# Generative Video at the Speed of Light

Keegan McCallum

AI Engineer World's Fair 2026

[Talk page](https://ai.engineer/talks/Xln-On3syJk-generative-video-at-speed-light) · [Video](https://www.youtube.com/watch?v=Xln-On3syJk)

## Generative Video at the Speed of Light

Fast, continuous video generation changes what an application can offer—but making it interactive requires GPU routing, streaming pipelines and controls that stay synchronized with the output.

Basic familiarity with React, Python async functions and model inference will help with the implementation discussion.

### From believable pictures to continuous generation

[0:20](https://www.youtube.com/watch?v=Xln-On3syJk&t=20s)

The 2023 clip of Will Smith eating spaghetti is an immediately recognizable starting point for generative video: distorted motion and imagery that nobody would confuse with reality. **Sora** in 2024 brought a substantial improvement, followed by **Sora 2** and then **Seedance**, whose output Keegan McCallum describes as photorealistic. That progression makes visual quality the obvious axis on which to judge new models.

But an interactive application needs more than a convincing finished clip. It needs video that arrives quickly enough to consume as it is generated, and that can continue rather than stop after a short shot. McCallum, who introduces himself as the founder of [uRun](https://urun.sh/), an inference provider for interactive media, shifts attention to **efficiency and long-horizon generation**: capabilities that determine whether video can become an interface.

### Helios makes speed visible

[1:16](https://www.youtube.com/watch?v=Xln-On3syJk&t=76s)

The **[Helios](https://arxiv.org/html/2603.04379v1)** demonstration separates two useful capabilities. In the bottom-right corner, one generation runs continuously; elsewhere, a collection of clips is generated faster than a viewer can consume them. McCallum puts their quality near that of the previous year's frontier models. He describes Helios as a distillation of **[Wan 2.1 14B](https://github.com/Wan-Video/Wan2.1)**. More precisely, its technical report describes architectural adaptation and token compression before distillation; the distilled checkpoint is distinct from Helios Base.

The next comparison asks the audience to distinguish a real-time result from one that took minutes to generate. Its complete reveal labels two portraits of women surrounded by flowers, using the same prompt and the same H100:

| Model | Generation throughput |
| --- | ---: |
| Wan 2.1-14B | 0.33 fps |
| Helios 14B | 19.5 fps |

McCallum judges the right-hand result to have arguably better motion and reports about one-hundredth of the cost. The displayed frame rates measure throughput; the talk does not give the billing calculation behind that cost comparison.

![Slide showing two generated portraits of women surrounded by flowers, labeled A and B, with model names and frame rates revealed below.](https://ai.engineer/api/talk-essay-media/72b679b881f083459dbc06cbdf319f731369f76234aa29dc3b34eb4bb5437fbe)

Same prompt: Wan 2.1-14B at 0.33 fps and Helios 14B at 19.5 fps on the same H100.

The subsequent charts address quality for both short and long video generation. Helios had arrived in March, making it a recent example of how quickly the efficiency frontier was moving. The important combination is sustained output and useful visual quality: speed opens up interaction only if the generated scene remains worth watching.

### From individual clips to hours of interaction

[2:42](https://www.youtube.com/watch?v=Xln-On3syJk&t=162s)

These capabilities extend beyond one model family:

- **World models** maintain consistency over long horizons while exposing fine-grained camera and viewport controls.
- **Avatar models**, such as those from [LemonSlice](https://lemonslice.com/blog), provide a conversational visual presence.
- **Video-to-video models** transform incoming imagery in real time, creating the effect of a magic mirror.

Each offers a different kind of interaction: navigating a scene, engaging with a character, or changing the appearance of a live input.

McCallum reports at least forty model releases that year across the emerging real-time and long-horizon video ecosystem. To put the economics in familiar terms, he asks who has spent $10 or $50 in an hour using Claude Code. He estimates that, with most of these video models, $10 buys three hours of continuous generation and $50 buys fifteen hours. Those are his usage estimates, without model-specific rates or billing conditions; the spoken fifteen-hour figure makes the slide's “full day” comparison concrete. The resulting opportunity is sustained visual interaction rather than purchasing isolated clips.

