Meta Superintelligence Labs Launches Muse Image and Muse Video

Meta Superintelligence Labs released Muse Image and Muse Video. The release focuses on instruction following and social context use.

Muse Image ranks as the current leader in controlled image output. It handles exact prompts, precise edits, and multi-image layouts. The model also pulls Instagram context to improve relevance. Users access it inside the Meta AI app, on the web, in Instagram Stories, and on WhatsApp. Availability starts in a limited set of countries.

Muse Video builds on the same base model. It adds native audio support while keeping high visual quality. Both tools include agent capabilities through Muse Spark, a coordination layer that enables the models to call external tools while generating media. For example, a designer could prompt Muse Image to create a product mockup that matches the visual style of recent posts from their brand’s Instagram followers, while a content creator might use Muse Video to produce a short clip that includes synchronized voiceover and background sound without post-production editing.

The announcement came via an X post from the official AI at Meta account on July 8, 2026. The post described the models as the first media generation systems from the new lab (@AIatMeta).

Release Details

Meta Superintelligence Labs built both models on a shared pre-trained foundation. Muse Image adds fine-grained editing and reference composition. Muse Video extends the same foundation to moving images with sound.

Early access is restricted. Users in supported regions can test the tools through existing Meta apps. The company has not released a full rollout schedule.

Technical Approach

The models emphasize agent tool use. Muse Spark acts as the coordination layer. This layer lets the models call external tools while generating media. Instagram context integration helps the system understand user style and preferences without extra prompts.

Visual fidelity in video output stays high. Audio is generated together with the visuals rather than added afterward. The shared training base reduces drift between image and video results.

User Access Path

People reach the models through Meta AI surfaces. Instagram Stories offers quick image experiments. WhatsApp supports longer conversations that include generated media. The web version gives desktop editing controls. Each channel respects the initial country limits.

What Remains Unclear

The post did not list model sizes or training data sources. Independent benchmarks have not yet appeared. It is not known how the models compare with current leaders on standard tests. The company has not shared plans for commercial licensing.

Next Signals

Watch for expanded country access in the next three months. Check whether third-party reviews appear on established AI evaluation sites. Track any updates on Muse Spark tool integrations. Monitor competitor responses from other large labs.

These signals will show whether the models move beyond early limited release.