Style Saver

Style for artist saver (artbysaver) for IllustriousXL v0.1.

Continuing the quest to make AI create underwater content.

Keyword: "saver \(artbysaver\)" is not always needed. Especially if you want weaker style but still benefit from the learned underwater stuff.

Use "artist name" in negatives to avoid signatures.

The model often outputs messed colors. It is a surprise that it can do colors at all considering how few colored images were used in training.

You can try nudging the style with tags "monochrome", "greyscale", "sepia" and "lineart" but results aren't always great.

The model learned the lineart style quite poorly. It fails to make the background white resulting in mostly grey image. The lineart is still there though so just converting grey->white in editing program fixes this. Most likely problem with base model not handling it well.

The usual quality tags like "worst quality" and "bad quality" may work poorly due to too different style.

Training details:

Trained on 664 images of which 337 were greyscale, 43 were sepia, 249 were lineart, and 35 were color. 6 epochs totaling 1992 steps.