Startup Opportunity: 2D>3D modeling, auto-rig, live-streaming

Jobs Volunteer Work
, , , ,
1

Technical Artist / Character Rigging — AI Avatar Startup

Help us solve a difficult 3D character problem

VIDA is an early-stage startup building a platform that turns a user’s 2D image or PFP into a 3D avatar that can be controlled by live motion tracking and used for streaming.

Our goal is:

2D Image → 3D Character → Automatic Rigging → Motion Tracking → Live Avatar

We already have a senior software engineer leading the application, backend, infrastructure, testing, and system integration and two junior devs.

Our biggest remaining technical challenge is the 3D character pipeline, particularly getting AI-generated 3D meshes automatically prepared for animation.

We are looking for a Blender/3D technical artist, character rigger, technical animator, computer graphics student, or technically minded Blender user who would like to work with us on this problem.

The Problem

Our 2D-to-3D pipeline can already produce usable character models, but the next step is difficult:

How do we automatically take an AI-generated 3D mesh and turn it into an animation-ready character?

We’re investigating:

  • Automatic skeleton generation

  • Automatic bone placement

  • Humanoid detection

  • Skin weights

  • Mesh cleanup and normalization

  • Facial rigs

  • Blendshapes

  • Motion capture

  • Motion retargeting

  • GLB/glTF

  • VRM

  • Blender Python automation

  • Computer vision

  • AI-assisted rigging

We’re particularly interested in people familiar with technologies/research such as:

  • RigNet

  • UniRig

  • HumanRig

  • RigAnything

  • AccuRIG

  • Mixamo

  • DeepMotion

  • Rokoko

  • MediaPipe

  • Blender’s Python API

You do not need experience with all of these.

What We’re Trying to Build

Ultimately, we want a user to be able to:

Upload a PFP

↓

VIDA generates a 3D character

↓

Character is automatically rigged

↓

User turns on their camera

↓

Their movements control the avatar

↓

They can stream with the avatar

The long-term goal is to make this process take minutes rather than hours of manual 3D work.

Who We’re Looking For

This opportunity could be a good fit for:

  • Blender users

  • Character riggers

  • Technical artists

  • Technical animators

  • Character TDs

  • Computer graphics students

  • Computer vision students

  • Game development students

  • Animation students

  • ML/AI developers interested in 3D

  • Self-taught Blender/Python developers

  • Recent graduates

You do not need to be a senior engineer.

We are especially interested in someone who enjoys solving technical problems and experimenting with Blender, Python, AI and 3D rather than simply creating finished artwork.

A talented college student or recent graduate who wants a serious project for their portfolio could be an excellent fit.

What You Would Work On

Initially, we’d like to work with you on a small technical investigation.

We can provide one of our current 3D character outputs.

The question is:

Can we automatically transform this mesh into a properly rigged character that can receive motion data?

You would help investigate:

  1. What is wrong with the current mesh?

  2. What preprocessing is required?

  3. Can an existing auto-rigging system solve the skeleton?

  4. Can we automatically generate usable skin weights?

  5. What should the XXX skeleton look like?

  6. How should the character be exported?

  7. How should motion be retargeted to it?

  8. What parts should we use from existing tools?

  9. What parts should XXX build ourselves?

We’re not expecting you to solve the entire system alone.

We’re looking for someone who can experiment, prototype, research existing solutions, and work with our senior engineer to establish a practical pipeline.

Existing Team

Our lead developer is a senior software engineer and manager with extensive experience in:

  • Python

  • Java

  • JavaScript/TypeScript

  • Web applications

  • Mobile applications

  • Automated testing

  • Broadcast automation

  • Audio streaming

  • Networking

  • Servers

  • Hardware/system integration

He is strong on the software and systems side.

We are looking for someone who can complement him and junior devs with 3D character, rigging, animation and Blender expertise.

Compensation

We are an early-stage startup.

We are looking for someone interested in contributing to an ambitious technical project and potentially becoming part of the team.

For the initial volunteer/experimental phase, the objective is to determine whether there is a strong technical and working fit.

For meaningful ongoing development, we are prepared to discuss:

Paid hourly work + potential equity

The exact arrangement will depend on experience, availability and level of contribution.

We want to be transparent about this from the beginning: we are not looking for someone to provide months of unpaid production work.

Our intention is to identify talented people who want to build something with us and compensate ongoing contributions appropriately.

Why us?

The problem we’re working on sits at the intersection of:

AI + 3D Generation + Character Rigging + Computer Vision + Motion Capture + Real-Time Streaming

We’re not trying to build another Blender plugin or another 3D modeling service.

We’re trying to make the process of creating a usable live avatar dramatically simpler.

If we can solve the automated rigging and animation pipeline, the technology could potentially be used across:

  • PFPs

  • NFT characters

  • VTubers

  • Streamers

  • AI personalities

  • Gaming avatars

  • Social media creators

  • Virtual influencers

What We’d Like to See

If you’re interested, please reply with:

  • Your Blender experience

  • Any character-rigging experience

  • Examples of your work

  • GitHub, ArtStation, demo reel, or portfolio if available

  • Any experience with Blender Python

  • Any experience with AI-generated 3D models

  • Any experience with motion capture or retargeting

  • Your current student/graduate status, if applicable

  • A brief description of the most technically difficult 3D problem you’ve solved

You don’t need an impressive resume.

