search

Doubiiu / DynamiCrafter

[ECCV 2024] DynamiCrafter: Animating Open-domain Images with Video Diffusion Priors
Apache License 2.0
2.09k stars 165 forks source link

In the future, will there be plans to train a frame interpolation model at a resolution of 1024？ #47

Open hben35096 opened 3 months ago

hben35096 commented 3 months ago

Thank you for training such a remarkable model; I can see its tremendous potential. In the future, will there be plans to train a frame interpolation model at a resolution of 1024? Will techniques like LCM, Turbo, or Lightning be employed to enhance video generation speed in these models?

https://github.com/Doubiiu/DynamiCrafter/assets/139383150/68f96d33-d9b5-45d4-a628-22acffda38f6

Doubiiu commented 3 months ago

Hi. Thanks for your recognition, which is very helpful for open-source projects. Currently, we are focusing on some research topics about interpolation and will do that for 1024x576 if the resource is available. Personally, I am very interested in those acceleration techniques but now it is not my focus and I hope community or someone else can help with that. Thanks again for your interest.

Jaythschild commented 3 months ago

Thank you for training such a great model. Currently, the test video generation time is about 2 seconds, reaching its limit. After that, the image will appear inverted. Is it possible to extend the generated content to look like a complete shot。 感谢训练出这么棒的模型，目前经过多次测试视频生成时长在约为2S到达极限，之后画面会出现倒放，请问有没有可能延长生成的视频的时长，至少看起来像一个完整的镜头

https://github.com/Doubiiu/DynamiCrafter/assets/127034579/0c188fea-bd7b-4c4b-97b9-331e830a53fb

Doubiiu commented 3 months ago

Hi. Did you make it generate longer videos by doing what in #36?