China open-sourced a model that reconstructs any scene in 3D from a regular video, in real-time.
one camera. no LiDAR. 10,000+ frames without falling apart.
just walk around with your camera and watch the entire world get rebuilt in 3D at 20 fps.
→ runs at ~20 FPS on a single GPU
→ Stable over 10,000+ frames
→ Beats optimization-based methods on benchmarks
→ Works on drone footage, driving videos, indoor walkthroughs
100% open source.
I've been experimenting with Seedance 2.0 to create seamless extensions. I created this sequence called UBUMBE FOREST. Seamless extensions can actually be quite tricky to do and requires some precise magic with editing to connect them. There is no shortcut here, you must experiment until you find the sweet spot to make the transition seamless.
Ubumbe Forest - During a routine foraging trip deep in the Ubumbe forest in Tanzania, a father guides his daughter off the beaten path. Here, hidden from the rest of their tribe, she steps into a breathtaking new world and comes face-to-face with insects and creatures she has never seen before, even encountering a rare Peafox.
🟢Extending a scene however is fairly simple:
- Generate your first sequence.
- Upload that video as an Omni reference.
- Type this at the start of the prompt "Naturally continue the movement and momentum from the previous clip. Keep the exact same art style, lighting, character appearance, and physics." and then explain what happens next in this sequence.
- Congrats, you now have a scene extension that naturally flows from where the last video ended. However this is only the start of the puzzle, you need to join these clips so they're seamless.
Seedance 2.0 is able to match the pacing, remember the characters and even their voices using this technique. The hard part is the extension but check my second post to see how I'm doing it in Capcut.
GPT Image 2 + Seedance 2.0 - Prompt Share
Even though it doesn't come close to the effects in the anime Witch Hat Atelier, it was a major source of inspiration for this idea. This time, since the storyboard panels didn't include many scene-specific details, I described each shot directly in the prompt. You can check the prompts below.
Rainer Maria Rilke-《杜伊諾哀歌》
For beauty is nothing but the beginning of terror, which we are barely able to endure, and we are awed because it serenely disdains to annihilate us."
(美不過是恐怖的開端,我們尚能勉力承受;我們之所以敬畏它,是因為它冷靜地不屑於將我們毀滅。)