Video Generation

Today's Stories

1. 1

   World Labs unveils Atlas, a single AI model that generates, reconstructs, and simulates 3D worlds from just a few photos

   World Labs unveiled Atlas, an omni-model trained on text, images, video, and 3D data that anchors every input to a specific position in 3D space. It generates up to one minute of 1440p video with user-controlled camera paths, reconstructs real scenes from as few as two or three photos, and can handle over a hundred inputs. This addresses a problem Fei-Fei Li outlined in a November 2025 essay: current multimodal and video models break data into 1D or 2D sequences, making simple spatial tasks needlessly hard. Atlas's shared spatial understanding separates it from pure language or video models, and it outperforms specialized models in human evaluations, scoring 75% against MiniMax H3 and 94% against Seedance 2.5.

   Atlas's native 3D output supports point clouds and 3D Gaussian splats, matching the representation in Marble. It also serves as a real-to-sim tool for robotics, building on an August 2026 real-to-sim-to-real engine that ran for an hour each on five robot platforms without human intervention. Atlas is currently in an early-access program for select partners.

2. 2

   World Labs unveils spatial AI model Atlas

   World Labs released Atlas, a world model trained from scratch to handle text, images, video, and 3D natively. It will power the company's spatial generator Marble. Atlas can generate new views from one or a few photos, reconstruct real spaces, create bullet-time videos, and simulate robot camera data. The company says it outperforms open-source reconstruction models on DTU, ETH3D, and ScanNet benchmarks.

   Atlas is in early access for select partners. Developers and companies can sign up via a form, and World Labs plans to integrate Atlas into future versions of its products, including Marble.

3. 3

   Visko raises $10M, launches live AI video model Orbis

   Visko raised $10 million in pre-seed funding from Llama Ventures and opened public access to its first foundation model, Orbis, a live model that generates long-form videos. Unlike conventional AI video generators that produce short, static clips, Orbis runs continuously, allowing real-time user interventions without breaking visual coherence. The startup claims it maintains quality for hours, addressing a key limitation of existing models.

   Orbis is available to try on Visko's website, streaming at 4K resolution at 24 frames per second. The company says it can generate worlds for hours without degradation, though costs for dynamic generation may be high.

4. 4

   AI video displaces actors in China

   Since ByteDance's Seedance 2.0 launched, AI-generated videos are replacing human performers in Chinese productions. In Q1 2026, about 128,000 short dramas were published, three times the total for all of 2025, with ninety-five percent AI-generated, per the China Netcasting Services Association. The industry employs 690,000 people directly, and 15 million list livestreaming as their primary job. Some performers are forced to "distil" their voice and likeness into AI tools before getting fired, and AI-related labor disputes have risen over the past two to three years.

   One minute of AI video now costs just $90 to $120, about ten percent of what human production used to cost, according to Tsinghua University professor Shen Yang. With Seedance 2.5 and Wan 3.0, AI video quality has jumped again.

5. 5

   What Counts as a World Model?

   A Reddit user asked the ML community to clarify what a world model is, noting that most current builds are fancy video generation models. The term is contested, with simulators, physics engines, emulators, and digital twins all being considered as possible world models, which creates confusion for non-experts.

   The discussion hinges on whether a world model must operate on learned representations, not just hand-crafted physics, leaving room for ML-based simulators.

6. 6

   BearJam: AI automation needs human oversight

   BearJam, an AI-powered video production company, shared its approach to responsible automation after two years of experience, emphasizing a hybrid model that combines AI with human judgment. The company argues that automation scales errors as fast as output, so human checkpoints and clear limits are essential to avoid legal and reputational risks, especially since liability for AI output rests with the deployer, not the platform.

   BearJam has turned down projects it could technically deliver, including misleading before-and-after imagery, and commits 1% of revenue from AI-intensive projects to nature restoration in Kent.