Re:Stable Diffusion Textual Inversion
by GaggiX on Monday August 29, 2022 @07:19PM
(#32644784)
Textual inversion also supports more than one embedding for identifier, just change num_vectors_per_token in the yaml config.
Example: https://www.reddit.com/r/StableDiffusion/comments/wzf1qk/sd_...
[ Parent
| on HN ]
Re:Stable Diffusion Textual Inversion
by lucidrains on Monday August 29, 2022 @07:47PM
(#32644977)
so is dreambooth worth open-sourcing then given textual inversion?
Re:Stable Diffusion Textual Inversion
by beyondarmonia on Monday August 29, 2022 @08:04PM
(#32645114)
From the textual inversion guy's own comment on Twitter
>The objective is similar, but it's: (1) A different approach - they also fine tune the model itself, and they get much much better identity preservation!
dreambooth retains higher fidelity as the model is finetuned, but to be honest I think textual inversion is actually more applicable as you can just add some embeddings to inject new knowledge into the model and not an whole new model just for a single concept (if you want to share it with others). Also I have not seen dreambooth being applied to replicate styles.