Chitose Fujinomiya is a character from a 2024 video game, Like A Dragon: Infinite Wealth.V1For this version, I did not use any captions for the dataset, following recommendation from CivitAI. I used default training parameters except for the Epoch, which was increased to 35 in order to reach training steps of almost 2000.To be honest, I'm not really satisfied with the result as it often produces cartoonified images, probably because Flux interpreted the sample images as cartoon instead of game screenshots due to having no captions. Also, it might be slightly difficult to reproduce some looks or outfits as they were not captioned during training.I might do V2 to use captions sometime in the future, if my Buzz allows it.V2Same datasets, but now with captions. As predicted, though it may still not be perfect, the results got much better than V1.Usage Tips (V1)Trigger word: Chitose FujinomiyaStrength: 0.8 - 1.0Hair: Short hair, ponytailOutfit: Black short-sleeved crop top, black and white bikiniUsage Tips (V2)<ul><li>Include video game screenshot in the prompt to get better accuracy. Otherwise, the image might be more realistic but the outfit and the hair might be a bit inaccurate.</li><li>For her hair, use bun hairstyle, short hair or slicked-back hair (this one is sometimes inaccurate).</li><li>For her default outfit, use black short-sleeved crop top, black sweatpants and grey sneakers.</li><li>For bikini, simply use black bikini.</li><li>For her casino dress, use blue dress.</li><li>For her Yokohama outfit, use navy bomber jacket with black crop top underneath.</li><li>If her necklace looks bad, it is recommended to include necklace in the prompt (not always needed though).</li><li>Strength: 0.8 - 1.0</li></ul>

Chitose Fujinomiya [FLUX] [Yakuza/Like A Dragon]

V1For this version, I did not use any captions for the dataset, following recommendation from CivitAI. I used default training parameters except for the Epoch, which was increased to 35 in order to reach training steps of almost 2000.To be honest, I'm not really satisfied with the result as it often produces cartoonified images, probably because Flux interpreted the sample images as cartoon instead of game screenshots due to having no captions. Also, it might be slightly difficult to reproduce some looks or outfits as they were not captioned during training.I might do V2 to use captions sometime in the future, if my Buzz allows it.

Files with SHA256: d641b96eb3d9c6dfb1d728a842df702304eb5949e4cb6353248c61a937326539