Recently, I embarked on my second attempt at FLUX DEV training with FluxGym. This time, I took a different approach that yielded significantly improved results.
My strategy involved retraining the model using a more diverse image set, surprisingly, with fewer images overall. This refined methodology proved to be a game-changer.
The outcome is a highly flexible set, particularly in its prompt applications. It’s truly exciting to discover the creative possibilities once you’ve successfully trained a LORA on a specific subject — the results can get really crazy!