Welcome to version 2 of my SvetaF embedding. Why a new model versus changing the version? I accidentally chose depicts a real person on V1 and I'm waiting for it to be reviewed to become live again. That being said this is a new Textural Inversion and the setting are a little different on this one. I may retrain at some point as it just isn't where I want it.

The Sveta models are kind of my passion project and I know what I'm looking for. That being said let me reiterate this embedding does not create images of an actual person, rather a likeness.

• SvetaF-v2 is trained on 24 images.  I used the same 10 images from my jp-gingerv1-275 embedding and another 14 images of a different model to create this embedding. 

• She is not a real person, rather is created from a series of handpicked and manually captioned images that I took of the subject over about 15 years to create the outputs you see.

• 44 images trained for 5000 steps.

• 10 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 (This is different than the V1 step).

• Batch size of 1 with 1 gradient accumulation step.

• Looks like I forgot to take her earrings out in the captions, you will need to do that in the negative if it is an issue for you.

I'd appreciate any feedback you may have and would love to see what you come up with. Thanks and enjoy! (i.e. review and share images please)