FLUX SRPO T2I

FLUX SRPO is a 12 billion parameter flow transformer fine tuned from FLUX.1 dev using Semantic Relative Preference Optimization, a technique that embeds reward feedback directly into the diffusion trajectory. It generates high quality, photorealistic images from text prompts and is tuned to reduce the plastic or overly synthetic look that can plague earlier models while improving first pass hit rates on complex prompts. SRPO delivers stronger aesthetics, natural skin and fabric rendering, and tighter alignment with human preference judgments. The model supports image sizes from 512 by 512 up to 1024 by 1024, with configurable inference steps around 28, guidance scale around 4.5, and optional safety checks. It is available via commercial APIs for both personal and production workloads.

September 24, 2025

Black Forrest Labs

Text to Image (T2I)