Trained on 97 illustrations from Nouveau dictionnaire encyclopédique universel illustré, A.K.A. the Trousset encyclopedia. Included were landscapes, cityscapes, and a handful of other scenes illustrated in the same style.FLUX.1 [dev] has decent performance with “woodcut illustration” already, but sometimes it uses mid-tones instead of proper black &amp; white shading.This LoRA is kinda scuffed but demonstrates some progress in influencing that style.Its effect may be more visible on something like <a target="_blank" rel="ugc" href="https://huggingface.co/wikeeyang/Flux.1-Dedistilled-Mix-Tuned-fp8">Dedistilled-Mix</a> or <a target="_blank" rel="ugc" href="https://huggingface.co/Anibaaal/Flux-Fusion-V2-4step-merge-gguf-nf4">Fusion</a> than stock FLUX.1 [dev]. It favors shading with fine lines over solid blocks of black and its 19th-century background may influence the architecture.Its performance at smaller resolutions can be much worse than normal for FLUX.1, probably because it was trained exclusively on 1+ megapixel images.Training captions began with the phrase “woodcut illustration.”Also <a target="_blank" rel="ugc" href="https://huggingface.co/keturn/woodcut-illustrations-Trousset-LoRA/blob/main/trousset-landscapes.epoch31.safetensors">available on 🤗Hugging Face</a> along with some more information about the training. Thanks go to <a target="_blank" rel="ugc" href="https://www.oldbookillustrations.com/">Old Book Illustrations</a> for collecting and labeling the images! (If you want to train on images like these, please use the <a target="_blank" rel="ugc" href="https://huggingface.co/datasets/gigant/oldbookillustrations">previously assembled dataset</a> instead of hammering the OBI server.)

Files with SHA256: a96949cf2bf05ec5144fabef10a864e0679e3071cd0f3fc1391e08fa1fbcb095