Star Wars: Visions (2021~) - The Ninth Jedi
2026/08/27
素材のjpgノイズを軽減させて再学習しました。データセットは全部同じです。
IL0.1版とanima版はcivitaiのデフォルト設定でトレーニングしてあります。
IL1.0版はちょっと大人っぽく服が不安定です。
This model has been retrained after reducing JPG compression noise from the source images. The dataset remains exactly the same.
- IL 0.1 & Anima versions: Trained with Civitai's default settings.
- IL 1.0 version: Characters may appear slightly more mature, and clothing rendering can be a bit unstable.
画像の出力に「VXP_IL_XL」と「Pancake Mix」、「WAI-NSFW-illustrious-SDXL」、「NJSmix_tengaeshi」を使用しています。
The images were generated using "VXP_IL_XL", "Pancake Mix" ,"WAI-NSFW-illustrious-SDXL"and "NJSmix_tengaeshi"
Description
FAQ
Comments (4)
What's the difference between "IL1.0 new" and "IL0.1"?
IL 0.1 has better character accuracy than IL 1.0, but IL 1.0 offers much better human anatomy with fewer errors and more composition options. Since outfit details don't matter for nude renders, I usually stick with IL 1.0.
@daretoku Thank you for the response. One last question: If that is the case, then why not just release the better one?.
@HoityToity Thanks for asking! I'm happy to share the full story behind my workflow.
To answer your question directly: the version I published here IS actually the best result among all my tests!
Because my local PC isn't powerful enough to train LoRAs locally, I rely on web-based training services (Civitai, PixAI, and Tensor.art, though I’ve currently paused my Tensor subscription). Each service has its own pros and cons:
PixAI: Can train from IL0.1 to IL2.0, but lacks tagging features, so character accuracy can be hit-or-miss.
Tensor.art: Supports IL0.1 to IL2.0 and tagging, but the complex settings require a lot of fine-tuning to get right.
Civitai: Only supports IL0.1, but offers the highest stability and overall quality.
For every model I create, I test them across these platforms and release whichever version captures the character's appearance best. For this specific character, the Civitai-trained IL0.1 version gave the absolute best results, which is why I put it out front!
PixAI actually works great for everyday characters with simple outfits—for example, this Miyuki model came out really well on PixAI, maybe even better than expected! But for complex Sci-Fi or RPG designs, PixAI struggles without tagging, so Civitai’s stability makes a huge difference.
(As a side note, I’ve also been trying out Anima recently since it’s been getting a lot of attention, and I was really impressed by how accurately it captures character details!)
I hope this gives you a better picture of why I chose this specific version to release. Thanks again for your interest in my work!