I don't know what you mean by embed.
Posted under General
Maybe we just use a general tag for the style if its an artist or character embed. Not sure about using an artist tag like post above said tho.
Besides that we will keep the embed in the prompt data for now.
i think having the tag "embedded" is fine.
It's different from prompts/ img2img because it's either a .bin or .pt, trained by the A, and could simultaneously be the same tag for trained hypernetworks.
There's an Asanagi style tag on Danbooru. Would it make sense to use that for this? https://danbooru.donmai.us/wiki_pages/asanagi_(style)
I actually came here wanting to ask the same question for Great Mosu, who doesn't have a style tag on Danbooru currently, and there are several generations being made that look arguably identical to work they would make. See these links for reference:
Not all these are embeddings either, so a tag like "embedded" wouldn't apply to everything replicating an artist's style. Textual inversion (embeddings), hypernetworks, and Dreambooth are several of the popular training/finetuning methods available currently. There are dozens of each of these being shared around, which will only continue to grow, so using a general style tag doesn't seem completely viable long term to me either, but as a temporary solution maybe it's fine for now?
I think mentioning the embed/hyper/dreambooth model in the metadata is the smarter option.
For a case like this, as more training methods are invented, it would be a mess to maintain several different style tags for one artist, who also potentially have been trained by multiple separate users (which has already been a frequent occurrence), and with newer versions potentially being made. Also I imagine most people searching with tags are interested in just the style, not the training method or very specific hash. If they find an image they like then they can check the metadata for more info on it.
Only additional thought is this might be an annoyance for some people if they're searching has:prompt to find the prompt only having an embed name and maybe a link, and not containing the actual prompt. Maybe there could be an additional field in the AI metadata? There's a lot of factors that can affect generation currently not listed in it that are also not generally listed in the PNG metadata (upscale latent space, sigma parameters) so that could be potentially useful to have, could end up being filled with junk data though. I'm personally open to anything as long as it doesn't clutter and make the tags/metatags less useful.
Updating the wiki pages I'm fine with, seems useful to link stuff there. I can probably do that for some of the existing artist style tags soon.