Meta Muse Video: Release Date, API Access, and What Meta Connect Might Change

Meta Muse Video has been in a public leaderboard for most of the summer, which is a strange place for a model nobody can call. Meta previewed it on July 7, 2026 alongside Meta Muse Image, promised access "soon to creators and in Meta AI," and then said nothing for eleven weeks. No release date. No public API. No pricing, resolution, clip length, or regional list. What it does have is a set of developer-platform traces pointing at an unlaunched Video Playground, and a Meta Connect keynote landing on September 23, 2026 — the day this is being written. This is a what-we-know-so-far piece: everything below is either vendor-confirmed, independently measured, or explicitly flagged as unverified.

What Meta has actually confirmed

Less than the volume of coverage suggests. Meta's own preview post is the only primary source, and it commits to a short list of claims:

• Native audio. Meta Muse Video generates sound in the same pass as picture rather than bolting it on afterwards — speech, ambience, sound effects, and music. This is the one capability Meta leads with, and it is what separates it from Meta Muse Image's silent output.

• A shared pretraining base with Meta Muse Image. Meta says the two share a foundation. It does not say which of Meta Muse Image's controls — instruction following, multi-reference editing, self-refinement — carry over to video.

• Two acknowledged weaknesses. Meta itself flags that fast-motion physics and audio-video synchronization are still improving. That is vendor-reported, and it is worth taking literally: "native audio" does not mean flawless lip sync.

• Content Seal watermarking. Meta's invisible provenance watermark ships in Meta Muse Image, and Meta has said it plans to extend it to video. Detection tooling is not fully documented.

That is the confirmed set. Everything else in circulation — clip duration, resolution, price, regions, export terms, commercial use — is inference.

The Video Playground leak, and what it does not say

The signal that put Meta Muse Video back on the radar came from TestingCatalog, which reported that three unreleased sections exist in Meta's developer platform: Video Playground, Voice Playground, and Connectors. Two of those three have since shipped or partially shipped. Meta opened Muse Connectors to developers on September 19, 2026, letting services such as Notion, Granola, and Stripe payments plug into the Muse agent. Voice Playground has no public counterpart yet.

What the report does not contain matters as much as what it does. There is no verbatim code, no API endpoint string, no request or response schema, no model identifier, and no Meta confirmation. A developer-platform section existing is evidence of intent to expose something to developers. It is not evidence that a video endpoint is built, priced, rate-limited, or scheduled. Treat the Video Playground as a strong directional signal and a weak scheduling one.

Three leaderboard snapshots that disagree with each other

Meta Muse Video's ranking is the most-reported and least-stable number attached to it. Three Arena text-to-video snapshots, spread across the summer, do not agree:

• July 5, 2026 snapshot — Meta Muse Video at #3 on the text-to-video board, Elo 1459, behind Gemini Omni Flash at 1527 and Seedance 2.0 at 1482, and ahead of Sora 2 Pro at 1366 and Veo 3.1 at 1364.

• August 14, 2026 snapshot — Meta Muse Video at #5, Elo 1457, across 616,845 votes and 45 models, with MiniMax H3 and HappyHorse 1.0 having moved ahead of it.

• September 22, 2026 snapshot — the board as we captured it directly, the day before Meta Connect. Meta Muse Video sits at #9 of 48 models with a score of 1456 and 2,188 votes, having been passed by Wan 3.0, Dreamina Seedance 2.5, and MiniMax H3. Gemini Omni 1.1 Flash leads at 1516.

Three things are worth pulling out of that. First, the score barely moved across eleven weeks (1459, 1457, 1456) while the rank fell six places — the field around it got better, which is the normal life of an unreleased model parked on a board it cannot be re-tested against. Second, the vote count is the tell: 2,188 votes is a rounding error against a board carrying 718,577 of them, so the confidence interval around that rank is wide and the number should not be quoted to three digits. Third, and most important, a board that scores a model with no public access is measuring something other than production readiness. Rank on a text-to-video board is not a statement about whether you can build on it, and it is not a substitute for running your own footage through it.

