Skip to content
Bud's Wiki

Muse Video preview — MSL video generation model announcement (Alexandr Wang)

TL;DR

Section titled “TL;DR”

Tweet 4/ in Alexandr Wang’s July 7 2026 Muse Image launch thread previews a companion video-generation model, Muse Video, from Meta Superintelligence Labs. The tweet claims it’s “competitive on prompt adherence, visual fidelity, temporal consistency” and says it is “coming to meta ai soon.” No timeline, no architecture, no benchmarks, no independent evaluations — a product tease sitting inside the same thread that formally introduces Muse Image (MSL’s second major release, after Muse Spark in April). Notable as the third MSL-branded model publicly announced and the first video-generation model in the Muse family.

Key claims

Section titled “Key claims”
  • Muse Video is under development and “coming to meta ai soon” — no ship date given [tweet].
  • The model is claimed by Meta to be “competitive on prompt adherence, visual fidelity, temporal consistency” — Meta’s own internal framing, no third-party benchmarks cited [tweet].
  • The tweet embeds two short generated video clips as the sole visual evidence [tweet, video 1, video 2].

Method

Section titled “Method”

Not disclosed. The tweet is a product preview embedded in a launch thread and contains no architecture, parameter count, training data, or benchmark details. Meta’s Muse-family framing so far — Muse Spark (LLM), Muse Image (image generation, launched today), Muse Video (previewed here) — suggests a coordinated product-line release strategy rather than a single-model release.

Results

Section titled “Results”

None reported. Two short generated video clips are shown; the axes claimed (prompt adherence, visual fidelity, temporal consistency) are the standard T2V evaluation vocabulary but no scores against Veo 3, Sora 2, Kling, Seedance, or Wan are provided in the tweet.

Why it’s interesting

Section titled “Why it’s interesting”

This is the third MSL model announcement filed on the wiki: Muse Spark (LLM, Muse Spark — first model from Meta Superintelligence Labs (MSL)), and now — same day as this preview — Muse Image (image generation, formally launched by MSL today per CNBC coverage). Muse Video previewed here extends the Muse family into video, mirroring Google’s Gemini-family fan-out (LLM → image → video) and Alibaba Qwen’s (LLM → image → Omni) rather than the closed-frontier T2V labs’ single-modality focus (Runway, Moonvalley, LTX, Kling). The framing sits in the same closed-but-API-accessible cohort that Open foundation-model releases tracks as the baseline against open T2V releases like Wan 2.2, LongCat-Video, and LTX-2 — extending the datapoint of “MSL first ships closed and promises eventual open-source” from LLM to visual modalities.

See also

Section titled “See also”