![Slide asking who has burned $10 or $50 of tokens in an hour, followed by the revealed comparison to three hours or a full day of live-generated video.](https://ai.engineer/api/talk-essay-media/6cb6b442da7a5503fac93a3cda8df505524a2a177bf7c80c6350a7b6a8fa31d6)

$10 ≈ three hours; $50 ≈ a full day of live-generated video.

### A webcam becomes a visual interface

[4:07](https://www.youtube.com/watch?v=Xln-On3syJk&t=247s)

A magic mirror starts with a webcam feed and changes what the user sees. McCallum's examples include trying an outfit, seeing yourself in a car you like, or previewing a haircut. Because the transformation is open-ended and happens on live input, the interface could support exploration while the user moves and reacts, rather than requiring a separate image-generation request for each variation.

That visual medium also suggests an accessibility opportunity. Much of today's AI interaction demands reading and writing, which can be difficult for some users and a poor fit for people who think or learn visually. Companions and other visual interfaces could let more people experience the benefits already familiar to coding-model users. This is a proposed direction for interface design, not an accessibility outcome demonstrated in the talk.

### Direct the shot while it is being generated

[5:02](https://www.youtube.com/watch?v=Xln-On3syJk&t=302s)

Content creation has often followed a slot-machine workflow: write a prompt, perhaps supply keyframes, generate a result, and try again if the shot misses. McCallum describes spending about $10 per minute trying to obtain the desired shot, without specifying a provider or billing basis. The creative problem is the separation between asking and seeing: the user discovers what to change only after committing to a generation.

McCallum says these models can respond to steering in under a second while generation continues. That changes the workflow into an ongoing feedback loop. A creator could pilot an agent, watch over its shoulder as it generates, and adjust the direction while the scene develops. The control becomes more granular because observation and intervention happen during the same generation.

He then points to **[Google Gemini Omni](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-omni-3-5-videos/)** as a way to render a more fully realized clip from that creative process. The proposed division of labor is useful: explore and direct with fast output, then produce a higher-fidelity result. Google's video-generation and editing demonstrations support the product reference, though the talk does not demonstrate an integrated refinement pipeline. McCallum's invitation is to consider applications beyond the familiar examples of navigating generated worlds.

### The infrastructure behind a smooth stream

[6:01](https://www.youtube.com/watch?v=Xln-On3syJk&t=361s)

Delivering that experience introduces several connected engineering requirements:

1. **Place GPU capacity near the audience.** A global application may need GPUs in multiple regions.
2. **Route users to serving capacity.** Each connection needs an appropriate destination and a GPU to run its workload.
3. **Establish media transport.** McCallum names WebRTC, ICE and TURN as parts of the connection setup.
4. **Connect models into continuous workflows.** More ambitious applications need several models working together through a real-time harness.
5. **Synchronize controls with frames.** User input must stay coordinated with generated output while the application maintains a smooth stream.

The last two requirements make this more than an endpoint that returns a video file. The application must coordinate computation, controls and media throughout an active session.

### A React surface with programmable Python underneath

[6:54](https://www.youtube.com/watch?v=Xln-On3syJk&t=414s)

uRun's proposed application boundary is a drop-in **React component** that provides interactive video while hiding the serving machinery. Behind it, McCallum describes a programmable **Python runtime** for composing complex pipelines with asynchronous generation. Avatar applications and live video transformations are two intended uses of that split: the frontend handles the user's interaction, while the backend coordinates the models.

For the webcam example, a minimal Python pipeline can express the composition with asynchronous iteration. The following application-level function accepts incoming frames, a transformation function and an output sink; it is not a uRun SDK signature:

```python
from collections.abc import AsyncIterable, Awaitable, Callable
from typing import TypeVar

Frame = TypeVar("Frame")

async def transform_webcam(
    frames: AsyncIterable[Frame],
    transform: Callable[[Frame, str], Awaitable[Frame]],
    send: Callable[[Frame], Awaitable[None]],
    current_prompt: Callable[[], str],
) -> None:
    async for frame in frames:
        prompt = current_prompt()
        transformed = await transform(frame, prompt)
        await send(transformed)
```

The prompt is sampled for each input frame, and each transformed result is sent before the next frame is processed. That preserves ordering in this small pipeline while allowing the prompt to change between frames. GPU scheduling, media transport and tighter synchronization remain responsibilities of the surrounding runtime. The abstraction makes the model operation programmable without making the application author implement every part of the streaming service.

