Meta Launches Muse Image & Muse Video To Rival Google's Nano Banana 2, Grok, & ChatGPT Images: All Details

In a major new development, Meta has launched Muse Image and has also previewed Muse Video, marking the first media generation models developed by its mega venture - the Meta Superintelligence Labs. Muse Image is described as the company's most advanced image generation model yet, capable of following instructions faithfully, editing with precision, composing from multiple reference images, and drawing on Instagram for social context. Muse Video is built on the same pretraining base and delivers visual fidelity with native audio support.

How does Muse Image work?

Instead of directly mapping prompts to images, Muse Image operates as an agent that invokes search and coding tools to improve accuracy, self-refines its own generations, and improves through scaling test-time compute. On the coding front, the model learns to write and execute code during reinforcement learning to produce accurate plots and QR codes, and it conditions on rendered figures to improve the accuracy of its generated images. Muse Image also integrates with Muse Spark, allowing the two models to combine code and media generation to create animated GIFs, websites with embedded images, and interactive visual games.

On search, Muse Image can search the web to ground its generated images in factual and real-time information and visual references, which Meta says improves factual accuracy on knowledge-intensive prompts, particularly those involving current events and real-world facts. The model also has a self-refining behaviour built into its chain of thought, which can take the form of a local edit when a small detail is off, a fresh generation from scratch when larger parts are wrong, or a switch to tool use for more factually accurate output. Meta says this behaviour was not explicitly designed but emerged during reinforcement learning training because it produced better images and higher reward.

Similar to language models, Muse Image improves the more compute it uses at inference time, reasoning more, using more tool calls, and applying more self-refinement steps as compute increases. Meta found that spending compute on deliberate reasoning scales considerably better than simply generating several images and picking the best one, an approach known as Best-of-N, which improves quality early on but saturates quickly.

Who can access Muse Image and Muse Video, and is it free?

Muse Image is available now across the Meta AI app, on meta.ai, on Instagram Stories in the US, and on WhatsApp in limited countries, and is coming soon to Facebook. Muse Video, on the other hand, is coming soon to creators and to Meta AI and has not yet been released for general use. Meta's announcement does not mention any subscription fee or paywall for accessing Muse Image, and it is being rolled out through Meta AI, which is generally free to use, though the company has not detailed whether any usage limits or premium tiers apply.

Muse Image and Muse Video: Key features

Muse Image supports multi-reference image composition, meaning it can combine elements from multiple input reference images in a single prompt, including people, objects, clothing, styles, and environments, and it supports interleaving text and images inline in prompts for complex compositions. The model also maintains coherence across editing turns, allowing for iterative refinement of an image over multiple prompts.

Meta says it is continuing to invest in areas where Muse Video currently has performance gaps, such as audio-video synchronisation and physically accurate fast motion. To help address concerns around AI-generated content, Muse Image includes a feature called Content Seal, an invisible watermarking system. Images created using Muse Image in the Meta AI app and on meta.ai carry a hidden provenance signal that Meta says remains intact even when the image is cropped, compressed, resized, or screenshotted. The company plans to extend Content Seal to video as well, and is previewing a separate detection tool that lets users check whether an image carries a Content Seal watermark.