A LoRA trained on older books and pages (~100-150 years old), since SDXL tends to generate very oversatured and unrealistic pages otherwise. Dataset is tagged, mention those keywords to encourage e.g. a double sided book cover (tags overlap):<ul><li>old, book: 147</li><li>text, text page: 97</li><li>left side, left page: 48</li><li>right side, right page: 43</li><li>title, title page: 6</li><li>illustration, illustration page: 34</li><li>cover, book cover: 29</li><li>single-sided, one-sided: 6</li><li>scribble, writing, annotation: 3</li></ul>

Old Books and Paper

Initial train, 147 images, 68 of them unique, 15 epochs trained on default settings.

Files with SHA256: e224c192551d9163e727135521892e3ee3b116e65ad255de7d2c399c5ed2d351