### Agents need access to the same building blocks

[7:39](https://www.youtube.com/watch?v=Xln-On3syJk&t=459s)

The final layer is tooling for agents that build applications. McCallum argues that platforms in 2026 also need to function as software factories, with interfaces that agents can use directly. He says uRun has built CLI and MCP server access for constructing these interactive video applications. These are capabilities presented in the talk; the public overview does not establish their current API contracts.

**The frontier is increasingly in how the models are served.** Continuous generation becomes a usable medium when users can connect, steer it and receive coherent output without managing the underlying GPU and streaming systems themselves. McCallum closes by seeking design partners who want to push human-computer interaction in that direction, alongside an invitation to join uRun.

## Resources

- [uRun](https://urun.sh/) — Overview of uRun's continuous video sessions, live controls and interactive application infrastructure.
- [Helios technical report](https://arxiv.org/html/2603.04379v1) — Architecture, training methods and benchmarks for a 14B model designed for fast, continuous video generation.
- [Helios code and checkpoints](https://github.com/PKU-YuanGroup/Helios) — Installation instructions, model checkpoints and text-, image- and video-conditioned inference examples.
- [Wan2.1](https://github.com/Wan-Video/Wan2.1) — The open video-generation project underlying Helios's initial 14B model.
- [LemonSlice avatar research and updates](https://lemonslice.com/blog) — Research and product announcements covering interactive avatars, model releases and integrations.
- [Google's Gemini Omni demonstrations](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-omni-3-5-videos/) — Examples of generating and editing videos through multimodal inputs and conversational instructions.

## Slides

### Something else is happening

The speaker shifts from video quality to two other areas of improvement: efficiency and long-horizon generation.

[1:01](https://www.youtube.com/watch?v=Xln-On3syJk&t=61s) [Text reads, “No wonder we talk about quality. But...something else is happening.”](https://assets.ai.engineer/sites/aiecode2025/sha256/639d5390e3c4fc72fed4c60d608ef084eb941098df5bd1493812a13c190a9082.webp)

No wonder we talk about quality.

But...something else is happening.

uRun

### Same prompt: one is real time

The same prompt produces two flower scenes: Helios 14B runs at 19.5 fps versus Wan 2.1-14B’s 0.33 fps on the same H100. The speaker estimates the Helios clip cost about one-hundredth as much.

[2:14](https://www.youtube.com/watch?v=Xln-On3syJk&t=134s) [Two flower-themed video frames compare Wan 2.1-14B at 0.33 fps with Helios 14B at 19.5 fps on the same H100.](https://assets.ai.engineer/sites/aiecode2025/sha256/5ed17ddb96323a82997f410382694d09a1defc95296f96f2da5cabc08000b1e0.webp)

Same prompt. One of these is real time.

A
B

A: Wan 2.1-14B — 0.33 fps · B: Helios 14B — 19.5 fps, same H100

### Not just one model

The techniques span world models with long-horizon consistency and camera control, avatar models, and video-to-video models that transform images in real time.

[2:40](https://www.youtube.com/watch?v=Xln-On3syJk&t=160s) [Four examples labeled LingBot-World, DreamX-World, LiveAvatar, and SANA-Streaming.](https://assets.ai.engineer/sites/aiecode2025/sha256/c11c035a5eab0288e91f3e6a3d1c9aa08b42e6041711e989febed8fb404c399a.webp)

Not just one model.

LingBot-World
DreamX-World
LiveAvatar
SANA-Streaming

Embedded image text: -World; D

### An explosion of innovation

Floor counts of real-time, DiT-based distilled video models rise from 2+ in 2024 to 40+ in H1 2026. The live arXiv/HF search undercounts models, especially in the latest period.

[3:09](https://www.youtube.com/watch?v=Xln-On3syJk&t=189s) [Bar chart shows at least 2 real-time distilled video models in 2024, 6 in the first half of 2025, 22 in the second half, and 40 in the first half of 2026.](https://assets.ai.engineer/sites/aiecode2025/sha256/e9884da37598d3d6ea8f3a0a3adcbacbb75e3ede366aa25294078febd99e5178.webp)

An explosion of innovation.

