The new standard.
In large-scale double-blind evals Luma Photon outperforms every model on the market in quality, creativity and understanding while being radically more efficient.
Charts below show the overall weighted preference scores from pairwise preference evaluations over a set of diverse prompts.
For example, 60.1% means images from Luma Photon is preferred 60.1% of times over other models.
To make this new generation of models possible, we have designed a new set of evals and metrics that are purpose-fit for creative use cases. Luma Photon models were evaluated on this set of evals. We plan to share more on this methodology in the future.
Upto 10x more efficient.
The Luma Photon family is a step function change in efficiency brought about by significant architectural innovations.
The Photon models are radically faster (and cheaper) at generating ultra high-quality images compared to similar models and services.
Luma Photon - 1.5 cent / 2MP 1080p image
Photon Flash - 0.4 cent / 2MP 1080p image
This marks the beginning of visual abundance in our world — every idea can be iterated on 100 times over for less than a dollar to arrive at that extraordinary creation. Luma Dream Machine service is built on this principle of visual abundance and is made possible by this new Photon family of models.
Built for those who build our world.
The Photon family of models have been designed by our creative, design, and research teams to achieve unique aesthetics that mark the end of the “AI look”.
These specially designed models let designers, movie makers, architects and visual thinkers explore vast idea spaces and achieve extraordinary things.
Flux Pro 1.1 and Stable Diffusion 3.5 Large results are not included as the model/API returned too many errors for detailed long prompts.
Film
View model preference chart
Design
View model preference chart
Art Styles
View model preference chart
Fashion
View model preference chart
The new benchmark in visual intelligence.
The Photon family establishes a new state of the art in understanding natural language instructions and, for the first time ever, brings a large context window to visual generative models.
This intelligence now makes it possible to build multi-turn and iterative workflows for ideation, and editing.
Prompt adherence
Iteration and memory
A powerful new image reference system.
Applications built with Luma Photon will be able to let users to express intent by prompting with multiple images.
This makes it possible for users to bring their existing works, inspirations into new generations without fine-tuning or tediously reproducing them through prompts.
Consistent characters from a single input image.
Luma Photon has the unique capability of creating consistent characters from just one input image (currently in beta) and placing them in any scene with simple instructions.
With this ability, it’s now possible to create stories and campaigns with the same character appearing across generations.
Available now for everyone.
Luma Photon and Luma Photon Flash models are available starting today in Luma API. For more information and documentation, see here.
You can also try Luma Photon in the new Dream Machine service, the faster Flash variant will be available there soon.
Comfy UI ready.
Use Comfy UI workflow powered by Luma Photon API. Access Text to Image, Character Reference, Style Reference, Image Reference and Modify.