WonderHuman: Hallucinating Unseen Parts in Dynamic 3D Human Reconstruction
TLDR: Gaussian full body Avatars using Score Distillation Sampling in both Canonical and Posed Space.
https://wyiguanw.github.io/WonderHuman/
ByteDance 🎶 has just released OmniHuman-1, and it’s insanely good. It takes a single image and audio to produce these results.
TLDR: Video Diffusion Model using a Multi-Modal Diffusion Transformer (MMDiT), trained with flow matching and using a ton of data.
https://omnihuman-lab.github.io/
“This guy is on a Zoom call, but not actually in front of the camera. And you can also run it with text to speech open-source model trained on your own voice, written by a language model listening to the other user’s voice input.
https://x.com/rohanpaul_ai/status/1886757493998805226
“The latest AI & 3D tools make it really fun to create for my VFX channel. Also grateful that my new AI-focused YouTube channel (Bilawal Sidhu) has crossed 10K subs. Here’s my 2025 creation stack: > Google Veo 2, OpenAI Sora (text-to-video) > RunwayML (video-t o-video) > Kling
https://x.com/bilawalsidhu/status/1886225400855925071
Inside the “Virtual Lab” where AIs and humans collaborate
https://www.freethink.com/artificial-intelligence/virtual-lab-interdisciplinary-research
Dress-1-to-3: Single Image to Simulation-Ready 3D Outfit with Diffusion Prior and Differentiable Physics – https://dress-1-to-3.github.io/
“Video-to-video AI is underrated. Stop wasting your time text prompting, and re-rolling the slot machine just to get the right composition. 3D is easy — you can literally greybox & kitbash your scene together in minutes. Can’t wait till this tech is running in real-time.
