jp-Sveta928-1075 is the latest in my attempt to learn TI's.

• This is yet another Svet(a) embedding. My favorite RL model.

• 928 in the name refers to the date the images were taken. I wanted to see if I could do a simple on in a couple of hours, I think it turned out well.

• I made this embedding with 12 images. Took the reference pictures with my I-Phone against a tan wall, simple stuff.

• This gets pretty close, reference photos are nude so it will produce nude images on it's own.

• 12 images trained for 1075 steps.

• 12 vectors per token.

• I used the following progressive learning rate 0.05:10, 0.02:20, 0.01:60, 0.005:200, 0.002:500, 0.001:3000, 0.0005

• Batch size of 4 with 3 gradient accumulation steps.

As always I appreciate your feedback and look forward to seeing what you all create.