[batchku]

Hi Victor!

I love this. I've been playing around with drawing-to-animation techniques for a long time (check out this playlist of videos from my lab at CMU from 9 years ago: https://www.youtube.com/watch?v=Uv0qX7N2sJw&list=PLLhv4FQp7z...).

I gave your system a try and it's wonderful. The quality of animations are very different than what you get from rigged rag-dolls.

There are pros and cons to these approaches; the nice thing about masking/cropping/rigging (as our old "Dranimate" system did and Meta's "Animated Drawings" does) is that you prepare once, and then you can animate as many times in as many different ways as you want; you can drive it with mocap data; or with real-time inputs as with the videos of Dranimate above). The bad thing is that the first time preparation takes time.

I can see a lot of magic emerging from the combination of the two. I have recently gotten deep into this again (becauase now we have two little ones that are drawing all the time, and we started a Preschool in coastal norcal so kids are all around and this sort of magic really works on them, IRL).

I would love to connect with you sometime if you are up for chatting. Maybe there is a there there.

[vunderba]

You didn't do anything wrong - Kling can occasionally return a static result particularly for 2D images. It might have been the "grid lines" but Kling's model is effectively a black box.

https://kling.kuaishou.com/en

Side note: Nice video - setting up the rigging mesh reminds me of the "puppet warp" deformation system in Adobe Photoshop.

@Vic:

Bit of advice, you need to build some kind of a static checker (either in terms of the total compressed size of the video - very small file sizes likely indicate a failed gen, or in terms of a hash as a measure of frame-to-frame changes) - to help set up an automated appeals for customers... having to manage this on a case-by-case basis will destroy any enthusiasm you might have for this project.