Flux.1-Dev-Asia

This is an experimental fusion model that incorporates my retrained Asian Lora.

The model is compatible with other Flux.1 Dev Loras, but you may need to adjust the Lora weights accordingly.

The FP8 version has undergone multiple adjustments, maintaining the diversity of Asian faces in the model while also making some aesthetic modifications. After around 1000 rounds of testing, the overall effect is quite good. However, some issues have been identified.

Despite my efforts to minimize the impact on the weights of other parts of the model, the overall brightness of the model has increased slightly. This has resulted in some differences in composition actions compared to the original model. Although the overall performance is still good, it is very regrettable that there has been a slight degradation in limb quality as well. I hope to find a better solution in the next version.