How is this Lora model trained?

#1
by Songssx - opened

Hello, I tried this LoRA model and it works really well. Could you please tell me how this LoRA model was trained, for example how the dataset was prepared?

I know that tools like FluxTrainer or kohya_ss can directly train LoRA models for Flux Dev, but from what I understand, the datasets prepared in this way usually consist of image-text pairs. However, the resulting Kontext LoRA seems to mainly enhance text-to-image capabilities, while editing capabilities seem to require datasets consisting of triplets like , , and .

May I ask if your LoRA training was done using such triplet datasets? Also, did you train the LoRA directly through Python code, or did you use any additional training frameworks?

Sorry, I just saw that there’s a video link provided in the description. Thanks for sharing — I’ll check it out and learn how to train it.

Sign up or log in to comment