Scale HUMAN animations with OmniHuman-1 (Technical Review)

Author(s): Deltan Lobo

Originally published on Towards AI.

Rethinking the Scaling-Up of One-Stage Conditioned Human Animation Models

This member-only story is on us. Upgrade to access all of Medium.

Video generation models have been fun till now. We’ve seen several video generation models that create content — especially HUMAN ANIMATIONS.

But those models focus only on either facial expressions or body movement — but not both.

TikTok’s parent company ByteDance bridges this gap by releasing OmniHuman[1]… Now we are able to try out a broader range of animations, from subtle lip-syncing to full-body motion by maintaining consistency in different body proportions.

Of course yeah, traditional models were capable of creating human videos. But as I said it was only limited to a specific body movement and also even if it was created, the characters in the video would look lifeless.

I hope you might be aware of some portrait videos giving explanations about something.

But OmniHuman solves this up to a certain extent. Whether you’re animating a portrait, half-body, or full-body character, this model adapts seamlessly. Supports both talking and singing, handles human-object interactions and challenging body poses, and accommodates different image styles.

Video generation has made huge leaps in recent years… Thanks to diffusion models[2] and transformer architectures[3]. These models work incredibly well for general video generation (like text-to-video systems).

But,… Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI