RuntimeWarning: invalid value encountered in cast

[ysxk]

系统：windows server 2022 显卡：a100 环境：秋叶包 模型已全部下载comfyui\models\echomimic\下的vae和sd-vae-ft-mse中都存放了模型，也试过把工作流vae的键值改为vae或者sd-vae-ft-mse 均会出现视频黑屏的情况 开关降噪都会黑屏 以下是完整日志 我注意到其中有一个远程主机断开连接，但我不知道是不是在下载vae？ rompt executed in 19.68 seconds got prompt ** refer in EchoMimic V1 mode!** loaded temporal unet's pretrained weights from E:\ComfyUI-aki-v1.4\models\echo_mimic\unet ... Loaded 0.0M-parameter motion module Use audio drive video! video in 5 FPS, audio idx in 50FPS whisper_chunks: (298, 50, 384) audio_fea_final: torch.Size([1, 298, 50, 384]) ref_image_latents shape: torch.Size([1, 4, 16, 64]) face_mask_tensor shape: torch.Size([1, 1, 1, 128, 512]) face_locator_tensor shape: torch.Size([2, 320, 1, 16, 64]) final_length is : 1 Moviepy - Building video E:\ComfyUI-aki-v1.4\output\409309_audio.mp4. MoviePy - Writing audio in 409309_audioTEMP_MPY_wvf_snd.mp4 MoviePy - Done. Moviepy - Writing video E:\ComfyUI-aki-v1.4\output\409309_audio.mp4

Moviepy - Done ! Moviepy - video ready E:\ComfyUI-aki-v1.4\output\409309_audio.mp4 savingE:\ComfyUI-aki-v1.4\output\409309_echo.mp4 at E:\ComfyUI-aki-v1.4\output\409309_audio.mp4 Pipelines loaded with dtype=torch.float16 cannot run with cpu device. It is not recommended to move them to cpu as running them will fail. Please make sure to use an accelerator to run the pipeline in inference, due to the lack of support forfloat16 operations on this device in PyTorch. Please, remove the torch_dtype=torch.float16 argument, or use another device for inference. Pipelines loaded with dtype=torch.float16 cannot run with cpu device. It is not recommended to move them to cpu as running them will fail. Please make sure to use an accelerator to run the pipeline in inference, due to the lack of support forfloat16 operations on this device in PyTorch. Please, remove the torch_dtype=torch.float16 argument, or use another device for inference. Pipelines loaded with dtype=torch.float16 cannot run with cpu device. It is not recommended to move them to cpu as running them will fail. Please make sure to use an accelerator to run the pipeline in inference, due to the lack of support forfloat16 operations on this device in PyTorch. Please, remove the torch_dtype=torch.float16 argument, or use another device for inference. Pipelines loaded with dtype=torch.float16 cannot run with cpu device. It is not recommended to move them to cpu as running them will fail. Please make sure to use an accelerator to run the pipeline in inference, due to the lack of support forfloat16 operations on this device in PyTorch. Please, remove the torch_dtype=torch.float16 argument, or use another device for inference. Prompt executed in 47.42 seconds Exception in callback _ProactorBasePipeTransport._call_connection_lost(None) handle: <Handle _ProactorBasePipeTransport._call_connection_lost(None)> Traceback (most recent call last): File "E:\ComfyUI-aki-v1.4\python\lib\asyncio\events.py", line 80, in _run self._context.run(self._callback, *self._args) File "E:\ComfyUI-aki-v1.4\python\lib\asyncio\proactor_events.py", line 165, in _call_connection_lost self._sock.shutdown(socket.SHUT_RDWR) ConnectionResetError: [WinError 10054] 远程主机强迫关闭了一个现有的连接。 got prompt ** refer in EchoMimic V2 mode!** loaded temporal unet's pretrained weights from E:\ComfyUI-aki-v1.4\models\echo_mimic\unet ... Load motion module params from E:\ComfyUI-aki-v1.4\models\echo_mimic\v2\motion_module.pth Loaded 453.20928M-parameter motion module when use echo v2,need choice a pose dir,for get error using default pose ! * infer length is 296 poses_tensor:torch.Size([1, 3, 296, 512, 512]) video in 5 FPS, audio idx in 50FPS ori: tensor(-3.9043, device='cuda:0', dtype=torch.float16) tensor(3.9043, device='cuda:0', dtype=torch.float16) latents shape:torch.Size([1, 4, 296, 64, 64]), video_length:296 final_length is : 296 **** Moviepy - Building video E:\ComfyUI-aki-v1.4\output\413088_audio.mp4. MoviePy - Writing audio in 413088_audioTEMP_MPY_wvf_snd.mp4 MoviePy - Done. Moviepy - Writing video E:\ComfyUI-aki-v1.4\output\413088_audio.mp4

