Media Models Become Agents
Generative image, video and audio models increasingly call external tools — code execution, search, retrieval — mid-generation rather than producing output in a single forward pass, so expect recurring tool-using media models, verification of factual content inside generated media, and the media stack converging on the same agentic scaffolding as text models.
weakening · confidence 36 · Emerging (watchlist) · tracking since September 13, 2026 · updated September 14, 2026
Score history
Daily conviction score, 0 to 100. Higher means the thesis is more strongly corroborated.
Now 36 · -2 since Sep 13 · ranged 36 to 38
Why the conviction moved
- Sep 13Strengthened
Meta Superintelligence Labs' first media models ship with Muse Image calling its own coding and search tools mid-generation, with Muse Video adding native audio and remaining in preview. Tool invocation inside the generation loop breaks the single-pass assumption that has defined diffusion-era media models.
Source trail
Supporting · September 13, 2026
Meta's Muse Image Calls Its Own Coding and Search Tools Mid-Generation
Meta Superintelligence Labs' first media models ship with Muse Image calling its own coding and search tools mid-generation, with Muse Video adding native audio and remaining in preview. Tool invocation inside the generation loop breaks the single-pass assumption that has defined diffusion-era media models.
Meta AI
Unlock full source trail, score history, and daily updates.
Unlock TrendsAffected regions & assets
Townsquare
Argue the thesis in Townsquare.