Tencent / MimicMotion

High-Quality Human Motion Video Generation with Confidence-aware Pose Guidance
https://tencent.github.io/MimicMotion/
Other
1.92k stars 162 forks source link

我这生成的视频咋是黑乎乎的呢? #73

Open ucas010 opened 3 months ago

ucas010 commented 3 months ago

采用这个code https://github.com/Tencent/MimicMotion/issues/13 做的,参考视频就是抖音原图的那种跳舞视频, 结果生成的视频是黑乎乎的, 而且我测试(如下)也是黑乎乎的视频 官方代码: python inference.py --inference_config configs/test.yaml

生成的视频在outputs里面:pose1_20240808170131.mp4 打开一看黑乎乎的,啥都没有

请大佬看下,这是咋回事啊? 感谢感谢

ucas010 commented 3 months ago

https://github.com/user-attachments/assets/b64efaca-25dc-4daa-9ff4-befb3d914cbb

ucas010 commented 3 months ago

重新装环境,又测试了下,采用的官方的图片,生成的视频如下

https://github.com/user-attachments/assets/c36c1cf6-db0c-4442-b597-c5b145aafd5b

https://github.com/user-attachments/assets/1277df9d-5105-4e0a-a27e-e581da546277

这个视频也没复现啊。。

ucas010 commented 3 months ago

参考图片, sample_1 参考视频, https://github.com/user-attachments/assets/5c428150-7f78-47da-b789-94430078283d

生成视频: https://github.com/user-attachments/assets/716adc8a-cde3-44f9-9b3e-e0c6bfe172da

ucas010 commented 3 months ago

我哪里设置有问题么?

CatDroid commented 3 months ago

我用comfyui的工作流 在2080的GPU上 由于显存紧张,改成2 context size =3 context overlay =2 生成的视频也是黑的。 用的工作流参考 https://github.com/kijai/ComfyUI-MimicMotionWrapper