NewsKling says AI video's bottleneck has moved to meaningresearch
Every SLOP TV News story about research.
NewsKling says AI video's bottleneck has moved to meaning
ExplainersAdobe and UT Austin's AgiWorld builds persistent video worlds
ExplainersTencent's WeVisCut edits a long video from a script with a 9B model
L3G-World
L3G-World generates 3D worlds at 9.02 FPS on one H100
The KAIST paper's latent 3D Gaussian memory tops DL3DV baselines and uses 266 to 417 times fewer 3D primitives than NeoVerse, but no weights are out yet.
filmflow
FilmFlow turns a text prompt into a Blender camera path
The Shanghai Jiao Tong University-led system ships with CinemaBench, 150 Blender scenes and 750 camera instructions, but no code is released yet.
solarrenderer
SolarRenderer turns game-engine block-outs into video in 4 steps
The team froze a 33B MiniMax-H3 video model, added two small LoRAs, and cut sampling from 50 steps to four, with a streaming variant at 768p.
Google
Google's co-director wants to end character drift in AI video
A research framework stacks planning agents over Gemini and Veo to keep a character looking like the same person across minutes of footage, and it is not a product you can use yet.
WanPE
WanPE boosts AI video preference 50.86 points at 30 seconds
A 397B prompt enhancer trained on 1.05 million real videos beats raw user requests by a far wider margin at 30 seconds than on shorter clips, and the Wan team has published no weights.
runway
Runway Wants AI Video to Stream as You Prompt It, Frame by Frame
A research post explains how Runway post-trains existing video models to generate causally and stream frames, and what of that is actually live today.
astronex
Astronex-World 1.0 Streams Open World-Model Video on a Single 48GB GPU
The Apache-2.0 release is built on Alibaba's Wan 2.2 prior and generates 832x480 video at 24 frames per second without a render queue, its authors say.