what am I doing wrong dreambooth training #1073
-
|
Hi I am trying to create a style based on vector graphics. But I don't know why the results are very poor. I don't know what I am doing wrong maybe someone can help I changed cfg, sample method, sampling steps and still the results are poor
settings 1 attempt to train: file names: omve
vector ilustration of Nissan Skyline r32 by omve settings 2 attempt to train: file names: omve1
vector ilustration of Nissan Skyline r32 by omve1
|
Beta Was this translation helpful? Give feedback.
Replies: 2 comments 1 reply
-
|
untick the box "style training" and train the unet for 1300 step, text enc for 350 and set the concept to 0 even though I think it captured the style very well, you just need to play with the prompt to get the right perspective |
Beta Was this translation helpful? Give feedback.
-
|
Now it is good :) but upscaling washes all image. What upscaler is good for such graphics ? |
Beta Was this translation helpful? Give feedback.
-
|
use img to img if you want a resolution below 1280x1280 |
Beta Was this translation helpful? Give feedback.
untick the box "style training" and train the unet for 1300 step, text enc for 350 and set the concept to 0
even though I think it captured the style very well, you just need to play with the prompt to get the right perspective