Real-time, DiT-based, distilled video models — at least this many, per period

2024: 2+
H1 2025: 6+
H2 2025: 22+
H1 2026: 40+

Floor counts from a live arXiv/HF search — undercounted, especially the most recent period.

### Token costs and live-generated video

The speaker compares an hour of token spending with estimates of three hours of continuous video for $10 and a full day of live-generated video for $50.

[3:35](https://www.youtube.com/watch?v=Xln-On3syJk&t=215s) [The slide compares $10 with approximately three hours and $50 with approximately a full day of live-generated video.](https://assets.ai.engineer/sites/aiecode2025/sha256/537757d33cb8db88105808ae8ac85e03ec4f5ab2e5a70350af72514aec65b331.webp)

Who's burned $10, or $50, of tokens in an hour?

$10 ≈ three hours · $50 ≈ a full day of live-generated video.

Interface text:
uRun

### Mirror, mirror—show me anything

A magic mirror could transform a webcam feed in real time, letting someone preview outfits, haircuts, or themselves in a car they like.

[3:50](https://www.youtube.com/watch?v=Xln-On3syJk&t=230s) [Text reads: Mirror, mirror—show me anything.](https://assets.ai.engineer/sites/aiecode2025/sha256/f9a98ee8f199a3bedaaabd6ccbc81039f16d1d7a79eaf06714f001d6348f942b.webp)

Mirror, mirror—
show me anything.

uRun

### A different way to reach everyone

Visual AI companions could make experiences like coding models more accessible to people who think and learn visually.

[4:30](https://www.youtube.com/watch?v=Xln-On3syJk&t=270s) [Text reads: A different way to reach everyone.](https://assets.ai.engineer/sites/aiecode2025/sha256/e3be28df5d5f62d9046cb9b87838a8e91531f80fd080d24d73939c7f253ba18a.webp)

A different way
to reach everyone.

### A generative camera

The speaker describes steering these models in under a second while they generate, giving creators finer control over shots than prompting and waiting for a result.

[5:00](https://www.youtube.com/watch?v=Xln-On3syJk&t=300s) [A generative camera—not a slot machine you pull and pray.](https://assets.ai.engineer/sites/aiecode2025/sha256/43347e143bde90755f766a9247e67f2e947a8bda646340d2d4297714e2fbdbd5.webp)

A generative camera—
not a slot machine you pull and pray.

Interface text: [illegible]

### Worlds from a single image

The speaker asks the audience to look beyond familiar world models, introducing the idea of worlds that dream themselves from a single image.

[5:46](https://www.youtube.com/watch?v=Xln-On3syJk&t=346s) [Worlds that dream themselves—from a single image.](https://assets.ai.engineer/sites/aiecode2025/sha256/4796dfe95ed07f5eb7daed887865ea0b65b64c5f0700e0f2963661a875ad4ce0.webp)

Worlds that dream themselves—
from a single image.

### What does an app like this need?

A global audience may need GPUs worldwide and routing to connect users. The stack also includes WebRTC, ICE and TURN, plus multiple models streaming in sync with user controls on every frame.

[6:00](https://www.youtube.com/watch?v=Xln-On3syJk&t=360s) [A list of requirements: worldwide GPUs, placement and routing, WebRTC, ICE, TURN, and synchronized models.](https://assets.ai.engineer/sites/aiecode2025/sha256/7ac0279134811556b9b10bd8c1e5d9b1a50a775a5a7e1dfd177a3ee4996a865b.webp)

what does an app like this need?

GPUs, worldwide.
Placement and routing.
WebRTC. ICE. TURN.
Multiple models, wired together.
Synced. Every frame.

Interface text:
urun

### A React component

The React example creates a live GPU session, renders it with UrunVideo, and updates the prompt mid-stream. The slide labels this a shipped API.

