Version 2.0:
- Improved from the first lora, the sword is less deformed and sword holding pose improved. Hand and fingers still a big problem, using hand lora can make it slightly better
Recommend using hires fix or adetailer, or both for best result (adetailer: https://github.com/Bing-su/adetailer.git)
- SUGGESTED prompt for flying pose: swordism, swg flying, 1girl, solo, full body, flying, above clouds,
- Setting:
VAE: usually I still use the 840000 ema pruned
Weight: 0.8 - 1
Clip skip: 2
Sampler: eula a
CFG: 7
Steps: 30+ (I'm usually bring it up to 32)
Upscaler: 4x-Ultrasharp (with Upscale by: 2, Hires steps: 15 and Denoising strength: 0,25)
Version 1.0:
- There will be a bad result because I didn't know what I'd done with this lora at the beginning. Ironically, this lora has a sword in its name, and most of the time, adding a sword prompt will lead to deformed hands and a weird sword appearance.
- Be careful with frame distance prompt: upper body, close up, cowboy shot, from side will give good quality on character, while full body and from behind will generated bad face.
- This lora can create NSFW content, and if you mind about NSFW, add "nsfw, nude" to the negative prompt just in case.
Weight: 0.8-1
Aspect: best with 512x512 or 512x768
Sampler: Eula a
Steps: 30
CFG: 7
Clip Skip: 1, or 2 if using with other lora
Description
- Improved from the first lora, the sword is less deformed and sword holding pose improved. Hand and finger still a big problem, using hand lora can make it slightly better
Recommend using hires or adetailer, or both for best result (adetailer: https://github.com/Bing-su/adetailer.git)
- SUGGESTED prompt for flying pose: swordism, swg flying, 1girl, solo, full body, flying, above clouds,
- Setting:
VAE: usually I still use the 840000 ema pruned
Weight: 0.8 - 1
Clip skip: 2
Sampler: eula a
CFG: 7
Steps: 30+ (I'm usually bring it up to 32)
Upscaler: 4x-Ultrasharp (with Upscale by: 2, Hires steps: 15 and Denoising strength: 0,25)
FAQ
Comments (4)
I'm not the best person to create images of Chinese culture, but I liked the images I created. Model works very well for the proposal it was created
Regarding holding weapons, it is common to have problems with 1.5 models, as the Stable Diffusion AI for 1.5 models does not behave very well with hands and feet, but I managed to generate some images holding a sword and I believe they turned out well.
Thank you so much for taking the effort to review my lora, the hands still a pain with 1.5 and even SDXL, not mention holding pose. I had trouble with the hand and sword deforming with the 1.0 version, then I doubled the data training for the 2.0 version and was able to get some decent images, but most of the time the fingers still getting error.
@Darknoice Civitai is a website for the community to help each other, unfortunately there are some trolls who try to denigrate other people's content, but don't give up because of these types of people.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.