CivArchive
    wan2.2 live 2d风格动态壁纸 - v1.0
    NSFW

    基于wan2.2 i2v low训练而成,这款 LoRA 的目标是重现类似动态壁纸的视频风格

    Based on wan2.2 i2v low training, this LoRA aims to reproduce a video style similar to dynamic wallpaper

    参考提示词:live 2d,动态壁纸风格,人物轻微摆动,随呼吸起伏,有规律的眨眼,头发随风飘动,人物位置不变

    Reference words:live 2d,Dynamic wallpaper style, the character sways slightly, rises and falls with breathing, blinks regularly, hair flutters in the wind, and the character's position remains unchanged

    Description

    FAQ

    Comments (22)

    Cano250Sep 15, 2025
    CivitAI

    我要吃你的批!

    monsternova06880Sep 15, 2025

    6

    kurokuroSep 15, 2025· 2 reactions
    CivitAI

    Interesting! Is there no 'high' noise variant?

    wwyhahaha
    Author
    Sep 15, 2025· 3 reactions

    I believe the high-noise version is unnecessary, as most people prefer a dual-sampling workflow with both high and low noise. The intended effect can be achieved simply by using the low-noise LoRA.

    kurokuroSep 15, 2025· 2 reactions

    @wwyhahaha Interesting - I'll give it a shot with just the low LoRA. I thought the high noise LoRA was for 'structure', which seems important for a concept / aesthetic LoRA like this; and the low noise LoRA was for fine details.

    keforce1Sep 15, 2025
    CivitAI

    提示词怎么写

    wwyhahaha
    Author
    Sep 16, 2025· 2 reactions

    参考提示词:live 2d,动态壁纸风格,人物轻微摆动,随呼吸起伏,有规律的眨眼,头发随风飘动,人物位置不变

    bossman777Sep 16, 2025· 2 reactions
    CivitAI

    Does it need a trigger word?

    wwyhahaha
    Author
    Sep 16, 2025· 2 reactions

    The trigger word is 'live 2d'

    renanhuaSep 19, 2025
    CivitAI

    给推荐个流吧,试了三个效果都不是太好,大概20次里只成功了三次

    wwyhahaha
    Author
    Sep 20, 2025

    我的流也可烂了,我想不用加速,但是不用加速我真的带不动啊

    wwyhahaha
    Author
    Sep 20, 2025

    实例视频都是原视频,是可以读出来工作流的,你要想试试也行

    renanhuaSep 21, 2025

    找到方法了,用首尾帧加速流,添加这个lora就可以稳定头发飘动了

    1983507035fyc933Jan 14, 2026

    @renanhua 兄弟怎么解决的,能给个流吗?

    jiangniaodawangOct 21, 2025
    CivitAI

    没有高噪和低噪版本么大佬

    wwyhahaha
    Author
    Oct 21, 2025

    只用低噪声效果就不错了,高噪声不是很必要

    endo123Nov 23, 2025
    CivitAI

    I'm curious about the comfyui workflow

    uqwDec 12, 2025
    CivitAI

    Не вижу разницы после применения Лоры

    white_razorswhit2619Dec 13, 2025
    CivitAI

    不太懂为什么低噪就可以控制,不是应该高低噪合成控制吗,怎么低噪单独使用捏

    AiMaosterMemeFeb 27, 2026
    CivitAI

    拖入你的视频(包含元数据)到comfyui,读取到了工作流,我是4070 12g VRAM,64g RAM,安装了sageattention cu128,开启了加速,由于直接OOM,所以我不能使用q8,只能使用q4km,我相信这也是绝大多数人的选择(8-16vram),并且我的分辨率保持1024(也许我还可以用其他外部软件进行补帧和高清修复,这点有很多软件比comfyui更专业),时长5s调整到4s。

    但我的加速报错module 'comfy.ldm.modules.attention' has no attribute 'attention',可能是因为某些插件或者pip库老了,我知道我的sageattention cu128没问题,我使用的预制轮子。

    所以只能把两个Patch Sage Attention KJ设置为disable。

    (注意,记得提前手动下载模型rife49.pth到ComfyUI\custom_nodes\ComfyUI-Frame-Interpolation\ckpts\rife,因为脚本的下载配置似乎过时了),rife49.pth你可以在谷歌随便搜索,有很多hugg平台用户提供的下载方式

    其他的设置没有改(也许可以试试不同的live2d lora或者描述)。

    我的输入和最终结果都在图床区,总体来说很不错,我的配置跑图很快乐,但视频不乐观,至少也得十几二十分钟,不过只要跑出来好的结果(并且不爆显存OOM),也能接受,本次的执行时间大约为49分钟以上,由于是1024的分辨率,所以很难受,但结果我感觉不错,还可以继续完善一下眼睛部位的描述。

    这种live2d很令人着迷。

    因为我自己正在研究一些素材资源,但我并不会使用spine2d以及live2d这些工具。

    也许未来我可以更好的进行游戏开发,壁纸制作?

    raidou88Apr 5, 2026
    CivitAI

    can anyone share a wan workflow that works with this? i upload a colored manga page but cant use this lora :(

    NiicolaApr 20, 2026
    CivitAI

    This is really awesome, though I would also like to see this for LTX 2.3, if possible :D

    LORA
    Wan Video 2.2 I2V-A14B

    Details

    Downloads
    2,980
    Platform
    CivitAI
    Platform Status
    Available
    Created
    9/15/2025
    Updated
    5/4/2026
    Deleted
    -

    Available On (2 platforms)

    Same model published on other platforms. May have additional downloads or version variants.