[6:52](https://www.youtube.com/watch?v=Xln-On3syJk&t=412s) [A React example creates a live GPU session, renders video, and updates its prompt mid-stream.](https://assets.ai.engineer/sites/aiecode2025/sha256/d8ebbe9bd4a5ee5ae7164ccb540fdaf658890fb128e69d2b1428cf8bcfe33450.webp)

What if it was just a React component?

function LiveScene({ prompt }) {
  const session = useApp().generate({ prompt })   // a live GPU session

  return <UrunVideo session={session} />          // that's it
}

session.doc('control').set({ prompt })             // re-steer, mid-stream

@urun-sh/react — real shipped API

urun

### Sessions programmable in Python

The LiveAvatar pipeline sends microphone audio through a realtime model, renders avatar frames, and combines them with audio for streaming at 24 fps. The slide says this code runs on B200s today.

[7:11](https://www.youtube.com/watch?v=Xln-On3syJk&t=431s) [Python code combines microphone input, a realtime model, avatar frames, and audio-video streaming.](https://assets.ai.engineer/sites/aiecode2025/sha256/df90f063b095b776cee5011c38229715393c5831bde00e38a641d6ad291c5c71.webp)

And behind it: sessions, programmable in Python

voice = ctx.stream("mic", codec="opus") \
    | ctx.realtime(model="gpt-realtime")

avatar = voice | ctx.window(2, stride=2) \
    | ctx.source(render_avatar) | ctx.pace(fps=24)

frames = [avatar, voice] | ctx.liveness(idle=render_idle)

pipeline = [frames, voice] | ctx.mux(fps=24) \
    | ctx.stream("av", codec="h264", audio_codec="opus")

This is LiveAvatar — real code, running on our B200s today.

Interface text: urun

### We need factories

The speaker calls for software factories that agents can interact with, and says the team built one accessible through a CLI or MCP server.

[7:37](https://www.youtube.com/watch?v=Xln-On3syJk&t=457s) [In 2026, we don't just need platforms. We need factories.](https://assets.ai.engineer/sites/aiecode2025/sha256/e4ca22a22d2b23cee97f1979e07cb94d9e1cbde2bac9191d1e213eb8e851d553.webp)

In 2026, we don't just need platforms.

We need factories.

### So we built one

The slide sketches a live avatar build: create a repo and PR, deploy, verify a decoded frame in a browser, and watch production logs. The speaker says the factory connects through a CLI or MCP server.

[7:47](https://www.youtube.com/watch?v=Xln-On3syJk&t=467s) [Slide showing a command to build a live lip-syncing avatar and steps for creating, deploying, verifying, and monitoring the application.](https://assets.ai.engineer/sites/aiecode2025/sha256/6d826644bb820710e649340d6bb714553812fd1b88f43177daafd7f1777a19f2.webp)

So we built one.

$ build_app("a live avatar that lip-syncs to my mic")

  → repo created, PR opened
  → deploy converges
  → decoded frame verified in a real browser
  → watching prod logs

MCP server · skills · CLI

Interface text:
uRun

### The frontier is serving models

With models available, the speaker identifies serving them as the frontier and calls for software factories that agents can access through a CLI or MCP server.

[7:59](https://www.youtube.com/watch?v=Xln-On3syJk&t=479s) [Slide reading, “The models are here. The frontier is serving them.”](https://assets.ai.engineer/sites/aiecode2025/sha256/737de633563d2d414be57feea847bfa12fb3fcadfa515a83bbe7177699c0b219.webp)

The models are here.

The frontier is serving them.

urun

### Design partners and hiring

uRun is seeking design partners working on human-computer interaction and is hiring. The speaker invites interested people to talk after the session.

[8:08](https://www.youtube.com/watch?v=Xln-On3syJk&t=488s) [Closing slide seeking design partners working on human-computer interaction and announcing hiring, with contact details.](https://assets.ai.engineer/sites/aiecode2025/sha256/047ea698178928ed5fa7c1cb80f9a0dd41822e63fb2fcbe2da5e00914408e7af.webp)

We're looking for design partners pushing the boundaries of human-computer interaction.

And we're hiring.

urun.sh · hello@urun.sh

urun

## Full transcript

[0:00](https://www.youtube.com/watch?v=Xln-On3syJk&t=0s) [on-hold jingle] I am Keegan. I'm the founder of uRun, um, a new kind of inference provider

[0:20](https://www.youtube.com/watch?v=Xln-On3syJk&t=20s) focused around, uh, interactive media. And I'm here to talk about generative video. So we hear a lot about generative video improving along the quality axis at the frontier. We have the classic Will Smith eating spaghetti from twenty-twenty-three.

