Flux

#6

black-forest-labs · seit 2024-08-01 · 16× · zuletzt 01. Juli 2026

50

Momentum

FLUX is a text-to-image model family by Black Forest Labs (Freiburg, Germany), founded by former Stability AI employees. FLUX.1 (August 2024) is built on a 12-billion-parameter rectified flow transformer; FLUX.2 (November 2025) extends this to a hybrid architecture combining a Mistral-3 vision-language model (24B) with a rectified flow transformer (32B total). The model family is released in variants with different licenses — from Apache-2.0 open-source (Schnell/Klein) to non-commercial open-weights (Dev) and proprietary API-only models (Pro/Max). Current models (FLUX.2 / FLUX.1 Kontext) support not only text-to-image generation but also multi-reference image editing and in-context image editing.

Preuves (16)