https://civitai.com/models/357976?modelVersionId=433798 ver1:https://pixai.art/model/1726910189856515884 2.0版本感谢稀里哗啦帮忙写的提示词检索和预览图生成脚本，帮助很大。124位角色，更少的触发词The 124 version of Genshin Impact introduced updates to several characters and bound hair and eye options to characters. Now, it is possible to summon characters with fewer trigger words.原神全角色使用指南 Genshin Impact Comprehensive Character Usage Guide 2024.3 | Civitai124版本新增的角色比如：aether_(genshin_impact) 旅行者男·paimon_(genshin_impact) 派蒙signora_(genshin_impact) 女士madame_ping_(genshin_impact) 阿萍bonanus_(genshin_impact) 水夜叉kiana_kaslana_(herrscher_of_the_void) 代替 unknown_god_(genshin_impact)本次训练，在4090上耗费了超过50小时This training session took over 50 hours on 4090.124位角色（含服装），以每个角色200张单人图为目标，最后一共收集到 22649 张图片。For the 124 characters (including costumes), the goal was to collect 200 individual images for each character, resulting in a total of 22649 images.请注意，虽然有着高达124个有效文件夹，但是并不是所有的原神角色都支持，部分角色因为素材过少可能无法触发。但是乐观的是，可以触发的角色或服装应该大于100位。Please note that although there are up to 124 valid folders, not all Genshin Impact characters are supported. Some characters may not be triggered due to limited resources. However, the optimistic aspect is that there should be over 100 characters or outfits that can be triggered.Folders with Insufficient Images (less than 200 images)- 2_zhiqiong_(genshin_impact) - Image Count: 10 - 2_eremite_floral_ring-dancer_(genshin_impact) - Image Count: 16 - 2_katheryne_(genshin_impact) - Image Count: 16 - 2_cicin_(genshin_impact) - Image Count: 18 - 2_kitsune_saiguu - Image Count: 28 - 2_mirror_maiden_(genshin_impact) - Image Count: 32 - 2_unknown_god_(genshin_impact) - Image Count: 32 - 1_muji-muji_daruma_(genshin_impact) - Image Count: 38 - 2_aloy_(horizon) - Image Count: 38 - 2_xingqiu_(aoi_no_okina)(genshin_impact) - Image Count: 42 - 2_kaeya(sailwind_shadow)(genshin_impact) - Image Count: 44 - 2_xingqiu(bamboo_rain)(genshin_impact) - Image Count: 44 - 1_dottore(genshin_impact) - Image Count: 46 - 2_madame_ping_(genshin_impact) - Image Count: 76 - 2_indarias_(genshin_impact) - Image Count: 84 - 1_dori_(genshin_impact) - Image Count: 104 - 1_guoba_(genshin_impact) - Image Count: 106 - 2_gaming_(genshin_impact) - Image Count: 116 - 2_bonanus_(genshin_impact) - Image Count: 128 - 2_guizhong_(genshin_impact) - Image Count: 140 - 2_kamisato_ayaka_(heytea) - Image Count: 142 - 2_azhdaha_(genshin_impact) - Image Count: 146 - 2_mika_(genshin_impact) - Image Count: 158 - 2_diluc_(red_dead_of_night)(genshin_impact) - Image Count: 168 - 2_lisa(a_sobriquet_under_shade)(genshin_impact) - Image Count: 170 - 1_baizhu(genshin_impact) - Image Count: 182 - 2_rukkhadevata_(genshin_impact) - Image Count: 196Training model: ponyInference model: AutismMix SDXL特别鸣谢：算力赞助 nieta.art / GPU Sponsor: NetaRecommended dataset collection and processing tools: https://github.com/deepghs/waifucPossibly the most user-friendly dataset processing UI currently in use： https://github.com/Eugeoter/sd-dataset-manager原神全角色，包含服装共107位。A total of 107 character folders were effectively trained in this session.感谢xianienie 制作的封面！本次训练，在4090上耗费了超过48小时，感谢 nieta.art 支持我的想法。如果反馈好，这个系列还会出2.0版本，或者考虑制作其他作品的全角色系列。SD1.5版本的全原神也在训练中，敬请期待！当使用pony模型直接推理时，还原度可以达到80%以上当使用类似T-ponynai3的特化或融合模型，还原度约为60%以上本次训练有效的角色文件夹共107个。使用全标训练，因此，角色的所有特征的还原，需要搭配服装TAG才能还原！107位角色（含服装），以每个角色200张单人图为目标，最后一共收集到18k张图片。我根据训练集标签的出现频率，把出现频率大于80%的标签整理出来，整理了一个角色触发词，仅供参考。link：https://civitai.com/articles/4584/pony-xl-genshin-impact-comprehensive-character-usage-guide-20243完整的角色服装触发词，可以参考其他作者的原神模型，我摆了。比如，kozue就训练了很多原神模型，大家去抄他的触发词吧Genshin Impact has a total of 107 characters, including costumes.Special thanks to xianienie for creating the cover!This training session took over 48 hours on 4090. Thanks to Neta for supporting my ideas.If the feedback is positive, there will be a 2.0 version of this series, or consider creating a comprehensive character series for other works.The full Genshin Impact dataset for SD 1.5 version is currently being trained. Stay tuned!When using the Pony model for direct inference, the fidelity can reach over 80%.When using specialized or fused models like T-ponynai3, the fidelity is approximately 60% or higher.A total of 107 character folders were effectively trained in this session.Since it was trained with full labels, the restoration of all character features requires the use of costume tags for accurate reconstruction.For the 107 characters (including costumes), the goal was to collect 200 individual images for each character, resulting in a total of 18k images.Based on the frequency of occurrence in the training set labels, I have compiled a list of commonly occurring tags as character trigger words, provided for reference only.link：https://civitai.com/articles/4584/pony-xl-genshin-impact-comprehensive-character-usage-guide-20243For a complete list of character costume trigger words, you can refer to other authors' Genshin Impact models. I have mentioned kozue as an example, who has trained many Genshin Impact models. Feel free to use their trigger words as a reference.Training model: PonyRecommend model: Incomplete anime pony /Pony Diffusion for Anime/T-ponynai3/AutismMix SDXLTraining resolution: 1024*1024Recommended output resolution: 1024*1280特别鸣谢：算力赞助 nieta.art / GPU Sponsor: NetaDiscord : https://discord.com/invite/CFpyHceF52Recommended dataset collection and processing tools: https://github.com/deepghs/waifucPossibly the most user-friendly dataset processing UI currently in use： https://github.com/Eugeoter/sd-dataset-manager