/Tech4h ago

Runway launches Runway Characters, its first autoregressive video model for real-time avatars

It pins KV caches to maintain CUDA graph compatibility.

122363.5K

Original post

A great technical blogpost about @runwayml's first of it's kind autoregressive video model:

One of the most cracked, under the radar research teams is at @runwayml.

They were on the OG Stable Diffusion paper and just released the first autoregressive video model.

Did a technical deep dive on this + some interesting inference optimizations:

https://www.amplifypartners.com/blog-posts/the-infrastructure-behind-runway-characters

9:27 AM · Jun 25, 2026 · 1.8K Views

Sentiment

Users dismissed RunwayML's new autoregressive video model as unnecessary amid existing market saturation in generative video tools.

Pos

0.0%

Neg

100.0%

1 comments with sentiment.

Cluster Engagement

Digg Deeper

No Digg Deeper questions have been answered for this story yet.

AMPLIFYPARTNERS.COMVia

Posts from X

Most Activity

VIEWS106RETWEETS1

Video models are hard. Real-time video models are even harder.

Loved this breakdown by @itunpredictable on how we productionized Runway Characters, our real-time interactive avatar model.

We had to figure out a lot of things like causal training, lipsyncing, parallelizing the VAE and pinning the KV cache so CUDA graphs don't break.

One of the most cracked, under the radar research teams is at @runwayml.

They were on the OG Stable Diffusion paper and just released the first autoregressive video model.

Did a technical deep dive on this + some interesting inference optimizations:

https://www.amplifypartners.com/blog-posts/the-infrastructure-behind-runway-characters

3h1.8K173

Strata@ChainZenit

@sarahcat21 @runwayml Another generative video model, just what the saturation needs.

4h9