Moviepy - Done ! Moviepy - video ready E:\ComfyUI-aki-v1.4\output\413088_audio.mp4 savingE:\ComfyUI-aki-v1.4\output\413088_echo.mp4 at E:\ComfyUI-aki-v1.4\output\413088_audio.mp4 Prompt executed in 61.82 seconds

[ysxk]

能不能通过修改代码来指定模型路径？

[smthemex]

你的图片有问题，不是VAE的问题，ref_image_latents shape: torch.Size([1, 4, 16, 64]) 这里可以看得出来，你的输入图片的格式不对，你换个一个别的图片，试试 不过我可以改一下代码，避免图片输入的不规范，以下是正常输入的图片尺寸512*512示例 VAE的地址栏，就是随便你填的，你填Z:/VAE ，C/VAE都行。只要这个目录下，有config文件和模型，你要看config文件是不是加了前缀

[ysxk]

又试了一次，使用的是E:\ComfyUI-aki-v1.4\custom_nodes\ComfyUI_EchoMimic\echomimic_v2\EMTD_dataset\ref_imgs_by_FLUX\woman 也就是仓库中自带的图片，依然是纯黑的视频 * infer length is 336 poses_tensor:torch.Size([1, 3, 336, 512, 512]) video in 24 FPS, audio idx in 50FPS ori: tensor(-3.9043, device='cuda:0', dtype=torch.float16) tensor(3.9043, device='cuda:0', dtype=torch.float16) latents shape:torch.Size([1, 4, 336, 64, 64]), video_length:336 final_length is : 336 *** E:\ComfyUI-aki-v1.4\custom_nodes\ComfyUI_EchoMimic\src\utils\util.py:98: RuntimeWarning: invalid value encountered in cast x = (x 255).numpy().astype(np.uint8)

Prompt executed in 30.76 seconds got prompt when use echo v2,need choice a pose dir,for get error using default pose ! * infer length is 336 poses_tensor:torch.Size([1, 3, 336, 512, 512]) video in 24 FPS, audio idx in 50FPS ori: tensor(-3.9043, device='cuda:0', dtype=torch.float16) tensor(3.9043, device='cuda:0', dtype=torch.float16) latents shape:torch.Size([1, 4, 336, 64, 64]), video_length:336 final_length is : 336 *** E:\ComfyUI-aki-v1.4\custom_nodes\ComfyUI_EchoMimic\src\utils\util.py:98: RuntimeWarning: invalid value encountered in cast x = (x 255).numpy().astype(np.uint8)

[smthemex]

numpy版本不对？ 你pip show numpy 看看版本是多少 官方要求是1.26.4

[ysxk]

numpy版本是对的，我刚刚看到这个有关numpy的报错特地看了一下v2的依赖文件，v2要求的是1.23.5，我装了1.23.5也是黑屏，我再装回1.26.4试试

[ysxk]

重装过1.26.4后是一样的情况

[ysxk]

装完之后有重启过comfyui的

[smthemex]

还是np相关依赖的问题，你这个最后的报错看着眼熟

[smthemex]

或者你下一个fp16的模型，然后去掉名字去掉_fp16

[ysxk]

fp16的echomimic2吗 还是vae？

[ysxk]

我找了下，echomimic 和vae都没在huggingface上找到标了fp16的版本

[ysxk]

又折腾了几个小时，用ubuntu server自己动手安装了comfyui和echomimicv2 还是一样的报错 final_length is : 118 /root/comfyui/custom_nodes/ComfyUI_EchoMimic/src/utils/util.py:98: RuntimeWarning: invalid value encountered in cast x = (x * 255).numpy().astype(np.uint8) 是我运气太差了么2333 新机器上也是一样的numpy1.26.4 我还怀疑过镜像源，但是换了好几个镜像源安装np都是如此

[ysxk]

....我想可能是我的工作流有问题，我在https://github.com/smthemex/ComfyUI_EchoMimic/issues/45看到一位朋友的工作流，我完全照抄了过来，现在已经能够正常出视频了

[ysxk]

暂时不清楚问题出在哪个参数，我打算明天睡醒再试试，到时候给您反馈，以便能帮助到更多人

[ysxk]

又测试了几遍，改了几个参数，目前没发现是哪个参数导致的黑屏，我能成功生成的时候也没有做多余操作，就是单纯的照抄了一下那个大佬的工作流，难道是我去上洗手间的时候他自己修好了？2333

[smthemex]

SDXL vae是老毛病了，所以官方要用ft的版本，主要是fp16的问题，可能是你的A100的缘故，按理，你从菜单直接模型加载和sampler节点，应该是能跑的。但是加入了其他节点，就不好说了