lixunsong
lixunsong
We haven't tested on Mac, so we're not sure if it will run. You can give it a try and welcome your feedbacks :)
Hello @cocktailpeanut , I'm not sure the OOM error was cuased by the limit gpu VRAM. You can first run a smaller generation size, eg 256x256.
Right now the gradio demo needs to update a pose(keypoints) video, not the raw video. You can use the script `tools/vid2pose` to extract it first.
What's the error message about this crash?
关于脸部扭曲，可以尝试用更高的分辨率生成，如 (768x1024)。的确这里我们与官方使用的 keypoints map 有所不同。 Regarding facial distortion, you can try generating at a higher resolution, such as 768x1024. Indeed, here we have some differences from the key points map...
We haven't tried poetry. Is there any problem for you then setting up the environment?
The VRAM size is up to the resolution. It requires about 15GB to generate a video with resolution of 512x768. You can reduce the resolution to lower VRAM usage.
It seems you are missing binaray files of stable-diffusion-v1-5. Is this file `stable-difusion-v1-5/unet/diffusion_pytorch_model.bin` exists in your directory?
Hello @FantasticMrCat42 , we have updated a script to download weights automatically. You can try it~
It seems the downloaded weights are broken