Black Forest Labs just released FLUX.2 Small Decoder: a faster, drop-in replacement for their standard decoder. ~1.4x faster, Lower peak VRAM - Compatible with all open FLUX.2 models

Black Forest Labs has launched the FLUX.2 Small Decoder, a significant performance upgrade for its popular text-to-image generation models. The new component serves as a direct, drop-in replacement for the standard decoder in the FLUX.2 architecture, offering approximately 1.4x faster inference speeds while simultaneously reducing peak VRAM (Video RAM) consumption. This efficiency gain is achieved through architectural optimizations rather than a reduction in model capabilities, making it an attractive upgrade path.

A key feature is its full compatibility with all existing open FLUX.2 models available on platforms like Hugging Face. Users can simply replace the decoder module without needing to retrain their models or adjust their generation pipelines. This plug-and-play nature lowers the barrier to adoption, allowing the community to benefit from faster image generation and lower hardware requirements immediately. The release continues Black Forest Labs' strategy of iterative, open improvements to the FLUX ecosystem.

The decoder's release follows the successful FLUX.1 and FLUX.2 model families, which have gained traction for their high-quality, stylistically diverse outputs. By focusing on decoder optimization, Black Forest Labs addresses a common bottleneck in diffusion-based image generation. This move could pressure other AI image model developers to prioritize similar component-level efficiency gains, potentially accelerating the overall pace of performance improvements in the open-source generative AI space.