[0:36](https://www.youtube.com/watch?v=Xln-On3syJk&t=36s) It is nightmare fuel and not something you would ever mistake for reality.

[0:42](https://www.youtube.com/watch?v=Xln-On3syJk&t=42s) In twenty-twenty-four, we got Sora, and it gets a little better. It'll-- still has a bit of, you know, an AI feel to it, but it, it's getting there. And Sora 2, you know, even better.

[0:54](https://www.youtube.com/watch?v=Xln-On3syJk&t=54s) But Seedance this year, um, absolutely incredible. So photorealistic and it's, it's no wonder that we talk a lot about quality. But I'm here to talk about another axis which models are improving along, which is efficiency and, uh, h- the long-horizon generations. [sighs]

[1:16](https://www.youtube.com/watch?v=Xln-On3syJk&t=76s) So what you're watching here is a demo for a model called Helios that we serve at uRun. Um, the generation in the bottom right corner, you'll see, is a long, continuous generation, and the other video, um, is a bunch of clips, um, that have been generated faster than you can consume them.

[1:35](https://www.youtube.com/watch?v=Xln-On3syJk&t=95s) Uh, and they're about at the same quality as the frontier models were last year. [inhales] They're-- Uh, Helios is a distill of Wan 2.1 14B, um, and I'll talk a bit about the techniques that are used in the various models that are hitting the scene right now.

[1:56](https://www.youtube.com/watch?v=Xln-On3syJk&t=116s) But there's been an explosion in just the last year, uh, in terms of efficiency and capabilities.

[2:03](https://www.youtube.com/watch?v=Xln-On3syJk&t=123s) Um, so like looking at this, I kinda ruined it with the last clip, but you can guess which one is real-time, um, and which one was generated in, uh, a number of minutes.

[2:15](https://www.youtube.com/watch?v=Xln-On3syJk&t=135s) And the one on the right, uh, is in-- arguably a bit better. It's got better motion, um, and it was generated for about a one-hundredth of the cost.

[2:28](https://www.youtube.com/watch?v=Xln-On3syJk&t=148s) And these are just some of the charts showing, um, the quality bar, uh, for both long and short video generation. Uh, Helios came out in March, and it's, it's pretty incredible to see how fast these are improving.

[2:42](https://www.youtube.com/watch?v=Xln-On3syJk&t=162s) But these are techniques that are being applied, uh, all over the place, not just to one model. Uh, there's world models which can keep consistency over long horizons, and you can control, uh, in a fine-grained way, uh, the camera and the viewport.

[2:57](https://www.youtube.com/watch?v=Xln-On3syJk&t=177s) Uh, there's avatar models like we just talked about with Lemon Slice, um, and there's video-to-video models that can, can transform, uh, what you're seeing in, in real-time, almost like a, a magic mirror.

[3:10](https://www.youtube.com/watch?v=Xln-On3syJk&t=190s) Uh, there's actually been an explosion of innovation. There's been, uh, at least forty models, uh, with real-time capabilities, uh, and long-horizon generation capabilities released this year. Uh, show of hands, who ha-- here has burned ten or even fifty dollars worth of tokens in an hour with Claude Code?

[3:33](https://www.youtube.com/watch?v=Xln-On3syJk&t=213s) A lot of people. Um, and so we're at a place right now where ten dollars can get you three hours worth of generated video continuously with most of these models, and fifty dollars would give you an entire day interacting with an AI in a visual medium.

[3:49](https://www.youtube.com/watch?v=Xln-On3syJk&t=229s) Fifteen hours. And so I wanna talk a little bit about the different things this enables in terms of the way that we interact with computers, um, a-and I'll talk a little bit about what we're doing at uRun to try and make it easier for folks to experiment and build out applications like this.

[4:07](https://www.youtube.com/watch?v=Xln-On3syJk&t=247s) So one such use case would be a magic mirror. You could have, uh, your webcam, and you could ask to see yourself in any outfit. You could ask to, uh, see yourself in a car you like or with a haircut you're considering.

