The new Phobos V6.0 is a specially trained model, which we are already familiar with from Chronos Dev, and in which we replaced 50% of the internal content of the underlying Krea 2 model on a one-to-one basis. We also compressed it so that the model is now 12 GB in size without any loss of quality.

With the help of the new Surgery technology, certain tensors have been restricted due to quality issues. This may cause problems with certain Loraks. To protect the model, a few special faces have been burned in; these can, of course, be modified for Lora's.

During the training process, we used a combination of special labels and natural-language captions, enabling the model to understand them both individually and collectively.

WARNING: THE TENSORART PLATFORM AUTOMATICALLY CONVERTS THE MODEL TO FP8, SO IT DOESN’T EVEN COME CLOSE TO THE ORIGINAL LOCAL QUALITY!