My rating vs expectation: 90%

(Very high quality dataset. Angles of the photos could be improved, which led to less than great likeness. Or perhaps it's still under-trained even at 472 epochs. Bummer it didn't turn out perfect. Also it's highly baked, so not very flexible with simple prompts.)

All TIs I make depict real adults. Please apply it with good judgement. If you're the model and want this removed reach out to me via link below. I welcome all criticisms and suggestions with curiosity. If you like my work, buy me a house:
https://ko-fi.com/futurafree

Experimenting with high step count, larger dataset and low learning rate and I'm having a bit more success with it. This TI was trained in sd1.5 base with 5 vectors, 52 images, 24525 steps, and 472 epochs. This was finetuned for epiCRealism evo3 and Realistic Vision V4.0, but may work on others.

TIPS:

• MUST be the first word in the prompt for likeness and consistency.

  • Example: futurahannah24525v5, portrait of a woman, long dress etc.

  • Not: Portrait of futurahannah24525v5, etc.

• Recommended strength:

  • (0.98-1.0)

• I use adetailer for all images, with mediapipe_face_full, blank pos and neg, 1024x1024, inpaint mask blur: 20, inpaint noise strength: 0.45.

• Again, prompt in adetailer is left blank most of the time; (*exception*: if Loras/Lycoris is used and has significant impact on the appearance of the model's face, copy paste pos prompt without Loras/Lycoris into adetailer).

• Upscaler I use: 4x_NMKD-Siax_200k.