Trained with 40 images from nvl artist.

V1.A uses 25 images, V1.B uses 40 images, that's it. At first, I trained using 40 images and the result is not exactly what I wanted. So, I reduced it to 25 images to make the style variant more cohesive and the results are very good. But after generating using the same prompt, I realised the difference between A and B is negligible or they are different because of CFG range.

But, personally, I think:

B is slightly better for outdoor background than A,
A is slightly better for rim lighting and shiny clothes.
didn’t test, but B dataset has a bit more poses than A.

So, I decided to keep both.

Weight: 1.0-1.3. Either Euler or Euler a is fine too.

Tested using following model: (low steps & resolution for cheaper token cost to test result)

Best: WAI v17 - 10 steps, CFG 3.0-4.5

Good, but: VXP IL Base Model (NSFW alert!) - 12 steps, CFG depending on Euler (~5.5) or Euler a (~4.5)

WAI v15 gives inconsistent result

Any other better checkpoint or control setting of your preference.

Enjoy!