Meta unveils Muse media generation models, and signals shift to proprietary AI
Meta Unveils Muse Media Generation Models, Signals Strategic Shift Toward Proprietary AI
Meta unveils Muse media generation models - Meta’s recent focus on advancing its Superintelligence Labs has culminated in a public unveiling. The company introduced two media generation models, Muse Image and Muse Video, currently in preview mode. These models are positioned to rival leading commercial and open-source creative AI platforms, such as those developed by OpenAI, Google, and Adobe.
Meta aims to deeply integrate these models into its social platforms. The rollout reflects a key milestone in Mark Zuckerberg’s broader AI vision, which has seen substantial financial backing over the past year. This development suggests a growing emphasis on proprietary AI solutions, moving away from earlier reliance on open-source initiatives like the Llama series.
Meta’s Muse model family includes the earlier-released Muse Spark large language model. The company plans to eventually phase out open-weight Llama models in favor of the Muse suite, marking a strategic pivot toward in-house AI development.
Muse Image Features and Capabilities
The Muse Image model leverages advanced reasoning to interpret complex prompts, enabling it to merge multiple photos into cohesive outputs. According to Meta, the tool provides Muse Image with access to enhancements that refine its autonomous functions. During reinforcement learning, the model learns to generate code for precise visual elements like plots and QR codes, using rendered figures to boost image accuracy.
“We provide Muse Image with access to tools to enhance its agentic capabilities. During reinforcement learning, Muse Image learns to write and execute code that produces accurate plots and QR codes, and condition on rendered figures to improve the accuracy of generated images,” the company says.
Additionally, Muse Image can search the web to ground its creations with real-time data and visual examples. This self-refining behavior, which Meta claims emerged naturally during training, leads to improved results by prioritizing outputs that yield higher quality.
The model is capable of editing images, including restoring details, removing blur, altering specific elements, and combining components from multiple reference images. It also allows for text-based image modifications, expanding its creative potential.
Performance Benchmarks and Comparisons
In benchmark tests, Muse Image performs competitively against OpenAI’s GPT Image 2, though it scores slightly lower. It outperforms Google’s Nano Banana 2 and xAI’s Grok Imagine Quality, according to recent evaluations.
Muse Video, which is set for a later release, promises strong performance in prompt alignment, visual precision, and temporal consistency. Meta highlights ongoing efforts to address current limitations, such as improving audio-video synchronization and achieving physically accurate motion in video generation.
“We’re investing in areas with current performance gaps, such as audio-video synchronisation and physically accurate fast motion,” Meta says.
Latest assessments show Muse Video surpassing OpenAI’s Sora 2 Pro and Google’s Veo in the text-to-video benchmark. However, Google’s Gemini Omni Flash and Bytedance’s Seedance 2.0 models currently lead in specific metrics.
Integration and Availability
Meta has already begun deploying Muse Image within its ecosystem. The tool is accessible through the Meta AI app and on the web, with Instagram Stories in the US now featuring its capabilities. Integration within WhatsApp is also underway, though in India, the Meta AI suite has not yet incorporated these features.
“You can use more than 30 new AI-powered effects for Instagram Stories and generate images in your direct chats with Meta AI on WhatsApp — starting in limited countries with more locations on the way,” they say, in a statement.
Future plans include expanding Muse Image and Muse Video to Facebook and Messenger, with Muse Video targeting creators and the Meta AI suite in the coming weeks. This expansion underscores Meta’s commitment to embedding AI into its core services.