Shaduf.Research preview
Real-Time AI Video/AlayaWorld browser route and slowed playout

Pool topic: Real-time, continuous, and interactive AI video generation. Question revision: 1. Exact question: What can generate continuous, interactive AI video today, and which setups work at which cost, latency, hardware requirements, and quality?

Checked 9 November 2026

AlayaWorld is steerable in a browser, but its B200 route plays at 9.7 FPS

Project says: its Reactor adapter takes a starting image, camera movement and prompt changes, then sends completed chunks over WebRTC. On one B200, the published camera-still setting delivers 9.7 frames per second to the browser. The service slows playback to generation pace because each 32-frame turn takes 3.2 seconds while its nominal 24-FPS media would last about 1.33 seconds.

Check which AlayaWorld path you mean

The repository’s route table says the browser adapter serves the DA3 path. Its newer v1.1 four-step DMD path writes a file through a validation loop. Do not attach a v1.1 speed claim to the browser route without a matched test.

The browser setup is verified on one B200 and reaches about 60 GB peak VRAM during load. The source does not give a matched consumer-GPU browser rate or a p95 control-to-visible delay. It labels the released code and weights academic/non-commercial under the LTX-2 Community License; check the exact rights before a paid stream.

Small next test

Before renting hardware, ask whether slowed motion is acceptable for your format. If you can test this route under its stated terms, record connection-to-first-frame, ten command-to-visible times, delivered FPS, stalls and visual quality over a fixed session. The pool has not run or paid for this setup.

Search published pools, pages, reports, and evidence.