GenerativeImmersion

The same people, the same place and synchronized sound, held together across long continuous shots.

Seedance 2.5 generates up to 30 seconds of video and sound in a single pass, and extends a scene round after round while keeping its characters and setting. It can draw on up to 30 images, 10 video clips and 10 audio clips at once, and on untextured 3D renders that fix the blocking and the camera path.

Edits can target a single timestamp, the camera move or a green-screen background, while the rest of the scene stays as it was.

For researchKeeping everything constant while one element changes is the logic of an experimental manipulation.

More from the same release

From scenes you watch to spaces you move through: generated as you explore, or built once and kept.

Google DeepMind · Genie 3: creating dynamic worlds you can navigate in real time

Genie 3 generates a world from a text prompt in real time, at 720p and 24 frames per second, as you move through it, and keeps it consistent for several minutes. Text prompts can change the world on the way: the weather turns, a person appears. Since January 2026, Project Genie lets Google AI Ultra subscribers create such worlds and explore them for 60 seconds at a time.

World Labs · Introducing Marble

Marble builds persistent worlds from text, photos, video or a rough 3D layout. They can be edited and combined, then exported as Gaussian splats or meshes, for example into a game engine.

For researchA world that keeps generating can respond to what a participant does, not only show what everyone should see.

Also moving

• Tencent · Apr 2026 · openHY-World 2.0 generates or reconstructs editable 3D worlds from text, images or video, as meshes or Gaussian splats.
• Meta Reality Labs · Nov 2025 · researchWorldGen builds traversable, interactive 3D worlds from a single text prompt.
• ByteDance · 2026 · not releasedA real-time world model built on Seedance and aimed at Pico headsets, reported in September and first shown in October 2026.