luma/photon models

photon by Luma — AI Model Family

The photon family represents Luma Labs' breakthrough approach to text-to-image generation, delivering photorealistic imagery at unprecedented speed and cost efficiency. Built on a revolutionary Universal Transformer architecture, photon solves a critical challenge in AI image creation: producing studio-quality visuals without the computational overhead and expense that traditionally plague high-resolution generation. This family is purpose-built for professionals, content creators, and enterprises who demand both speed and visual fidelity in their creative workflows.

The photon family currently includes two primary variants optimized for different use cases: Photon (premium quality) and Photon Flash (speed-optimized). Both share the same core architecture and advanced capabilities while offering distinct trade-offs between quality, generation speed, and cost.

photon Capabilities and Use Cases

Photon generates ultra-high-quality images at 1920×1080 resolution (1080p) with exceptional prompt comprehension and character consistency. The model excels at understanding complex, multi-element descriptions—processing relationships between objects, lighting conditions, and spatial arrangements with semantic depth that traditional diffusion models struggle to achieve. For example, a prompt like "A professional product shot of a minimalist ceramic vase on a marble surface with warm golden hour lighting, shallow depth of field, and soft shadows" produces photorealistic output with accurate material properties, natural lighting interactions, and clean edges.

Photon Flash trades marginal quality for dramatically faster generation and lower costs, making it ideal for rapid iteration, bulk content creation, and ideation phases. At just $0.004 per image compared to $0.015 for standard Photon, Flash enables cost-effective exploration of multiple creative directions.

Both variants support multi-image reference inputs, allowing users to upload character references, style references, or compositional guides to maintain consistency across generations. This is particularly valuable for brand campaigns, character design, or maintaining visual coherence across large content libraries. The models also support natural language iterative editing—after generating an initial image, users can refine specific aspects through conversational instructions like "make the lighting warmer" or "change the background to an urban setting" without regenerating from scratch.

The architecture produces minimal artifacts, with particular strength in rendering realistic human anatomy (including hands and facial features), accurate material properties (metallics, fabrics, glass, leather), natural lighting and shadows, and detailed textures with proper depth. This represents a significant leap over earlier-generation models that struggled with distorted hands, inconsistent lighting, and unnatural skin textures.

What Makes photon Stand Out

The Universal Transformer architecture is the technical foundation that sets photon apart. Unlike traditional diffusion models requiring extensive compute resources, this design enables photon to generate high-resolution images up to 10 times more cost-efficiently than competing alternatives—while generating images 800% faster than some competitors.

The large context window for visual generative models is a distinctive innovation, allowing simultaneous processing of multiple reference images and complex instructions. This capability enables granular creative control that few competitors match.

Character consistency across generations maintains facial features, body proportions, and distinctive characteristics with remarkable accuracy, solving a persistent pain point in AI image generation. Combined with multi-image references and iterative editing, this creates a professional-grade workflow for character-driven projects.

The model demonstrates exceptional prompt understanding, reducing the need for extensive prompt engineering. Users can describe scenes naturally and receive outputs that match their intent with minimal refinement cycles.

photon is ideal for marketing teams creating product visuals, creative agencies producing campaign assets, e-commerce platforms generating lifestyle imagery, game developers prototyping visual concepts, and any professional requiring production-ready imagery at scale.

Access photon Models via each::labs API

All photon variants are accessible through each::labs, the unified platform for AI model access. Rather than managing separate accounts across multiple providers, you can integrate photon directly into your workflows through a single, consistent API.

The each::labs platform provides Playground for interactive experimentation and SDK support for seamless integration into applications and pipelines. Whether you're building batch processing workflows, connecting image generation to databases for personalized content at scale, or triggering generation from external applications via webhooks, each::labs provides the infrastructure to deploy photon efficiently.

Sign up to explore the full photon model family on each::labs and unlock professional-grade image generation with the speed and quality your projects demand.