jp-Fluer is the latest in my attempt to learn TI's.
The name is random... I'm up to F at this point and she is trained from stock images.
She is not a real person and does not reproduce any of the training images exactly.
14 images trained for 500 steps. 14 vectors per token. Learning rate off .005 with batch size of 7 with 2 gradient accumulation steps.
I'd appreciate any feedback you may have and would love to see what you come up with. Thanks and enjoy!
Description
FAQ
Comments (1)
This is great and looks really consistent. Nice job! Could you give some tips on training one of these? You mentioned using stock images and gave some info about training parameters, but can you give more detail on the process and which tools were used to conduct the training, what makes a good training set of images (different angles, lighting conditions, etc?), how those training images are tagged, etc. Just looking for latest info/tips on how to do this. I'd love to give it a shot.