Show us what you’ve built.

If you’re a student or recent graduate and this sounds like the kind of problem you’d enjoy working on, we’d particularly like to hear from you.

Please put "VIDA" in your reply so we know you’ve read the post.

2

You don’t, because AI doesn’t make animation-ready meshes. Instead you make your own mesh, which will actually save you hundreds of hours over time, because AI cleanup is a massive task

3

Joseph,

When you say “make your own mesh,” I assume you mean creating a clean, animation-ready base topology rather than using the raw geometry produced by an AI 3D generator, like a humanoid one, however we’re wanting to be versatile in changing any NFT/PFP into an avatar.

Our objective with is to automate as much of the process as possible. We’re currently getting reasonably usable 3D characters from 2D PFPs, but the major problem is getting those meshes into a reliably riggable/animatable state.

Do you think there is a practical middle ground between manually building every character and using the raw AI mesh? For example, could we use a standardized humanoid base mesh/topology and automatically transfer the generated character’s shape, textures, clothing, etc. onto it?

Or is there an automated AI/Blender workflow you would recommend for taking an imperfect AI-generated mesh and converting it into an animation-ready mesh?

I really appreciate your response, it reinforces thoughts we’ve had about the process.

4

It’s not an easy question. If you could figure this out perfectly, you would be the first. I have some thoughts, but they’re just ideas, I haven’t tried them.

Now, with that said, I don’t think you need to reinvent the wheel here. There are already a lot of solutions out there for generating a animation ready mesh based on custom parameters, that predate AI generation in some cases by decades. I am thinking specifically of things like character creator 4,
DAZ Genesis, unreal engine metahuman, etc. many of these are also already rigged, including facial blend shapes compatible with AR kit, which would save you additional effort.

It would be easier to customize something that already has correct animation topology and is already rigged, rather than starting with something bad and trying to manhandle it into something good.

That’s the extent of my thoughts, I could look into it from a consultant perspective if you want, happy to chat about that

5

Joseph,

You’re the man for lending me your knowledge and insight. Thank you. I’d absolutely be interested in speaking more about our consulting position. I’m the founder and do not have the technical knowledge so I’d want to arrange a meeting with you and Jordan. I there a way to connect outside of Blendr that you would like to share?

Alex Oberman

8

Joseph,

Hope all is great. I tried to contact you on your website and I don’t believe its gone thru. What’s a good contact? AOberman@TheTritium.com if you’d like to email me there.

AO

9

Joseph is right about AI meshes not being ready for animation. I ran the numbers on this instead of just guessing.

It was a structural check on over 2,000 AI-generated 3D assets (from 23 different generators) just to see how they hold up. The two biggest red flags I found for rigging were:

Non-manifold edges: About 10% of the AI models had these, the worst hit 53.5%. For context, I ran the exact same test on about 150 human-made models and only found 1.3%. The gap is huge, and it’s definitely a recurring problem.

Missing UVs: Several of these generators just don’t output UVs at all. Not even a messy layout, they are just blank.

Those non-manifold edges are what I’d worry about most for your specific setup. Auto-rigging tools and skin weight solvers rely on clean geometry to figure out where bones go. If an edge is shared by three faces, the software gets confused. The tricky part is that it doesn’t just crash.

I’m pretty sure you were recommended MetaHuman or CC4, but I’d push back on that a bit. Those programs work great because they rely on perfect, pre-made base meshes. It sounds like your goal is to use unique, AI-generated geometry. If that’s the case, you really have to follow a strict order: measure → repair → rig.

I actually built a tool that does this measuring and repairing offline, and you wouldn’t even have to install anything on your end.

If it helps, feel free to send me a handful of your generated characters. I’ll run them through the tool, let you know what it found, and send back a repaired version if it’s fixable. This is completely free with no strings attached. I really want to see the weird ways these meshes fail so I can improve my tool for the future. To be clear about scope: it’s one batch as a favour, not me joining the investigation phase for now, you said you’re not after months of unpaid work and I want to match that.

Just a quick heads-up on two things the tool can’t do:

  1. It doesn’t know if a mesh is actually made well artistically (things like edge flow, silhouette, and good deformation are invisible to it). It only cares if the structure is mathematically valid.

  2. It doesn’t do the actual rigging. It is just the cleanup step to get your model ready for the rigger.

10

Alza,

Thanks for the response, I really appreciate it. I’ve forwarded over your reply to my lead dev and will get back to you here. I think we definitely take you up on your offer, again, thank you. Do you ahve an email or contact method. John.Oberman@gmail.com

Back to you soon, this sounds promising.

JO

11

Hello,

You can contact me at alzastrategy@gmail.com . Thanks for considering my offer and I look forward to hearing from you.

12

Sending you an email now. Right on Alza.