Flux Schnell

Flux Schnell is the distilled, speed-focused variant within the Flux family of text-to-image models from Black Forest Labs. Built as a 12-billion-parameter rectified flow transformer, it takes a natural language prompt and produces images directly, trading some of the configurability of larger diffusion systems for very low step counts. The underlying architecture was trained using latent adversarial diffusion distillation, a technique that compresses the iterative denoising process into a small number of sampling steps while preserving visual fidelity and prompt adherence.

The practical appeal of Flux Schnell is its balance of quality and throughput. In typical hosted deployments it accepts a text prompt along with controls for aspect ratio, the number of outputs per request, output format and quality, and a step count ranging from one to four, with four steps recommended as the sweet spot for image quality. Because of the distillation it is well suited to interactive creative tools, rapid prototyping, and high-volume asset generation where latency matters more than maximum detail, while still aiming for output quality competitive with closed-source image generators.