Muse Image and Muse Video are media-generation models from Meta Superintelligence Labs (MSL), the Meta AI division led by Alexandr Wang, announced July 7, 2026. They are the first media-generation models developed by MSL and, following Muse Spark (April 2026), the lab's second model release overall (Source: ai.meta.com; siliconangle.com). Muse Image was released at announcement, free in the Meta AI app; Muse Video, built on the same pretraining base, was previewed as "coming soon" to creators and Meta AI (Source: ai.meta.com; cnbc.com). CNBC reported that Meta's internal tests place Muse Image behind OpenAI's GPT Image 2 but ahead of Google's Nano Banana 2, and framed the launch as part of Meta's bid to court advertisers and subscribers (Source: cnbc.com).
Snapshot
Benchmark position (Meta-reported)
| Date | Claim |
|---|---|
| 2026-07-05 | Muse Image holds the No. 2 spot on Arena for text-to-image, single-image editing, and multi-image editing, by human-preference Elo (Source: ai.meta.com) |
| 2026-07-05 | Muse Video ranks No. 3 on Arena in human-preference Elo for text-to-video (Source: ai.meta.com) |
Rankings are as reported in Meta's announcement, as of July 5, 2026; independent leaderboard positions move frequently.
Capabilities
Meta describes Muse Image as its most advanced image-generation model: it follows instructions, edits with precision, composes from multiple reference images, and draws on Instagram for social context. The model supports agentic tool use — SiliconANGLE's coverage highlights coding and search capabilities built into image generation — and integrates with Muse Spark, with the two models sharing tools and planning jointly. Meta cites combined code-and-media generation uses such as animated GIFs, websites with embedded images, and interactive visual games (Source: ai.meta.com; siliconangle.com).
Muse Video shares Muse Image's pretraining base and is described as delivering high visual fidelity with native audio support (Source: ai.meta.com).
In the announcement, Meta reports that quality scales approximately log-linearly in Elo with combined training compute across the two modalities — text tokens for reasoning and visual tokens for generation (Source: ai.meta.com). This is a developer-reported observation from internal ablations, not an independently verified result.
Availability
At launch, Muse Image was available in the Meta AI app and on meta.ai, in Instagram Stories in the US, and in WhatsApp in a limited set of countries, with Facebook availability described as coming soon. Muse Video was not yet released (Source: ai.meta.com). Meta withdrew the feature on July 10, 2026 amid the consent backlash described below (Source: reuters.com).
Provenance and watermarking
The launch also drew privacy criticism: Muse Image lets any user generate AI images incorporating photos from public Instagram accounts unless the account holder opts out via the "Sharing and reuse" setting. The default-on setting drew criticism from privacy advocates, and step-by-step opt-out guidance circulated widely by July 9, 2026 (Source: techcrunch.com; wired.com). See AI and Privacy. On July 10, 2026 — three days after launch — Meta discontinued the Muse Image feature in its apps, following objections from SAG-AFTRA, CAA, Public Citizen, and privacy advocates to the opt-out use of public Instagram likenesses (Source: reuters.com; axios.com). Critics quoted in the withdrawal coverage included actor Hannah Einbinder, and SAG-AFTRA said anything short of "a clear and conspicuous opt-in" was unacceptable given "the dangers of nonconsensual digital replicas"; Meta said the feature "missed the mark" (Source: reuters.com).
Images created by Muse Image in the Meta AI app and on meta.ai carry Content Seal, Meta's invisible watermarking system — a hidden provenance signal that Meta says survives cropping, compression, resizing, and screenshotting. Meta plans to extend Content Seal to video and is previewing a public detection tool (meta.ai/identification) for checking whether an image carries the watermark (Source: ai.meta.com). Unlike Muse Spark, which shipped with a 158-page safety-and-preparedness report, no standalone safety report for Muse Image had been identified at launch; the announcement's stated safety measures center on provenance.
Context
The release extends MSL's closed-weight direction — begun with Muse Spark's break from the Llama open-weight tradition — into media generation, and came during a period in which Meta's AI reorganization was under public scrutiny: at an internal town hall on July 2, 2026, Mark Zuckerberg said AI agent development was progressing slower than expected and that the benefits of the company's AI-focused restructuring had not yet materialized (Source: reuters.com). The launch places Meta in more direct competition with the incumbent frontier video and image generators (see Sora and Veo (Video Generation Models)).
Relationships
- developed-by: Meta AI (Meta Superintelligence Labs)
- depends-on: Muse Spark (Meta Superintelligence Labs) (tool sharing and joint planning; second MSL release after it)
- competes-with: Sora and Veo (Video Generation Models) (frontier media-generation landscape)
- related: AI Content Saturation (\"AI Slop\") (synthetic-media volume), AI Benchmarks and Evaluation (Arena Elo claims)