Tavus's Falcon cuts conversation video latency to 430ms
The avatar model generates video one chunk at a time so it can interrupt and react, and Tavus measured a 430-millisecond audio-to-video delay on a B200.
Everything we've published, newest first.
The avatar model generates video one chunk at a time so it can interrupt and react, and Tavus measured a 430-millisecond audio-to-video delay on a B200.
The 5B model turns rough game footage into detailed frames at more than 24 frames per second on a single RTX 5090; the 14B build runs at 480p.
The Y Combinator-backed company opened a $1 million credit pool for the model, which it projects will cost more than 50% less than comparable systems.
The release describes four stages, discovery, creation, distribution and monetization, and names no price, launch date, customers or usage figures.
Studios suing Midjourney get a file-level look at a shipped AI video model's training data under an agreed protocol, not a ruling, and their damages claim is unquantified.
The API pricing page still says its plans don't support Kling 4.0, leaving Runway and Runware to carry Kling 3.0 instead.
The advocacy group's petition to ban lifelike AI-generated depictions of people or events that never existed had more than 17,300 signatures as of October 16.
AFP's fact-check found a SynthID watermark showing Google's AI made the TikTok video, and no credible report of any such Jakarta rally.
The board now ranks audio-enabled models as AA-Video-I2V v1.1, its Elo values no longer carry over, and Vidu Q4 Preview is missing from it.
fal says the model ties for first on Artificial Analysis's image-to-video board, which lists it second, two Elo points below MiniMax H3.
Irreplaceable, a group launched in late August, wants an immediate halt to frontier AI development until people have a say, a stake and safety guardrails.
Runway and ElevenLabs retired their local MCP servers in the same week Unity shipped 32 official skills for coding agents. Every media-creation MCP server and skill pack we track, with the date each figure was read.
A Toronto company taught a machine to put a boy in a village and walk him across it, physics built in, for less than a cent a scene. I have been typing for fifty years, and I cannot do it in six weeks.
The Tech Transparency Project says test accounts downloaded 135 apps that strip clothes or generate explicit video, parental controls or not.
Builds keep appearing under it, including one from NVIDIA Research, while a rival open video model ships with no territory clause at all.
The platform keeps a ledger of every AI and human action and scans output for copyright conflicts while the work is made.
The Wall Street Journal reports a roughly 75 percent premium on Decart's May valuation, weeks after Anthropic walked away from a $6 billion deal.
The company also launched Supercomputer, an agent it says routes each production step across models from OpenAI, Anthropic and Google.
Dubai's second $1M AI Film Award closes entries on November 20, and it is the only award on this list that requires one specific video model.
SLOP TV ran no tests of its own; in week one, independent testers praised Kling 4.0's motion and dialogue and questioned its 30-second single take.