Krea opens LoRAs for Krea 2, its most precise fine-tuning system

Krea 2 now supports LoRA training in beta, enabling Max and Business subscribers to fine-tune Krea 2 Medium and Large models with high-precision datasets.

Krea now offers LoRA training on its Krea 2 model, in beta. The LoRA technique, which involves specializing an image model using a small dataset, here allows for reproducing a given style, object, or character with increased fidelity. The platform already offered this feature on other models like Flux, Qwen, or Wan, but claims a higher level of quality and precision on Krea 2 than what previously existed in its ecosystem.

Regarding usage, training is accessed via the site's dedicated page, by creating a new LoRA and then importing a set of images into the tool. Two variants of the model are available for selection, Krea 2 Medium and Krea 2 Large, and a training step setting allows for adjusting the intensity of the specialization.

Access is currently reserved for Max and Business subscribers during the testing phase.