What is still unknown

If you are trying to decide whether Meta Muse Video changes a production plan this quarter, these are the open variables, and none of them are announced:

• Release date — no day, no window, no commitment beyond "soon."

• Access path — whether it arrives as a consumer feature inside Meta AI first, a developer endpoint, or both simultaneously.

• Clip length and resolution — undisclosed. Closed-beta samples have circulated at short durations; nothing has been published as a limit.

• Pricing and caps — no credit structure, subscription tier, or rate limit has been described.

• Regions — Meta has not said which countries, apps, or business products get it first.

• Commercial terms — unresolved. Do not assume Meta Muse Image's terms carry over.

• Inputs — image-to-video, reference images, extension, and editing are all undocumented.

Any page quoting a specific launch day or a price for Meta Muse Video without a newer Meta announcement is speculating.

Where it would live if it ships

Meta has not described distribution, but the shape of the Muse stack makes the likely order guessable. Meta Muse Image already runs across selected Meta products and regions, and the natural first surfaces for a video model are Meta AI and meta.ai, followed by the Vibes feed, Instagram, Facebook, and the Edits app. A free tier would not be unprecedented — Meta Muse Image has one.

The cautionary data point is Meta Muse Image's Instagram launch, which was pulled on July 10, 2026 — three days after it went out — after SAG-AFTRA objected to automatic enrolment of creators. Meta said it missed the mark; Meta Muse Image itself stayed live. A video model with native audio and face-adjacent output will meet a stricter reception than a still-image editor did. Expect the first rollout to be narrow and reversible.

The stack Meta Muse Video would join

It is easier to reason about Meta Muse Video as the missing piece of an already-shipped family than as a standalone launch. Meta Muse Spark 1.3 landed on September 2, 2026 with native video understanding and a 1M-token context — that is a model that watches, not one that generates. Meta Muse Voice Transcribe handles speech-to-text. Meta Muse Image handles stills and editing. The Meta Muse agent, which shipped September 8, 2026 and briefly held #1 on the US App Store Top Free chart, is the surface that would consume all of it. Meta's current Model API covers Meta Muse Spark 1.3, Meta Muse Image, and Meta Muse Voice Transcribe. There is no video-generation model and no voice-generation model on it today.

That gap is the whole reason the Video Playground trace is interesting. It is the first evidence that the API layer, not just the consumer app, is being prepared for one.

For teams that want to move before Meta commits, the practical pattern is to keep the generation step swappable. OrcaRouter exposes 200+ models behind a single OpenAI-compatible endpoint with automatic failover, so a pipeline can call a video model today and re-point at Meta Muse Video later without a second contract or a rewrite — and because we pass provider list pricing through at 0% markup, a vendor price cut shows up on our side the same day rather than at the next invoice cycle. To be explicit about what that does and does not mean: Meta Muse Video is not routable on OrcaRouter, and it will not be until Meta ships it. What is routable today from Meta's family is Meta Muse Spark 1.2, which is the model to reach for if you want the Muse line in a working pipeline right now.

What to do before the keynote

Meta Connect runs September 23–24, 2026 at Menlo Park, with the Zuckerberg keynote on Wednesday September 23 at 4:00 PM PT and developer sessions Thursday September 24 at 10:00 AM PT. If Meta Muse Video appears at all, that is where.

The useful preparation is not watching the stream. It is having a comparison ready. Write down three briefs before the announcement — one face-driven, one product shot, one fast-motion — and define what "good enough" means for each in terms of audio sync, temporal consistency, and retry cost. When access opens, run the same three briefs through Meta Muse Video and through whatever model you are using now, and compare approved clips rather than demos. A model that ranks #9 on a board and one that ranks #3 can produce identical output on your material, and the only way to know which is to have the test already written.

Until then, the honest summary is this: Meta Muse Video is a real model with a real capability gap it fills, a real developer-platform signal behind it, and zero published terms. The release date is not knowable from the outside. The preparation is.