# Meta releases Muse Spark 1.2

Meta's new evals and demos show Muse Spark 1.2 turning visual inputs into code, guiding robots, and adding audio-visual understanding.

- Adds audio-visual understanding for video-heavy enterprise workflows.
- Generates web pages and other artifacts from images or video, refining outputs through a continuous self-improvement loop.
- A specialized variant acts as a robot brain, planning sub-tasks such as tidying a desk and distinguishing brushes.
- Internally used for media generation, producing caption training data for Muse Image and Muse Video.

## Why it matters

Developers and enterprises building multimodal workflows now have a model that can drive tools, generate deployable code from visuals, and control robots.

## Sources

- [Meta AI: Meta releases Muse Spark 1.2](https://x.com/AIatMeta/status/2090485743034716420)

---

Summarized by dstilled on 2026-08-20. https://dstilled.ai/story/e5e8f528-2323-4e31-a988-191aaa88ba7b