This is fantastic, thanks for putting this together!
I'm curious if you've played with the prompt and the language model used to significantly steer the attention or format of the captioning?
I've been using MNeMoNiCuZ/joy-caption-batch before finding this repo, and when I use mlabonne/Meta-Llama-3.1-8B-Instruct-abliterated then changing the PROMPT value seems to nudge the wording very slightly, but it still mentions all the same properties.
This is fantastic, thanks for putting this together!
I'm curious if you've played with the prompt and the language model used to significantly steer the attention or format of the captioning?
I've been using MNeMoNiCuZ/joy-caption-batch before finding this repo, and when I use mlabonne/Meta-Llama-3.1-8B-Instruct-abliterated then changing the PROMPT value seems to nudge the wording very slightly, but it still mentions all the same properties.