Triggers: @mark bagley, western comics style, usm_style,Mark Bagley style, preliminary version, trained on a sample of cut-out frames from Ultimate Spider-Man (roughly the first 30 issues).This is really just a test version, which does a decent job of imitating the style, but due to poor tagging via the tagger, Jessica Drew has been labelled as Spider-Gwen, and the characters have no markup, so they can be copied using very random tags (a bit later I’ll upload a more refined version where I’ll try to mark up the characters and expand the dataset with other series illustrated by Mark Bagley, and I’ll also make a separate version for covers).The style isn’t particularly aggressive and can easily be overridden by the model’s style or additional LORA (though in my opinion, that’s actually a good thing).

Mark Bagley Style

It’s simply a data set that’s three times the size (volumes 1–3 of Ultimate Spider-Man plus around 100 panels from other comics, such as All-New X-Men or Alias/Superior Spider-Man).As it turned out, one artist alone isn’t enough and we still need to find the right colourists and inkers, so this version is a sort of ‘watered-down’ version and is something that the model could have produced out of the box if the authors had initially incorporated knowledge about the artist into the model.The visual style is slightly less sharp and closer to the modern issues drawn by Bagley, a bit more varied in terms of character proportions, and should cope quite well with rendering female behinds and men’s muscular chests. But the layout is still at the mercy of random, erratic taggers, so drawing a specific character is unlikely to work.The plan is to remove a bunch of rubbish tags and add the missing ones, as well as replace the ‘spider-man’ tag with ‘spidersuit’, so it’s easier to generate characters regardless of gender (because finding cool shots by Bagley without Spider-Man is quite the challenge), and to categorise ‘dialogue’, ‘action’ and ‘other’ frames so that the ‘Bagley style’ works properly

Files with SHA256: efc3f46b12830e01e9defd7a82b5355fa5da00e99b29df3aa6e7449c35669cf6