Flux:

Trained on FLUX1.dev with 237 images

Tested with ComfyUI and Forge

SDXL:

Trained on SDXL 1.0 vae fix with 281 images.

The result of another test training trying to reduce the file size without compromising (too much) on likeness/quality and without resizing the model afterwards...

And again, there is still a lot of room for improvement..

No VAE, hiRes fix or refiners were used for the example images...