Summary
This roundup from Theoretically Media covers the latest AI video and image news, starting with ByteDance. The company is bringing back OmniHuman with a 2.0 version of its avatar and lip-sync technology, now with cinematic camera movement and more emotion. The host finds it improved but still behind Seedance 2 and 2.5, and notes it is image-driven only and currently region-locked, with a Dreamina release expected later.
Bytedance also gave a first look at an unnamed world model at an event in Beijing. Footage shows mouse-based navigation, prompting objects into a scene, and camera retracking of an existing shot from Stephen Chow’s King of Comedy, which fits with Chow’s recent AI video licensing deal with ByteDance.
The video then turns to Krea 3, which launched quickly after Krea 2. Features like Allora and style transfer are still listed as coming soon, but its precise edit mode stands out. The host uses it to remove grain from GPT-generated images, change a character’s outfit without content pushback, and create last frames for first-frame, last-frame video workflows. The episode also touches on the return of Sora.
📺 Source: Theoretically Media · Published October 15, 2026
🏷️ Format: Roundup
