Luma Labs Launches Uni-1: The Autoregressive Transformer Model that Reasons through Intentions Before Generating Images

Luma Labs released Uni-1, luma Labs released a new foundational image model called Uni-1.

Uni-1 uses decoder-only autoregressive transformer architecture, uni-1 uses a decoder-only autoregressive transformer architecture that treats text and images as interleaved tokens.

Predicts next token in interleaved text-image sequences images are quantized into discrete visual tokens and the model predicts next tokens whether words or visual elements.

Performs a reasoning phase prior to pixel generation the model executes a reasoning phase to plan spatial layout before generating high-resolution details.

Led human preference rankings against Flux Max and Gemini on reported benchmarks, luma Labs evaluated Uni-1 on RISEBench and ODinW-13 and reported leading human preference rankings versus Flux Max and Gemini.