Krea 3 to Welcome Its First Expert Model Tomorrow
Krea announced that Krea 3 will welcome its first expert model tomorrow, and said that a final clue remains in the video.
Matched fieldsSummary、Body
←Search results
Search “Video”
Gemini 4 Carbon Tops the Artificial Analysis Intelligence Index…
Artificial Analysis evaluations show that Google's Gemini 4 Carbon tops Intelligence Index v5 with 59 points, ranking first in four of the 9 evaluations: HLE-Diamond, Terminal-Bench-Science, AA-Omniscience, and AutomationBench-AA. On the Coding Agent Index, Gemini 4 Carbon (high) paired with Antigravity CLI scores 59 points, 3 points behind Claude Code paired with Claude Opus 5.5 (max, 62 points). The model is currently available only to some users and has not yet been publicly released; it is priced at $4/$20 per 1M input/output tokens with a 50% discount, has a 1M-token context window, and supports text, image, video, and voice input.
Artificial Analysis evaluations show that Google's Gemini 4 Carbon tops Intelligence Index v5 with 59 points, ranking first in four of the 9 evaluations: HLE-Diamond, Terminal-Bench-Science, AA-Omniscience, and AutomationBench-AA. On the Coding Agent Index, Gemini 4 Carbon (high) paired with Antigravity CLI scores 59 points, 3 points behind Claude Code paired with Claude Opus 5.5 (max, 62 points). The model is currently available only to some users and has not yet been publicly released; it is p
Matched fieldsSummary、Body
PixVerse showcases Kling 4.0 generation results
PixVerse released a joint generation example with Kling 4.0, noting that AI video still has shortcomings in the sense of physical impact and motion consistency, while this generation presents a heavier sense of impact and more stable details in the action, coming closer to the texture of live-action footage.
Matched fieldsSummary、Body
Maxinsights raises $30 million seed round to build a training data platform for physical AI
Maxinsights has completed a $30 million seed round led by PeakXV, with participation from Tola Capital, South Park Commons, INCE Capital and Alumni Ventures, to expand its physical AI data platform that collects real human activity data for robots and AI models. The company has collected over 2 million hours of real human activity data, of which 1.5 million hours can be used directly for model training, serving more than 30 AI lab and foundation model developer clients. The company has also launched the MaxLoop quality monitoring system and the MaxGlove tactile sensing glove; MaxLoop reduces review costs by about 80% and can process 450,000 hours of video per month.
Maxinsights has completed a $30 million seed round led by PeakXV, with participation from Tola Capital, South Park Commons, INCE Capital and Alumni Ventures, to expand its physical AI data platform that collects real human activity data for robots and AI models. The company has collected over 2 million hours of real human activity data, of which 1.5 million hours can be used directly for model training, serving more than 30 AI lab and foundation model developer clients. The company has also laun
Matched fieldsSummary、Body
Tokyo University of Science completes bladder cancer assisted surgery using XREAL One Pro AR glasses
Tokyo University of Science successfully completed a bladder cancer surgery assisted by XREAL One Pro AR glasses; the patient was a man in his 50s, and postoperative progress was smooth. The system was jointly developed by the university's Department of Renal Urology and the Department of Electrical and Electronic Engineering team, superimposing endoscopic images within the doctor's field of view and connecting to Sony's NUCLeUS platform to centrally manage operating room video signals. Other participants can wear xbx a01+ to view information synchronously.
Tokyo University of Science successfully completed a bladder cancer surgery assisted by XREAL One Pro AR glasses; the patient was a man in his 50s, and postoperative progress was smooth. The system was jointly developed by the university's Department of Renal Urology and the Department of Electrical and Electronic Engineering team, superimposing endoscopic images within the doctor's field of view and connecting to Sony's NUCLeUS platform to centrally manage operating room video signals. Other pa
Matched fieldsSummary、Body
Solink raises over $100 million to extract business data from security cameras with AI
Video intelligence company Solink has completed a new funding round of over $100 million, led by Vertu Capital with participation from Goldman Sachs through its Growth Equity arm, bringing its cumulative funding to more than $198 million. Its platform uses AI to extract information from the existing security cameras of retail, restaurant and warehouse customers; currently about 500,000 cameras are connected across more than 35,000 customer sites in 65 countries. In February this year the company launched an edge AI agent for perimeter security and loss prevention, and in March it added a vision-language model to its VerifEye command center to filter out false alarms.
