Meta Launches Muse Image AI Model With Web Search Feature, Previews Muse Video

Meta has launched Muse Image, its latest AI image generation model, and previewed Muse Video, a video generation model developed by Meta Superintelligence Labs. 

Muse Image is available on the Meta AI app, Meta AI website, Instagram Stories in the US, and WhatsApp in select countries, while Muse Video will be rolled out to creators and Meta AI at a later date.

Unlike conventional image generation models that map prompts directly to outputs, Meta said Muse Image operates as an AI agent capable of invoking tools such as web search and code execution before producing an image. 

According to the company, the model can refine its own outputs during inference and improve image quality by increasing test-time compute. It uses web search to retrieve factual and real-time information for prompts involving current events or knowledge-intensive topics.

“We’re excited to launch Muse Image and preview Muse Video, the first media-generation models developed by Meta Superintelligence Labs,” the company said in a statement. “Instead of directly mapping prompts to images, Muse Image operates as an agent: it invokes search and coding tools to improve accuracy, self-refines its own generations, and improves through scaling test-time compute.”

Muse Image can generate QR codes, plots, GIFs, websites with embedded images, and interactive visual experiences by writing and executing code during inference. 

The company said Muse Image integrates with its multimodel AI Muse Spark, allowing both models to share tools and jointly plan media generation tasks. Meta added that the image model supports iterative editing, multi-reference image composition, and precise modifications while maintaining consistency across multiple edits.

Meta also introduced Content Seal, an invisible watermarking system that embeds provenance information into AI-generated images. According to the company, the watermark survives cropping, compression, resizing, and screenshots. Meta said it plans to extend the technology to AI-generated videos and is previewing a detection tool that allows users to verify whether an image contains a Content Seal watermark.

On benchmarks, Meta said Muse Image ranks second on Arena’s human preference Elo leaderboard behind OpenAI’s GPT-Image 2 for text-to-image generation, single-image editing, and multi-image editing.

Muse Video ranks third on the text-to-video leaderboard, behind Gemini Ombi Flash and Seedance 2.0.

Alongside Muse Image, Meta previewed Muse Video, which is built on the same pretraining foundation and supports native audio generation. The model focuses on prompt adherence, visual fidelity, and temporal consistency, as the Big Tech company continues to improve its audio-video synchronisation and fast-motion accuracy.

“Muse Image connects deeply with the Meta ecosystem,” the company said. “Our ongoing investments in image and video generation will further enable creators and businesses to generate dynamic content across Meta products.”