This is my first time training a model and the quality product is far from ideal. Nevertheless, it does encourage the diffusion model to better understand what Shogi is, compared to not using a LoRA.

Other than the trigger words I listed, these words were also included in training but seems to be unsuccessful:
starting_board, playing_board, shogi_piece, shogi_board_on_wall

Techincal challenges for Shogi LoRA:

• 9x9 board not promised

• weak understanding on glyphs (kanji & kana)

• pieces arragement is random, and could generate unlikely results