Video intelligence company Solink has completed a new funding round of over $100 million, led by Vertu Capital with participation from Goldman Sachs through its Growth Equity arm, bringing its cumulative funding to more than $198 million. Its platform uses AI to extract information from the existing security cameras of retail, restaurant and warehouse customers; currently about 500,000 cameras are connected across more than 35,000 customer sites in 65 countries. In February this year the company
Matched fieldsSummary、Body
A Conversation with "Burying AI" Author Xianyu: How to Roast China's AI Industry
ChinaTalk talks with "Burying AI" author Xianyu about how he satirizes China's AI industry from a subcultural perspective. Xianyu says that since the second half of 2025 he has completely stopped using AI for writing, because while mainstream models' coding abilities have improved, the originality of their writing has clearly declined. He also built a website that lets ordinary people rank video models by their meme proficiency, and he disagrees with the host over which side, China or the US, has more severe hype.
ChinaTalk talks with "Burying AI" author Xianyu about how he satirizes China's AI industry from a subcultural perspective. Xianyu says that since the second half of 2025 he has completely stopped using AI for writing, because while mainstream models' coding abilities have improved, the originality of their writing has clearly declined. He also built a website that lets ordinary people rank video models by their meme proficiency, and he disagrees with the host over which side, China or the US, ha
Matched fieldsSummary、Body
Bites raises $40M Series A to bring AI training to frontline workers
Bites has raised a $40 million Series A led by PeakSpan Capital, bringing its total funding to $55 million. Its AI Studio can turn documents and standard operating procedures into branded training videos within minutes; each "Bite" is up to 90 seconds long, supports more than 150 languages, and is delivered via WhatsApp, SMS, Microsoft Teams and Slack. The company says customers typically achieve over 90% engagement within three weeks, with a net revenue retention rate of 161%.
Matched fieldsSummary、Body
Google DeepMind releases Gemini 4 Carbon, topping Artificial…
Google DeepMind has released Gemini 4 Carbon, which ranks first on the Artificial Analysis Intelligence Index v5 with a score of 59 and leads in 4 of 9 evaluations including HLE-Diamond and TB-Science. The model focuses on cybersecurity and long-horizon tasks, claiming 88.2% on SEC-bench Pro, 54.8% on ExploitGym, and 71.0% on CWE-bench, and is initially available only to trusted cybersecurity personnel through the Fairwind Program. In terms of specifications, it supports a 1M token context and 1M maximum output, with input covering text, images, video, and speech, and output in text; Artificial Analysis lists a standard price of $4/$20 per million input/output tokens, plus a 50% launch discount.
Google DeepMind has released Gemini 4 Carbon, which ranks first on the Artificial Analysis Intelligence Index v5 with a score of 59 and leads in 4 of 9 evaluations including HLE-Diamond and TB-Science. The model focuses on cybersecurity and long-horizon tasks, claiming 88.2% on SEC-bench Pro, 54.8% on ExploitGym, and 71.0% on CWE-bench, and is initially available only to trusted cybersecurity personnel through the Fairwind Program. In terms of specifications, it supports a 1M token context and 1
Matched fieldsSummary、Body
Iluvatar CoreX Tianyao 300 Benchmarked Against H200: Training Throughput Reaches 1.33x That of H100
Iluvatar CoreX held its first live demonstration of the Tianyao 300 at a media open day on October 28, claiming that in comparative tests on large language models such as GLM-5.3 as well as text-to-image and text-to-video scenarios, it outperformed the H200 overall. On the training side, in an 8-GPU full-system FP8-precision demonstration based on Qwen3-32B, the Tianyao 300 achieved per-GPU effective throughput 1.33 times that of the H100. The Tianyao 300 was unveiled at this year's World Artificial Intelligence Conference in July, features 144GB of memory, and its IXL144 supernode integrates 144 GPUs with roughly 10.1TB of memory in a single standard rack, reducing the cost per million effective tokens by about 40% compared with mainstream international solutions.
