Editorial desk
Who signs the articles of this publication, what it covers and how to write to it.
The articles of this publication are signed by its editorial desk, not by a named person. The desk is the publication itself: it chooses the stories, writes them and corrects them, under the publication’s name.
How stories are chosen and checked
A story is covered only when it has a primary source: an announcement on a company’s or a lab’s own pages, a paper, a code repository, an official text. Every claim is checked against the passage of its source before publication, and every figure in a data table comes from a primary source or a public API. The page How we work describes each step, and the page Methods explains how the tables are built. Corrections are listed on the page Corrections.
Writing to the desk
Write to the address above to report an error, suggest a source or ask about an article.
What the desk covers
Text-to-video and image-to-video models, AI video editing, avatars and lip-sync, open video models, prices and benchmarks.
Latest articles
- FastFrames paper reports 36 ms per frame for a 1.3B video model on one B200
- Paper reports oracle tracks let EGS close 43% and 56% of the gap
- One Shot Any View reshoots clips on new camera paths; authors report lower pose error than two baselines
- PhysResponse benchmark: ten video models miss physical parameters, authors report
- H3Remover, an RL-tuned video object remover, reports large gains over original H3
- ScoutCache, a training-free method, reports up to 2.28× speedup on Self-Forcing
- MovieRoll long-video framework: authors report up to 78% less KV-cache memory
- Paper: a demand score predicts when a higher guidance scale fixes failed text-to-video clips
- LynnReal-Hand: hand-controlled video world model, with a fast VAE decoder
- Pika details H3 Flex, a post-trained MiniMax H3 model, and its plan prices