Daily · Jul 28, 2026

Seedance 2.5 reaches the platform layer; World Labs builds generated worlds for robot learning

Model access and spatial intelligence advanced together: Runway previewed Seedance 2.5, while World Labs showed generated environments becoming training grounds for embodied systems.

Industry01

Runway previews Seedance 2.5 on its multi-model platform

Runway announced Seedance 2.5 was coming to its platform, extending its strategy of hosting competing frontier models alongside its own. Broad distribution shortens the gap between model release and production access, but it also makes the surrounding workflow more important than the logo on the endpoint. Re-run a fixed shot suite when it lands and compare usable-shot rate, not launch-reel quality.

Source Runway @runwayml · 2026.07.28

Industry02

World Labs turns generated worlds into robot training environments

World Labs shared early work using generated worlds to scale robot learning after integrating SceniX. This moves world models beyond visual exploration toward persistent geometry, interaction and repeatable physical tasks. For virtual production and previs teams, the same requirement is useful: judge a generated scene by what actors and cameras can do inside it, not only by how convincing one view looks.

Source World Labs @theworldlabs · 2026.07.28