Luma Uni-1

• 🆕 Luma's unified image model combining understanding and generation
• 🧠 Multimodal system that reasons over a prompt before generating
• 🔧 Prompt-based editing rather than mask-based inpainting or outpainting
• 👁️ Single architecture handles both image generation and in-place edits
• 📚 Character references and seeds enable consistent image sequences
• 🌐 Available through the Luma API
• 🎯 Edit by describing what to change and what to preserve

Luma AI

Luma AI is a research company focused on generative visual intelligence, building models that produce high-quality imagery and creative media. Its work centers on pushing the frontier of how machines understand and synthesize visual content, serving creators,…

Luma Uni-1 is Luma's unified image model, which the company positions as a multimodal reasoning system that combines image understanding and generation within a single architecture. Luma describes it as reasoning over a prompt before producing an image, rather than treating generation and editing as separate pipelines. This entry covers the editing capability, accessed through the same model that also handles text-to-image creation.

A notable design choice is that Uni-1's editing is prompt-based rather than mask-based: instead of painting a region for inpainting or extending a canvas for outpainting, you give clear instructions about what to change and what to preserve. Using character references and seeds, it can also generate coherent image sequences that maintain consistency across scenes, such as showing a subject from different angles.

Uni-1 is reachable through the Luma API. Luma has continued iterating on the unified-intelligence approach within the same family, including a later Uni-1.1 release that the company introduced as offering more directable intelligence and refined aesthetic control over the original Uni-1. Because this family is built around a single model that both interprets and creates imagery, editing and generation share the same underlying capabilities rather than relying on separate specialized tools.