The Cost of Motion

How Move AI turned iphone footage into character animation.

Author

Nicolás Cerdeira
October 29, 2026

Hey, it’s Nico.

Quick note before we get into this week’s issue: I’m now in San Francisco, and I’ll be around until November 20.

If you’re in town, I’d love to connect! Always happy to grab a coffee, meet founders and people working on interesting things in consumer or AI, and learn about what you’re building. Just hit reply and let’s make it happen.

Here are the 3 most important things of today’s issue:

  • Move AI, a startup that turned videos into 3D meshes, shut down — find out why below

  • How Cursor, Ramp and Cognition hire their teams

  • Genie 4 is out! — learn what this means for the future of World Models below

What I saw at AcceleratorCON AD

Last Friday I flew to New York for AcceleratorCON, a one-day conference where startups, accelerators, and investors meet face-to-face.

A few things that stuck with me:

  • The Draft Experience is the most original part. It's basically the NFL Combine for startups: selected founders get 1:1 pitch meetings with investors and accelerators, all in the same building, on the same day.

  • The theme was "AI is the new fire," and almost every conversation came back to it.

  • The tables between talks were as busy as the stages. A lot of the good conversations happened there.

  • There was an archery station 🏹

If you want to meet the people who fund and accelerate startups, in one room, it's worth going. Next stop is San Francisco. Check out AcceleratorCON here.

This Week In Startups

🔗 Resources

How Cursor, Ramp and Cognition hire their teams

📰 News

Mistral releases Large 4’s model weights for self-hosting

Anthropic officially files for its public debut

💸 Fundraising

Bedrock Robotics, a startup that retrofits construction equipment to operate autonomously, raised $600M in Series C funding

Upwind, a startup that secures cloud infrastructure and AI workloads, raised $400M in Series C funding

Serval, a startup that uses AI agents to build and run enterprise workflows, raised $150M in Series C funding

Rendezvous Robotics, a startup that builds modular spacecraft designed to assemble themselves in orbit, raised $20M in seed funding

Fail(St)ory

The Suit is Optional

Move AI turned videos of people into 3D animation data, letting performers leave the motion-capture suit behind.

Its technology made it into Just Dance and a Sony Music concert inside Fortnite.

It raised a $10 million seed round. This month, its assets sold for just £525,000.

What Was Move AI:

Imagine you're animating a character for a game. You need it to move like a real person, down to how their weight shifts during a turn. Motion capture records a performer's movements so an animator can apply them to that character.

Move used AI to identify body positions in video and reconstruct movement in 3D. Animators could import the resulting data into tools such as Blender, Maya or Unreal Engine and apply it to an existing character.

Its Move One app let individual creators capture a performer with only an iPhone and then get the animated 3D meshes.

For game developers and production studios, Move offered more advanced systems that filmed performers from several angles. Its newest product, Genesis, let studios process the footage on their own computers and keep production data in-house.

Performers could wear normal clothes, without a suit covered in tracking markers. That gave studios more freedom to record people dancing, acting or playing an instrument, then use those movements in games and virtual performances.

For studios, using Move still meant setting up cameras, cleaning up animation and fitting the results into an existing production process. Move's team also provided installation help and technical support, adding hands-on work to the business of selling software.

Genesis was announced for April 2026. By then, Move's latest accounts showed negative net assets. Move needed to grow its studio business while the company was already under financial pressure.

The Numbers:

Reasons for Failure: 

  • Deployment still needed people: Move's commercial lead described in-person assistance and installation support. That suggests each professional sale could bring technical work alongside license revenue. Customer savings depended on everything from setting up cameras to cleaning the final animation, making adoption harder than downloading an app.

  • Genesis arrived while Move was already under financial pressure: Move announced its new studio product for April 2026, after its accounts showed a weakened financial position. It still had to pay engineers and win customers while the new product established itself. That suggests Move had little room to wait for sales to grow.

  • Studios were delaying purchases: The company that bought Move’s assets reported lower spending by film, TV and games studios, along with delayed contracts. Move sold to similar customers. Slower buying decisions could have meant waiting longer for revenue while salaries and development costs kept coming.

Why It Matters: 

  • Measure savings through the finished animation. Include camera setup, calibration, cleanup and integration when calculating what the AI actually removes from a customer's budget.

  • Track technical-support hours alongside license revenue. A growing list of studio customers can still require more engineering help than the business can afford to provide.

Trend

Artifical Experience

Google announced Genie 4 this week, letting users explore generated worlds, change a scene while it's running, and hear characters respond.

Google also wants these worlds to become places where other AI systems learn.

An agent could practice a task, see the consequences of its actions, and try again. That would give AI developers a way to generate some of the experience their models need.

Why it Matters

  • Robot developers could repeat difficult scenarios without staging every attempt on physical hardware.

  • Providers could sell simulation access, adaptation and testing to teams building physical AI.

  • Buyers need evidence that simulated results predict performance on their own tasks.

Genie 4

Genie 4 is Google's latest model for generating worlds that respond as you explore them. Genie 3 already showed the basic idea: you move through an AI-generated environment, and the model keeps predicting what you should see next.

The new version gives those worlds sound and dialogue. Characters can converse, and the model generates ambient audio alongside the visuals.

You can also direct the scene while it's running: introduce an object or change the weather, then keep exploring the world you've just altered.

Google brings Gemini into the process to help these worlds follow a sequence of events. Gemini watches the scene, breaks goals into steps and tracks what's happening as Genie generates the next frames. In a robotics experiment, Gemini passes the planned steps to Genie as text instructions while the simulation renders.

That ability to set up a situation and steer what happens gives Google a starting point for building training environments. A developer could assign an agent a task, let it attempt the task, then vary the conditions and try again. With a system to score the results, those attempts could become experience the agent learns from.

Other Signals

Three developments this month show different parts of that workflow taking shape:

  • NVIDIA's Cosmos added simulation models for driving and two-armed robots. They accept control inputs and generate subsequent observations, giving developers a way to connect a controller to a learned environment for training or testing.

  • Odyssey's PROWL-2 addresses what happens when the environment gets something wrong. Its training method checks simulated experience against reference executions, withholds unreliable predictions and uses those failures to repair the world model. It was tested in games and simulated robot coordination.

  • Berkeley's SuperMimic trains a humanoid controller through a learned model of body dynamics, then demonstrates it on a physical Unitree G1. Its world model learns from a conventional simulator, showing how learned dynamics can become another stage in robot training.

The Trend

AI companies are building environments where other AI systems can practice. Genie 4 matters because Google wants to make those environments easy to generate, interact with and change on demand.

I think we’ll see more startups selling training environments for specific jobs. A team building a warehouse robot could buy access to a simulated warehouse tailored to the objects, layouts and failures its machine needs to handle, then pay for new scenarios as the robot takes on more work.

Help Me Improve Failory

How useful did you find today’s newsletter?

Your feedback helps me make future issues more relevant and valuable.

Login or Subscribe to participate in polls.

That's all for today’s edition.

Cheers,

Nico