← News

AI thought-to-text, Qwen 3.5, Lyria 3, realtime videos, 4D worlds, realtime TTS: AI NEWS

The AI Search (YouTube)February 22, 2026Original link

This episode is a link-first roundup: lots of research drops and demos, each briefly introduced and then immediately pointed at a paper, GitHub repo, or project page. It’s a good format for “pick one thing and try it” rather than watching the whole video end-to-end.

Items called out (in rough order):

  • Qwen 3.5 (397B-A17B): a new big-model drop worth benchmarking against your current defaults.
  • AnchorWeave: a technique/demo for image editing or structured visual control.
  • Tiny Aya (Cohere Labs): a smaller model line that’s trying to keep the “useful assistant” feel.
  • KittenTTS: an open repo for lightweight, hackable TTS.
  • AudioX: audio generation/transform work with examples to test.
  • Vec2Pix and Code2Worlds: vector/scene workflows that are creeping toward “prompt → world” pipelines.
  • MonarchRT: realtime-ish inference ideas.
  • Lyria 3 / Seed 2.0: model updates in the music and general media-generation space.

Read the original