Help request: embed training for redshift model

#24
by SamTyurenkov - opened

Hi there, thanks for the awesome model.

Im trying to train an embed on custom subject. I want to achieve photorealistic tiger tank, but so far it still comes out with some artefacts:

00008-375976945.png
00007-3012214692.png
00005-333016989.png
tmpz_4yfx6r.png

This last attempt was a training with based on single angle of tiger tank on different backgrounds. I used keywords describing background in filenames, selected a default subject filewords template, 30 vectors per token and "photorealistic tiger tank" init text. Trained for 5000 steps.

Most artefacts appear on the gun + it cant add the subject into background as cool as in prompts without embed.

Any advice is appreciated, im also willing to purchase a paid consultation if possible.

Thank you again!

Sign up or log in to comment