Some questions about the Lora training Process
First of all i would like to personally thank you for creating an amazing collective LORA models.
I'm fairly new to this LORA training and i'm stuck(not literally) on one of the process which is image captioning . I walkthrough a tutorial from Aitrepreneur and in his video he used BLIP captioning to do the job and modified it to get the accurate result.
The problem start here, in his tutorial, he used captions like " a woman wearing a black jacket and a white shirt etc...." So as for the Anime Characters or styles do i used this the one in the tutorial or use danbooru tags ?
That was the main question.
Can u provide a sample of captioning if possible?
Also i'm thinking about using
20 SFW Images
10 SFW+N Images (kinda like ecchi )
30 NSFW Images
Do you thinks it's over kill or still need more?
I've been using a hybrid of the two for captioning and have been finding it works well for me. though i have never tried only booru tags alone, i have seen noticeable improvement adding them to the end of a prompt over a prompt alone.
Its probably not the best way to do this but here is an example of what i normally do.
Emma Misteltein a girl with pink hair and blue eyes holding a sword in her hand and looking at the camera with a smile on her face, with a sword in her hand, with a white background, 1girl, blue_eyes, earrings,braid jewellery, long_hair, looking_at_viewer, pink_hair, portrait, simple_background, smile, solo
Ok that's an interesting take, i've tried tagging with both danbooru and also the traditional way of captioning but i;ve never combined both of them. Both are good but i prefer the danbooru since it work very well with anime models which i'm currently testing out. Thanks anyway.
