English/中文
This LoRA model takes inspiration from transparent textured photos generated by X-ray radiography. It learns the unique texture and applies color processing to ultimately not only have the ability to directly generate X-ray radiographs but also generate colored photographs with the distinctive transparent texture of X-ray radiographs.
Instructions:
Transparent Photography: Generates photographs with a unique transparent texture. Adding "skeleton" can generate characters with transparent skin. To achieve better color effects, you can add the negative word "monochrome," but there is a chance it may generate an oil painting style.
X-ray Films: "skeleton" plus "black background" or "white background".
Trigger word: xc
Weight: 0.8-1
Example image generation parameters: Steps: 20, Sampler: DPM++ 2M SDE Karras, CFG scale: 7, Size: 512x768, Denoising strength: 0.5, Clip skip: 2, Hires upscale: 2, Hires upscaler: R-ESRGAN 4x+.
Recommended positive prompts: innocent, shadow, fantasy, dream, masterpiece, highres, best quality, highly detailed. I also encourage everyone to share their own images and explore various effects generated by different prompts! (Caution: X-ray images should be used with care, and some specific actions may not accurately generate results.)
Note:
This LoRA model is a mixture of multiple concepts, which inevitably leads to some elements being overfitted. When undesired elements appear, you can add corresponding elements in the negative terms to resolve them, for example: curly hair, navel, headband, sunglasses, text, etc.
Please use negative embeddings like "bad-hands" and "easynegative" with caution as they may affect the image generation results.
Due to limitations in the dataset, generating animal or fictional creature skeletons may be inaccurate and may result in the generation of animal heads with human bodies or human heads with animal bodies.(which maybe shouldn't be considered a drawback, haha)
This LoRA is trained based on SD1.5, so it can be applied to most checkpoints based on SD1.5. There may be slight differences in the results generated by different models (some 2D models are also effective). Among the three large models I tested, I recommend using "cyberrealistic_v31".
本LoRA以X线平片摄影产生的透明质感照片为灵感，学习其独有的特殊质感并加以色彩的加工，最终使其不仅具有直接生成X线平片的能力，同时也能够生成带有X线平片独特透明质感的彩色摄影图片。
使用方法：
透明感摄影：可产生独特透明质感的摄影图片；加skeleton可以生成肌肤透明的人物；若想要更好的实现彩色效果，可添加负面词monochrome，但有概率会生成油画风格
X光片：skeleton加black background 或 white background
必需触发词：xc
权重：0.8-1
例图生成参数：Steps: 20, Sampler: DPM++ 2M SDE Karras, CFG scale: 7, Size: 512x768, Denoising strength: 0.5, Clip skip: 2, Hires upscale: 2, Hires upscaler: R-ESRGAN 4x+,
推荐正面提示词：innocent,shadow,fantasy,dream,masterpiece,highres,best quality,highly detailed,也希望大家多多分享自己的图片、探索不同提示词产生的各式效果！(X光片图慎用，部分指定动作无法准确生成)
注意：
1.该LoRA由多种概念混合而成，不可避免的导致部分元素过拟合；当你不想要的元素出现时，可在负面词添加相应元素以解决，例如：curly hair,navel,headband,sunglasses,text等
2.请谨慎使用如bad-hands、easynegative等负面嵌入，它们可能会对图片生成效果产生影响
3.由于数据集限制，动物或虚拟生物骨骼生成会不准确，会有生成兽头人身的情况
4.该LoRA以SD1.5为底模训练，故可适用于大多数以SD1.5为基础的大模型，不同模型产生的效果会有少许差异(部分二次元模型也有效)。我测试的三个大模型中，推荐使用cyberrealistic_v31。