[4:23](https://www.youtube.com/watch?v=Xln-On3syJk&t=263s) Um, a lot of different, uh, possibilities because these are open-ended models that can transform, um, what they're seeing on a webcam in real-time. I also think about accessibility a lot with these models.

[4:34](https://www.youtube.com/watch?v=Xln-On3syJk&t=274s) Um, you know, working with AI involves a lot of reading and a lot of text. Uh, for some people, that's more difficult. Uh, for some people, they just don't think, um, in, in text.

[4:47](https://www.youtube.com/watch?v=Xln-On3syJk&t=287s) They think visually and learn better that way. Uh, so there's more opportunities to have companions or visual mediums that are gonna allow more people to experience, uh, the things a lot of us have with coding models.

[5:02](https://www.youtube.com/watch?v=Xln-On3syJk&t=302s) And I'm excited about content creation. Um, so far, we very much had a slot machine type approach where you're setting up a prompt and maybe some keyframes and spending about ten dollars a minute, uh, to try and get the shot that you want.

[5:17](https://www.youtube.com/watch?v=Xln-On3syJk&t=317s) But with these models, you can actually steer them in real-time, um, in under a second while they're generating and get the actual shots that you want. Maybe you're piloting an agent that it-- you're able to look over its shoulder and see what it's generating in real-time.

[5:33](https://www.youtube.com/watch?v=Xln-On3syJk&t=333s) Um, but you're able to more granularly control the content you're generating. And with modern models like Google, uh, Gemini Omni, you can actually render these out as a more full-fidelity clip.

[5:47](https://www.youtube.com/watch?v=Xln-On3syJk&t=347s) And of course, we all are thinking about world models, uh, but I wanna take the, the focus off of just kind of the, the, the basic, uh, world models that we talk a lot about and just try to expand, uh, the horizons of what we can do with this technology.

[6:01](https://www.youtube.com/watch?v=Xln-On3syJk&t=361s) And so what does it look like to actually build an application like this? So you're gonna need GPUs all over the world, potentially, if you've got a global audience that's gonna be using these.

[6:14](https://www.youtube.com/watch?v=Xln-On3syJk&t=374s) You're gonna need to think about, uh, where you're connecting the, uh, users to, what GPUs you're gonna use to serve them. You're gonna need to set up probably WebRTC and ICE and TURN.

[6:27](https://www.youtube.com/watch?v=Xln-On3syJk&t=387s) Um, and for the most interesting use cases, you're gonna wanna model, um, wire multiple models together in continuous streaming workflows, um, building those real-time harnesses. And you're gonna want, um, things synchronized with your controls that you're providing to your end users, um, with every frame, uh, and continually providing a smooth streaming experience.

[6:54](https://www.youtube.com/watch?v=Xln-On3syJk&t=414s) And so our idea is what if there was just a React component that you could drop into your application, uh, to make it easy to provide video interactively inside your applications with any model.

[7:13](https://www.youtube.com/watch?v=Xln-On3syJk&t=433s) And behind the scenes, there's a programmable Python runtime that lets you easily build these complex pipelines generating asynchronously, um, so that you can build avatar models, you can build these video-to-video transformation models.

[7:32](https://www.youtube.com/watch?v=Xln-On3syJk&t=452s) You can experiment and, and build whatever you can really imagine on top of these.

[7:39](https://www.youtube.com/watch?v=Xln-On3syJk&t=459s) And I argue that in 2026, we don't just need platforms, we need software factories and ways for our agents to interact with these. And so we've actually built one that will let folks hook into a CLI or an MCP server and build these kinds of applications.

[7:58](https://www.youtube.com/watch?v=Xln-On3syJk&t=478s) And so the models are here, and the frontier is really in how we serve them. Uh, I

[8:09](https://www.youtube.com/watch?v=Xln-On3syJk&t=489s) went way over-- I went way under time. Um, [laughs] but we are looking for design partners who wanna push the boundaries of human c- human-computer interaction, and we're hiring at uRun.

[8:20](https://www.youtube.com/watch?v=Xln-On3syJk&t=500s) Um, so come see me after the talk if, uh, if you're interested in chatting more. [upbeat music]
