Google DeepMind just unveiled GenCeption. This AI turns video into depth, segmentation, 3D keypoints, and searchable 4D worlds. 8 wild examples:

2. The same prompt-steered model switches between depth, surface normals, segmentation, and camera rays Even across humans and robots.

3. One model, any video task

4. Video becomes a queryable 4D scene. It predicts depth, geometry, camera motion, and segmentation... then finds the object in 4D.

5. Prompt the reconstructed 4D It grounds objects directly inside the 4D scene.

6. Emerging Behavior - Sim-to-Real Generalization - Generalization to multiple instances - Generalization to unseen objects

7. 3D keypoints under extreme motion

8. Depth. Segmentation. 2D keypoints. 3D keypoints. One video-generation backbone becoming a general-purpose vision model.

Discuss on HuggingFace Project Page:

If you enjoyed this thread, Follow me @minchoi and please Bookmark, Like, Comment & Repost the first Post below to share with your friends:

this is so cool. we are building esports/gaming agent and are deep into vision models, who is building on new experimental models, which vision models worth testing? aiming to build this for the likes of GTA, Valorant, League of Legends, CS and many others.

searchable 4D worlds is the part that changes what "video search" even means

I don't buy the searchable 4D worlds yet till I see the query speed, how fast does GenCeption search them???

Depth-driven visuals unlock deeper design insights

Great!!

Wild Examples 💯

💯

Google is searching 3D space now. depth, segmentation, keypoints - infrastructure for making reality queryable. your camera becomes a search engine

Querying a 4D scene in natural language is a big jump

Wow. This is crazy. So basically I can use this to help me plan how to design and decorate the interior my place?

Turning video into searchable 4D worlds is a genuinely big capability jump.

AI vision is evolving incredibly fast

This changes everything wow

Very interesting. Love this

🙏

Video turning into a searchable 3D world is wild. It also quietly resets what counts as basic AI skills every few months.

😧!!

‼️

AI keeps opening up new creative possibilities.

video into searchable 4d worlds is wild. searchable is the word doing the work there. a 4d scene you can query is what robotics and mapping teams actually build on, not just a clip that looks cool.

日本語で解説を書きました Japanese breakdown here:

That’s impressive

I don’t understand why they don’t add a “teacher AI” that interprets the footages and adds a “Geometry pass” to the data, this would allow for better robot training or even real time reading of world space for robots.

this is crazy