Iluvatar CoreX held its first live demonstration of the Tianyao 300 at a media open day on October 28, claiming that in comparative tests on large language models such as GLM-5.3 as well as text-to-image and text-to-video scenarios, it outperformed the H200 overall. On the training side, in an 8-GPU full-system FP8-precision demonstration based on Qwen3-32B, the Tianyao 300 achieved per-GPU effective throughput 1.33 times that of the H100. The Tianyao 300 was unveiled at this year's World Artifi
Matched fieldsSummary、Body
Thomas Wolf jokes: this robot deserves an Oscar
Thomas Wolf shared a robot video with the caption "give this robot an oscar", without further explanation.
Matched fieldsSummary
Odyssey: World Models Will Bring New Discoveries
Odyssey posted that world models will bring new discoveries, along with a video.
Matched fieldsSummary
Rohan Paul jokes about the "slacking" agent in multi-agent RL
Rohan Paul posted a joke about multi-agent reinforcement learning: one of the agents gets tired and wants to go eat lunch, along with a video.
Matched fieldsSummary
Utopai X Tops AA-Video-I2V v2.1 at Second Place
Utopai X, post-trained by Utopai Studios based on MiniMax H3, ranks second on the AA-Video-I2V v2.1 leaderboard, behind only Dreamina Seedance 2.5, and is the only model to simultaneously enter the top two of both AA-Video-T2V v2.1 and AA-Video-I2V v2.1. The model performs best on Physics and Multi-Scene & Narrative prompt categories as well as in the Frontier, UI/UX & Motion Design, and Productivity & Knowledge Work scenarios. Utopai X is available only on the PAI platform, at 93 PAI credits per second for 1080p video, with no public API; PAI subscriptions start at 1,000 credits for $15/month.
Utopai X, post-trained by Utopai Studios based on MiniMax H3, ranks second on the AA-Video-I2V v2.1 leaderboard, behind only Dreamina Seedance 2.5, and is the only model to simultaneously enter the top two of both AA-Video-T2V v2.1 and AA-Video-I2V v2.1. The model performs best on Physics and Multi-Scene & Narrative prompt categories as well as in the Frontier, UI/UX & Motion Design, and Productivity & Knowledge Work scenarios. Utopai X is available only on the PAI platform, at 93 PAI credits pe
Matched fieldsSummary、Body
Tripo AI Releases Real-Time Generative Renderer EdenLive and Open-Sources It
Tripo AI has launched EdenLive, a real-time generative renderer dedicated to interactive scenes, as part of its interactive environment world model Project Eden, and it is now open-sourced to developers. EdenLive decouples scene evolution logic from visual rendering; rendering can be started with just a 2D/3D scene or reference video, a target first frame, and text instructions, and it supports both 480p and 720p resolutions; the 5-billion-parameter version can achieve real-time generation at over 24 frames per second at 720p on a single consumer-grade RTX 5090. The open-source release includes the 5-billion and 14-billion parameter model weights, real-time video generation code, the EdenGolem v0.1 skill for converting text into simple games, and a complete prompt-to-game front-end toolchain; the code is released on GitHub under the Apache 2.0 license, and the weights are available on Hugging Face.
Tripo AI has launched EdenLive, a real-time generative renderer dedicated to interactive scenes, as part of its interactive environment world model Project Eden, and it is now open-sourced to developers. EdenLive decouples scene evolution logic from visual rendering; rendering can be started with just a 2D/3D scene or reference video, a target first frame, and text instructions, and it supports both 480p and 720p resolutions; the 5-billion-parameter version can achieve real-time generation at ov
Matched fieldsSummary、Body
Ant inclusionAI Open-Sources the UBTree Project Page: Using Unigram and…
Ant inclusionAI has launched the UBTree project page on GitHub, introducing a speculative decoding method that uses Unigram and Bigram models for parallel tree drafting (arXiv:2609.39972). The page includes demo videos of Ling3 Tiny and Flash, comparing four methods—AR, DFlash, DSpark, and UBTree—under the same target model; the videos are 1440×760, 20 FPS, at 0.25× playback speed. The figures and table numbers are all driven by static/js/data.js.
