0304/0305-03010 multi lora,controlnet ,视频测试
- issac zhang
- Mar 4, 2025
- 3 min read
Updated: Mar 10, 2025
3.3 最后的universal stylelora测试（不改了！） | 3.4 字体lora | 3.5角色lora | 3.6 道具lora | 3.7lora修改 | 3.8故事板 | 休息-徒步 |
3.10故事板确定（跟项目一样） | 3.11ccomfyui，多lora，controlnet测试 | 3.12 styleframe | 3.13styleframe | 3.14styleframe（只做一个镜头） | 3.15 栅格AE动画？？？或者其他字体动画？？ | 休息 |
AE字体动画？？ | AI人角色动画 | AI角色动画 | AI角色动画 | 激光打印 | 激光打印 | 激光打印（成品） |
用了这个工作流进行测试，发现效果可以。得继续测试前后帧。这个可以用于到时候武术动作和精准的角色设定控制。然后放到kling里面进行测试。
今天晚上继续进行Lora训练！
0305
prompt：traditional Chinese Gongbi painting style, inspired by ancient woodblock prints. A historical scene depicting two craftsmen operating an old mechanical device with large wooden gears and a pulley system. The animation features the two men actively turning the gears and pulling levers, their facial expressions subtly shifting with effort. The wooden wheels rotate rhythmically, and the entire contraption moves smoothly, mimicking ancient Chinese craftsmanship. The background has a soft, textured parchment effect, reinforcing the historical atmosphere. maintaining the intricate ink details and delicate brushwork of classic Chinese illustrations
这张图是我用最新的multi-lora工作流做出来的，放入到Wan之后直出的视频是这样的效果。可以看到这个效果非常好，动画效果比kling要强很多。估计kling1.5的训练集的问题，会导致出来的画面有点在3d上面贴材质的感觉，很不自然。但是wan输出的动画很自然，并且对于上传图片有做很好的继承。
0306-310
图片都剪裁于古籍，经过多次的测试之后。整个数据集的处理流程一共有这么几步：
裁剪-PS去掉水印/字体/多余人物 -手工提高线条精度（加强对比）-后期处理成单色（去除多余texture，只留线条） - 统一size（1024*1024/768*1024…） -caption
已经处理完了最新的角色lora，准备继续进行lora测试。
不打标版本效果
自动打标版本效果
带锅版 Fast trainingV1
从这个版本开始，经过手动加强线条和添加角色道具。我直接尝试去训练动画角色
带锅版 Fast trainingV2
带锅版 手打caption版
Figure 1 processed Chinese traditional Physiognomy image from Dataset
Figure 2 an image generated without any LoRA
Figure 3 a generated image using Character LoRA （strength 0.6）
Figure 4 a generated image mixed with character LoRA and universal concept LoRA
Figure 5 a generated image mixed with character LoRA and universal concept and universal style LoRA
Prompt： ABAAA style,ohwx man,BBCCC style.
A Chinese woodblock print of a ohwx man's face from the front view, The face is divided into sections, including the forehead, eyes, nose, mouth, and chin, each annotated with dense Chinese characters predicting aspects such as longevity, wealth, marriage, personality, and career.The ohwx man wearing a cooking pot as a hat, and the inscriptions appear to associate facial features with aspects of fate, personality, or fortune.
The illustration is simple yet detailed, with smooth linework and a beige background, enhancing the appearance of an ancient manuscript or historical text
Fixed Seeds： 207131067904716
Reflection： 综合测评下来，角色的LoRA手动的caption效果是最好的，而且使用角色Lora的时候需要同时加上风格颜色描述词才能出发这种效果。综合了以上三个的LORA之后能够达到我想要的中国版画视觉效果，实现了对古代木版画风格的良好继承。
貌似我的大lora应该要试试去训练一版concept...
Comments