fast-DreamBooth Colab not creating good results compared to Runpod

[afrofail]

The google collab is more convent and has been easier to use, however the runpod tutorial I followed by Aitrepreneur has drastically different results that are working way better for what I'm trying to generate. I trained a character in both and the fast-DreamBooth collab results just the character with no other prompts. The runpod blends the characters face with prompt and style flawlessly. Is it because it's a 1.4 model?

I'm doing 20 images, 500 step increments, at 1000 steps it's already not implementing other prompt keywords unless I turn down the cfg scale to 1-3, but as soon as I hit 1500 steps, it's hard to get any keywords in unless I'm doing image2image. Am I doing something wrong or should I convert back to 1.4 model?

0.00001, 50 train text encoder.

[TheLastBen]

set the text encoder to 10% and do 6000 steps

[afrofail]

I'll give it a shot and see if this helps. Thanks for the input and keeping tabs with issues all the time, I know it's really hard to keep up with everyone and there personal issues.

[0xdevalias]

See also:

• https://github.com/TheLastBen/fast-stable-diffusion/issues/80
• https://github.com/TheLastBen/fast-stable-diffusion/pull/150
• https://github.com/TheLastBen/fast-stable-diffusion/issues/448
• https://github.com/TheLastBen/fast-stable-diffusion/issues/518