EchoMimic V3 - 一张照片一段音频生成半身数字人说话视频 支持手势动作 支持50系显卡 一键整合包下载
EchoMimicV3 是蚂蚁集团开发的一款基于人工智能的多模态人类动画生成系统，通过 13 亿参数的模型实现“声音+文本+图像”的统一驱动，生成高度逼真的虚拟人像动画（如说话的头部、肢体动作等）。简单来说，它能让虚拟角色根据输入的语音、文字或参考图像，自动生成同步的口型、表情和身体动作，像真人一样自然。
应用领域
虚拟数字人
为虚拟主播、AI 客服、智能助手提供自然动作和口型，提升交互体验。
示例：电商直播中，虚拟模特根据商品介绍自动调整表情和手势。
影视与游戏制作
快速生成动画片段，降低传统动作捕捉的成本和时间。
示例：为游戏角色添加语音驱动的实时表情，或修复影视中口型与台词不匹配的问题。
教育与研究
辅助语言学习（如生成带口型的外语教学动画），或用于心理学、人类行为研究。
示例：分析不同语调对虚拟人表情的影响。
无障碍技术
帮助语言障碍者通过文本或图像生成表达动作，或为听障人士提供可视化语音内容。
示例：将语音转换为带口型的动画，方便听障者“观看”对话。
使用教程：（建议N卡，显存12G起。支持50系显卡，基于CUDA12.8）
上传一张人物照片（支持头像和半身照）和驱动音频，设置参数，提交即可。
默认不带模型，解压后需手动下载模型。
下载地址：
夸克网盘：https://pan.quark.cn/s/f7f3026f28ee
百度网盘：
**** 本内容需购买 ****
大佬，购买的资源，后面更新后，也要再支付一次吗？ wujunchen 发表于 2025-8-15 11:29
大佬，购买的资源，后面更新后，也要再支付一次吗？
是的
如果是刚需，可以开个年费会员，所有包都是免费下载 我的是特斯拉t10显卡，驱动都正常，但是这个还是直接用cpu跑起来啦，慢的不能用，自己不会修改设置。哎
weilaiyilai 发表于 2025-8-18 17:59
我的是特斯拉t10显卡，驱动都正常，但是这个还是直接用cpu跑起来啦，慢的不能用，自己不会修改设置。哎
...
这卡好像不支持bf16
tesla很多老架构卡，很多模型都跑不了 无言以对 发表于 2025-8-18 18:22
这卡好像不支持bf16
tesla很多老架构卡，很多模型都跑不了
哦，谢谢回复 -----------------------------------------------------
更多前沿AI技术和应用，访问 https://deepface.cc
软件WebUI正在启动中，请稍后...
-----------------------------------------------------
*一键包制作：https://deepface.cc
* 请合理合法使用AI技术，不要用于违法用途
* 否则由此带来的任何责任后果，请自行承担
2025-09-06 00:28:53.594201: I tensorflow/core/util/port.cc:113] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.
WARNING:tensorflow:From E:\EchoMimic_v3\deepface\lib\site-packages\keras\src\losses.py:2976: The name tf.losses.sparse_softmax_cross_entropy is deprecated. Please use tf.compat.v1.losses.sparse_softmax_cross_entropy instead.
loaded 3D transformer's pretrained weights from models/Wan2.1-Fun-V1.1-1.3B-InP\./ ...
