Meta ships Muse Image and Muse Video, its first in-house generative media models

Meta released Muse Image and Muse Video on July 7, its first in-house generative image and video models, built by Meta Superintelligence Labs and rolling out across Instagram, WhatsApp and the Meta AI app.

The launch ends Meta’s reliance on outside models for the AI image and video features baked into its apps, giving the company control over the systems that will generate media for billions of users and, eventually, advertiser campaigns.

Muse Image is already live in the Meta AI app, on meta.ai, in Instagram Stories in the US and on WhatsApp in a limited set of countries, with Facebook support coming soon, Meta said. Muse Video is due to reach creators and Meta AI in the near term. Both models embed Meta’s Content Seal invisible watermark and will power a range of AI visual effects across Instagram and WhatsApp.

Muse Image runs coding and web-search tools as part of generating an image and uses extra compute at inference time to refine its own output before returning it — an agentic approach Meta says improves quality on harder prompts.

Meta’s performance claims should be read as the company’s own. It says Muse Image ranks second on the Arena leaderboard for text-to-image, single-image editing and multi-image editing, and that Muse Video ranks third for text-to-video generation as of July 5. Those standings are self-reported and not independently verified, and Arena rankings shift as new entries arrive. Whether Muse holds up against OpenAI, Google and specialist video startups will be tested once it is generating creative at ad scale.