Luma Photon is a text-to-image foundation model developed by Luma AI, designed to produce high-fidelity, photorealistic images for professional creative workflows. Built on a bespoke Universal Transformer architecture, it is engineered to generate artifact-free 1080p images with a focus on natural aesthetics, aiming to minimize the artificial appearance common in generative AI. The model family includes Photon Flash, a variant optimized for high-efficiency and rapid-turnaround generation.
A core feature of the model is its sophisticated image reference system, which enables precise style transfer and structural guidance from one or more input images. It includes dedicated character consistency capabilities, allowing users to maintain a specific identity across multiple generations by referencing a single character image. This is complemented by improved text rendering and high dynamic range (HDR) lighting.
The model demonstrates advanced natural language understanding, resulting in high prompt fidelity even with complex, descriptive instructions. It supports flexible aspect ratios and is intended for applications in film production, advertising, and architectural visualization. For maintaining character identity, users can utilize the consistency feature by using specific tags such as @character in their prompts when a reference image is provided.