### missing keys: 93;
### unexpected keys: 0;
['audio2token.proj.weight', 'audio2token.norm.weight', 'audio2token.norm.bias', 'blocks.0.cross_attn.k_audio.weight', 'blocks.0.cross_attn.v_audio.weight', 'blocks.0.cross_attn.norm_k_audio.weight', 'blocks.1.cross_attn.k_audio.weight', 'blocks.1.cross_attn.v_audio.weight', 'blocks.1.cross_attn.norm_k_audio.weight', 'blocks.2.cross_attn.k_audio.weight', 'blocks.2.cross_attn.v_audio.weight', 'blocks.2.cross_attn.norm_k_audio.weight', 'blocks.3.cross_attn.k_audio.weight', 'blocks.3.cross_attn.v_audio.weight', 'blocks.3.cross_attn.norm_k_audio.weight', 'blocks.4.cross_attn.k_audio.weight', 'blocks.4.cross_attn.v_audio.weight', 'blocks.4.cross_attn.norm_k_audio.weight', 'blocks.5.cross_attn.k_audio.weight', 'blocks.5.cross_attn.v_audio.weight', 'blocks.5.cross_attn.norm_k_audio.weight', 'blocks.6.cross_attn.k_audio.weight', 'blocks.6.cross_attn.v_audio.weight', 'blocks.6.cross_attn.norm_k_audio.weight', 'blocks.7.cross_attn.k_audio.weight', 'blocks.7.cross_attn.v_audio.weight', 'blocks.7.cross_attn.norm_k_audio.weight', 'blocks.8.cross_attn.k_audio.weight', 'blocks.8.cross_attn.v_audio.weight', 'blocks.8.cross_attn.norm_k_audio.weight', 'blocks.9.cross_attn.k_audio.weight', 'blocks.9.cross_attn.v_audio.weight', 'blocks.9.cross_attn.norm_k_audio.weight', 'blocks.10.cross_attn.k_audio.weight', 'blocks.10.cross_attn.v_audio.weight', 'blocks.10.cross_attn.norm_k_audio.weight', 'blocks.11.cross_attn.k_audio.weight', 'blocks.11.cross_attn.v_audio.weight', 'blocks.11.cross_attn.norm_k_audio.weight', 'blocks.12.cross_attn.k_audio.weight', 'blocks.12.cross_attn.v_audio.weight', 'blocks.12.cross_attn.norm_k_audio.weight', 'blocks.13.cross_attn.k_audio.weight', 'blocks.13.cross_attn.v_audio.weight', 'blocks.13.cross_attn.norm_k_audio.weight', 'blocks.14.cross_attn.k_audio.weight', 'blocks.14.cross_attn.v_audio.weight', 'blocks.14.cross_attn.norm_k_audio.weight', 'blocks.15.cross_attn.k_audio.weight', 'blocks.15.cross_attn.v_audio.weight', 'blocks.15.cross_attn.norm_k_audio.weight', 'blocks.16.cross_attn.k_audio.weight', 'blocks.16.cross_attn.v_audio.weight', 'blocks.16.cross_attn.norm_k_audio.weight', 'blocks.17.cross_attn.k_audio.weight', 'blocks.17.cross_attn.v_audio.weight', 'blocks.17.cross_attn.norm_k_audio.weight', 'blocks.18.cross_attn.k_audio.weight', 'blocks.18.cross_attn.v_audio.weight', 'blocks.18.cross_attn.norm_k_audio.weight', 'blocks.19.cross_attn.k_audio.weight', 'blocks.19.cross_attn.v_audio.weight', 'blocks.19.cross_attn.norm_k_audio.weight', 'blocks.20.cross_attn.k_audio.weight', 'blocks.20.cross_attn.v_audio.weight', 'blocks.20.cross_attn.norm_k_audio.weight', 'blocks.21.cross_attn.k_audio.weight', 'blocks.21.cross_attn.v_audio.weight', 'blocks.21.cross_attn.norm_k_audio.weight', 'blocks.22.cross_attn.k_audio.weight', 'blocks.22.cross_attn.v_audio.weight', 'blocks.22.cross_attn.norm_k_audio.weight', 'blocks.23.cross_attn.k_audio.weight', 'blocks.23.cross_attn.v_audio.weight', 'blocks.23.cross_attn.norm_k_audio.weight', 'blocks.24.cross_attn.k_audio.weight', 'blocks.24.cross_attn.v_audio.weight', 'blocks.24.cross_attn.norm_k_audio.weight', 'blocks.25.cross_attn.k_audio.weight', 'blocks.25.cross_attn.v_audio.weight', 'blocks.25.cross_attn.norm_k_audio.weight', 'blocks.26.cross_attn.k_audio.weight', 'blocks.26.cross_attn.v_audio.weight', 'blocks.26.cross_attn.norm_k_audio.weight', 'blocks.27.cross_attn.k_audio.weight', 'blocks.27.cross_attn.v_audio.weight', 'blocks.27.cross_attn.norm_k_audio.weight', 'blocks.28.cross_attn.k_audio.weight', 'blocks.28.cross_attn.v_audio.weight', 'blocks.28.cross_attn.norm_k_audio.weight', 'blocks.29.cross_attn.k_audio.weight', 'blocks.29.cross_attn.v_audio.weight', 'blocks.29.cross_attn.norm_k_audio.weight']
### All Parameters: 1707.215168 M
### attn1 Parameters: 0.0 M
Missing keys: 0, Unexpected keys: 0
### missing keys: 0;
### unexpected keys: 0;
[] []
Press any key to continue . . . 没理解，问题出在哪里，模型都全了，缺权重？ tnt56466939 发表于 2025-9-6 00:32
-----------------------------------------------------
更多前沿AI技术和应用，访问 https://deepface.c ...
手动设置虚拟内存，32G起 无言以对 发表于 2025-9-6 10:18
手动设置虚拟内存，32G起
谢谢，我晚上测试一下，另外咨询一下，为什么加载时电脑会特别卡，配置挺高的还在加载时特别卡呢？
页:
[1]
2