FLUX.1 Krea [dev] dramatically improves the inference efficiency of models through an innovative guided distillation technique. This technique allows models to maintain generation quality while dramatically reducing computational resource footprint, enabling large models with 12 billion parameters to run on consumer GPUs. In a standard configuration, it takes only 10-20 seconds to generate a 1024×1024 image using an RTX 3090 graphics card. For devices with insufficient video memory, the model also supports CPU-assisted computation via the enable_model_cpu_offload() method, which moderately slows down processing (~1-2 minutes/image) but ensures proper operation on all types of hardware.
This answer comes from the articleFLUX.1 Krea: Free Open Source Tool for Generating Highly Realistic ImagesThe
English 
简体中文 
日本語 
Deutsch 
Português do Brasil