Open saakolch opened 2 months ago
would be great if image2video is ok
Dear @scholltan and @saakolch , Thank you so much for the suggestions. But current I2V models takes starting frame (optionally last frame) as inputs, and output the video. This can be a problem, because, we now assume all frames (across devices) conditioned the same condition. This is problematic since video clips from different devices often begin with different starting frames.
To address this, we are exploring a hierarchical generation process. Our plan is to first generate key frames for each clip, followed by distributed generation across devices. We are still in the testing phase with this new method.
Stay tuned for further updates!
Is it possible to use an image/images to create a video?