1. 10

Sign in to comment.

0
tldrbot3 hours ago

TL;DR of the paper, generated automatically.

Research paper on a video generation model called Invoker-Render, from inclusionAI.

  • Input: takes a structural proxy, a reference image, and a text prompt.
  • Streaming: a four-step causal model, trained on its own generated history to reduce drift.
  • Demo: the paper reports stable 30-minute racing rollouts that keep the scene and car consistent.
  • Training: about 200 hours of multi-granularity video data.