Kling4.0 by Kuaile AI: Supports 8000 Tokens and Stereo Lip Sync

On September 28, Kling AI officially announced that the newly upgraded Kling4.0 will be fully launched in October. Currently, the Kling4.0 Flash version is available for a limited number of Black Gold Annual Card members to experience first. This upgrade has made significant improvements in three core dimensions: visual realism, creative controllability, and narrative completeness, offering creators a more professional "all-around team."

Comprehensive audio-visual performance breakthroughs, approaching cinematic quality

Kling4.0 significantly improves stability and continuity when dealing with large-scale camera movements such as high-speed motion, continuous action, tracking shots, and surrounding shots. In terms of audio, it now supports high-quality audio and dual-channel stereo sound, greatly enhancing the accuracy of character lip-syncing. In terms of video output, it not only supports ultra-wide 21:9 aspect ratio and 4K resolution, but also the 10-bit HDR format at 1080p will be released soon.

Extreme multi-modal control, reshaping commercial video creation

In terms of creative control, the new version brings unprecedented freedom: it supports up to 15 multi-modal reference inputs per generation (including 10 images, 5 videos, and 7 subject combinations). It not only allows precise editing of the subject and background of the original video, but can also intelligently analyze the cinematography and narrative structure of the reference video, making it easy to generate new commercial-level videos.

Strong long-take storytelling and multi-language support

Kling4.0 now supports generating native videos up to 30 seconds in a single session, perfectly meeting the needs of long takes and continuous storytelling. With up to 10 keyframe inputs, creators can precisely control the characters' states and plot development. Additionally, the prompt input limit has been significantly expanded to 8000 Tokens, supporting multiple languages including Chinese, English, Japanese, Korean, Spanish, Portuguese, German, French, and Hindi, as well as various accents and dialects. The ability to extend videos up to 2 minutes through multiple continuations will also be released soon.

Completely renewed workflow, delivering an ultimate user experience

Anthropic released Claude Sonnet 5.5 before OpenAI's Developer Conference, half the price of Opus and even passed Pokémon

Anthropic released Claude Sonnet5.5 before OpenAI's annual developer conference, positioning it as a versatile assistant for daily work. It breaks the 'impossible triangle' of speed, cost, and quality: it is 30% faster than Sonnet5, with up to a 30% reduction in cost per task, priced at half that of Opus5.5, with performance almost matching, and outperforming its high-end models in several tests such as Terminal-Bench4.0; the article also uses games as an example to illustrate that its intelligence is sufficient.

Bet on the Physical World AI: AMD Acquires Fei-Fei Li's Team and the Spatial Large Model Atlas

AMD announced it will acquire AI startup World Labs through a stock deal, valued at about 8.2 billion US dollars. After the transaction, co-founder and Stanford University professor Fei-Fei Li will serve as AMD's Executive Vice President and Chief Scientist, working with Chairman and CEO Sue Mi Terry to advance cutting-edge AI strategies. World Labs was founded by Fei-Fei Li, Ben Mildenhall, and Justin Johnson in early 2024, aiming to break through the boundaries of traditional language models and build AI that can understand and reason about the physical world.

AI Team Restructuring Frequently Tops Hot Search: Doubao's Response to the Rumor of Cutting Half of the Conversation Team Is Untrue

Doubao trended over rumors of downsizing its General Session team, with claims of layoffs, halving the dialogue team, and marginalization. PR head Liu Xing said reports reflect organizational restructuring with some functions split, denied layoffs, and called self-media interpretations inaccurate.....

Meta Admits Rarely That Its Product Design Was Deeply Influenced by OpenClaw After Being Accused of Copying It

Nat Friedman, product director of Meta's Super Intelligence Lab, responded that the AI agent Muse was indeed influenced by OpenClaw, but was created from scratch. Previously, Ansh Nanda, co-founder of an AI application, posted that Muse was OpenClaw for ordinary users, and shared a conversation showing that the AI itself acknowledged similar system files, not coincidentally but as a match. The community also found more similarities in Muse.

AI Infrastructure Backfires: $68 Billion Data Center Projects in Q2 Face Public Opposition

Data center construction driven by AI faces strong public opposition in the United States. Research shows that about $68 billion in projects were delayed or blocked in the second quarter of this year, compared to $13 billion in the first quarter, with opposition now affecting half of the projects. Data Center Watch reported that 45 data center development projects were hindered or delayed in the second quarter.