that, it built a demo form and landing page with no errors. I had to ask it to make the demo interactive so users could submit a testimonial and preview it. The landing page looked like AI and was pretty basic, but it worked and it was done in a fraction of the time of the others. Total tokens used: 33k Codex with GPT-5 At the end of the 30 minutes I just could not get Codex to produce a working app. It got stuck in a loop of not being able to set up Tailwind 4 and despite many, MANY, attempts, I ended up with a "failed to compile" error. Total tokens used: 102k Cursor Agent with GPT-5 This was the slowest agent by far and a couple of times I actually thought it got stuck in a loop and was close to Ctrl+C'ing to cancel it. The TUI is really nice though, especially how it shows diffs and it did eventually build a working app (after one or two slight errors that needed fixing) The demo was interactive and it had a very minimal design that looked bare but also a lot less like an "AI generated" app than the Opus 4.1 design. It also wasn't too chatty and just did what it needed to do! Code quality was on a par with Opus 4.1, but it did use 5.5x as many tokens to get there. Still cheaper than Opus on a direct comparison but not when you factor in a Claude Code Max subscription. Total tokens: 188k I'll be able to do a proper comparison and record some videos when I'm back from holiday but for now, Opus is still the more capable model out of the box and Claude Code is the more complete CLI product. It will be interesting to see how Cursor evolve their CLI though with commands and subagents because I think with GPT-5 they have a real shot at providing competition for Claude Code if they can optimise output to get similar quality with less tokens. Jump to 0:40 in the video to see the two apps. Which do you think is which? ;)

Ian Nuttall

195,173 views • 1 year ago