正在加载视频...

视频加载失败

We just tested what might be the most accurate motion transfer workflow we've seen so far. It's a little more involved than a standard workflow, but the results are pretty impressive. Here's the basic workflow: 1. Start with your reference footage. Create character and environment references. 2. Convert the...

30,770 次观看 • 1 个月前 •via X (Twitter)

9 条评论

NuoNuo YSN 的头像
NuoNuo YSN1 个月前

Why not just upload the original video as a reference instead of processing it first

BongBong 的头像
BongBong1 个月前

So shouldn't the generative services be creating a depth map at the same time as the generated video? Might be something they should offer.

DENJIN 的头像
DENJIN1 个月前

The problem with the skeleton workflow (in addition to the occasional glitchiness) was that it was overlayed on top of the depth map footage and then being fed to the AI, which is why you were getting clones/duplicates in the output.

W. S. Waldo 的头像
W. S. Waldo1 个月前

Excellent video. Detailed. Thorough. And I like that you didn't show everything magically working. The failures are more informative than the successes. Thanks for this!

AI Mastery Guide 的头像
AI Mastery Guide1 个月前

Depth maps making motion transfer this clean is genuinely impressive work 🎥

Moez Zhioua 的头像
Moez Zhioua1 个月前

Using depth maps as the motion cue feels right, I noticed far fewer artifacts than pure skeleton feeds. The extra conversion step adds overhead but the consistency is worth it. Rule: let depth drive the transfer

Armaan | AI Systems 的头像
Armaan | AI Systems1 个月前

the reason depth-first beats skeleton-only tracking: depth maps carry occlusion and body volume, not just joint positions, so the generator has real spatial info instead of guessing what's behind an arm or how a character fills out a frame. stacking depth + segmentation + pose is multi-signal conditioning — removing ambiguity before generation even starts, not just throwing more tools at it.

Dir-BillSun 的头像
Dir-BillSun1 个月前

The depth pass preserves something skeleton guidance keeps losing here: the actor’s weight through turns and occlusions. I’d test the face separately—the body can be accurate while the intention of the performance disappears.

Pranav Joshi 的头像
Pranav Joshi1 个月前

The annoying part is that the workflow is still pretty involved, but the result makes it look worth it. Motion transfer is getting scary good.

相关视频