Matched fieldsSummary、Body
Ant Group's inclusionAI Open-Sources 2B Visual Forensics Model DefakerOne2
Ant Group's GuangJian team has open-sourced DefakerOne2 on GitHub, a 2B-parameter unified visual forensics model built on the Qwen3.5-2B multimodal backbone, covering AI-generated content, DeepFake faces, document tampering, natural image tampering, face and ID anti-counterfeiting, and AI-generated video detection. The model was trained on approximately 50 million samples, employs a three-stage Tide-Style Tuning process from LoRA to full parameters and back to LoRA, and uses entropy-based test-time routing to select between the video and frame pathways. The team claims that it achieves SOTA on 40 of 53 detection benchmarks and on 11 of 14 localization benchmarks, and that the weights and the RRVBench benchmark have been released on ModelScope.
Ant Group's GuangJian team has open-sourced DefakerOne2 on GitHub, a 2B-parameter unified visual forensics model built on the Qwen3.5-2B multimodal backbone, covering AI-generated content, DeepFake faces, document tampering, natural image tampering, face and ID anti-counterfeiting, and AI-generated video detection. The model was trained on approximately 50 million samples, employs a three-stage Tide-Style Tuning process from LoRA to full parameters and back to LoRA, and uses entropy-based test-t
Matched fieldsSummary、Body
SenseTime Open-Sources SenseNova 6.8 Code2Video Model and Skill
SenseTime has released and open-sourced SenseNova 6.8 Code2Video, a lightweight model that creates and renders short videos by generating executable code, and has already achieved preliminary results in visual composition and motion design. The accompanying Skill supports turning ideas into videos and modifying them through follow-up instructions, and can also reconstruct reference videos into editable code. SenseTime has also open-sourced the model weights, the Skill, and the prompts and complete generation trajectories behind each demo video; these videos were all generated in a single run from a single prompt.
SenseTime has released and open-sourced SenseNova 6.8 Code2Video, a lightweight model that creates and renders short videos by generating executable code, and has already achieved preliminary results in visual composition and motion design. The accompanying Skill supports turning ideas into videos and modifying them through follow-up instructions, and can also reconstruct reference videos into editable code. SenseTime has also open-sourced the model weights, the Skill, and the prompts and comple
Matched fieldsSummary、Body
OpenAI Releases GPT-6.1 Terra, Priced 50% Lower Than the Previous Version
OpenAI has released a new model, GPT-6.1 Terra, priced at $1 per million input tokens and $5 per million output tokens, 50% cheaper than GPT-6.1 Sol's $2 and $10. OpenAI says the model is fast and efficient, combining visual reasoning with precise computer operation, and excels at tasks such as drafting proposals, filling out forms, and editing videos. This is OpenAI's fourth consecutive model price cut, and Anthropic also cut the price of Haiku 5.5 by 90% in October.
Matched fieldsSummary、Body
iOS 27.2 Upgrades Apple Photos: Shared Albums Add Four Features and Support for SynthID Watermarks
iOS 27.2 brings two upgrades to Apple Photos: Shared Albums add four features—editing photo and video captions, editing comments, preventing duplicate photo additions, and reminders before an album expires; it also adds support for the SynthID watermark developed by Google, which can identify whether an image was generated or modified by generative tools in Photos and Image Playground. The SynthID watermark is invisible but detectable. Using the new Shared Albums features requires first upgrading the album to iOS 27.
iOS 27.2 brings two upgrades to Apple Photos: Shared Albums add four features—editing photo and video captions, editing comments, preventing duplicate photo additions, and reminders before an album expires; it also adds support for the SynthID watermark developed by Google, which can identify whether an image was generated or modified by generative tools in Photos and Image Playground. The SynthID watermark is invisible but detectable. Using the new Shared Albums features requires first upgradin
Matched fieldsSummary、Body