144 articles
Google's Gemini now lets macOS users dictate, edit, and generate visuals by voice in any active window, making multimodal AI a daily workflow tool.
Google DeepMind's Gemini Robotics 2 lets robots walk, grab, and collaborate, transferring skills to new robot bodies in just hours.
MiniCPM-o 4.5 delivers real-time audio-visual AI on a single consumer GPU, letting the model listen, watch, and speak simultaneously without waiting for.
Black Forest Labs confirmed FLUX 3 Dev will be open-weight, but no release date, parameter count, or license terms have been announced yet.
Google's Gemini Robotics ER 2 watches continuous video to reason over physical tasks in real time, hitting 91.3% accuracy at identifying critical moments.
Speech emotion recognition can now detect feelings from voice tone alone, with deep learning pushing accuracy high enough for real use in therapy and.
AI is now listening to therapy sessions in real time, analyzing speech and facial cues to detect depression, PTSD, and relapse risk before clinicians can.
The best AI vision models aren't the most capable ones; they're the ones built for workflows where data is genuinely spread across formats.
AGIBOT's WITA-Omni beat Google Gemini and Alibaba on audio-visual AI, hitting 85.21% accuracy by unifying perception, speech, and movement in one model.
ClinFusion, a domain-specific medical vision model, outperforms general-purpose AI on 20 of 24 clinical benchmarks, with radiologists ranking its reports.
AI surveillance is growing 18% yearly, hitting $10.88 billion by 2032, driven by edge AI cameras, smart cities, and a shift to real-time threat detection.
Kling AI's image-to-video platform generates 4K clips in under a minute, and a $20 billion spin-off signals it is reshaping how video gets made.
FLUX 3 unifies video generation and robot control in one AI model, already running production tasks at Audi factories with no permanent performance.
Ropedia raised $30 million to scale wearable cameras that capture synchronized multimodal data, cutting physical AI training costs by 50 times.
Sony will fund up to $100,000 per project for university AI research targeting audio-visual integration, with applications opening July 15, 2026.
Alibaba's two-tier video stack pairs the closed, leaderboard-winning HappyHorse with open-source Wan 2.7, targeting both quality and cost as Sora exits.
Google's Gemini is proving surprisingly effective in construction and manufacturing, where vision language models offer real-time visual guidance for.
AI avatar video generators are splitting into two specialized tools: one built for fast social ads, another for long-form content, and each delivers.
Black Forest Labs launches FLUX 3 video generation, beating Runway Gen-4.5 in 77% of tests, but withholds pricing and full benchmarks.
OpenAI's Presence platform resolves 75% of enterprise calls without human help by prioritizing control and guardrails over raw AI intelligence.
Smarter routing in audio-visual AI lets a 35-billion-parameter model specialize by modality, boosting efficiency without ballooning compute costs.
AI pet robots need both vision and voice fused together; without multimodal intelligence, they feel incomplete and fail to form real emotional bonds.
Sonilo's Sound Effects 1.0 watches your video and auto-generates synchronized audio, ending the manual sound design grind for AI video creators.
Seedance 2.0 now owns 58 percent of video generation after Sora's shutdown, making it the new default for AI video across 1.68 million users.
VLAW, a vision-language-action model at ICML 2026, lets AI continuously learn from real-world interaction instead of staying frozen after training.
ToolSciVer uses vision-language models to verify scientific claims with 81% accuracy, cutting AI token use by 51% and flagging errors human peer reviewers.
Choosing the best video generation model can sink your project; durability, input types, and API lifespan matter more than benchmark scores.
Vision language models can now listen and respond mid-conversation in real time, with OpenAI, Meta, and open-source teams all shipping streaming.
AI agents are failing enterprises not because models are weak, but because retrieved visual assets like diagrams and recordings are often years out of.
Multimodal AI is exploding toward a $280 billion market by 2035, reshaping content creation and coding as 93 percent of creators say it accelerates their.
Thinking Machines Lab released Inkling, an open-weight multimodal AI model processing text, audio, and video using just one-third the compute of rival.
AI video generation now costs small businesses under $5 per clip, replacing stock footage by animating their own product photos with authentic.
Former DeepMind researcher Andrew Dai raised $55M at a $300M valuation for Elorian, a visual AGI startup with no product, betting vision AI is the next.
AI is transforming healthcare simulation debriefing by automatically detecting communication gaps, leadership failures, and safety threats from.
ByteDance's Seedance 2.5 generates 30-second videos in a single pass via API, but unresolved Hollywood copyright disputes cloud enterprise adoption.
AI-powered microphones that fuse audio and visual data are eliminating technician labor at live events and classrooms across Southeast Asia.
Nvidia's Jetson Thor T3000 and T2000 edge AI chips will bring vision language model power to robots at half the size and cost of current hardware.
Vision language models fail dramatically when tested on baby head-camera footage, revealing that infants possess learning mechanisms today's AI cannot.
A solo director built a feature film for $50,000 using AI video tools, costing 5,000 times less than Nolan's rival Odyssey adaptation releasing the same.
PixVerse raised $439M to build an AI game engine that generates worlds in real time, targeting a $500B gaming market after Sora's costly shutdown.
Google scrapped months of Gemini 3.5 Pro development to restart from scratch after the model failed recursive tool-calling, a core agentic AI requirement.
Gemini Omni Flash accepts text, images, audio, and video at once, outputting synchronized video and audio in a single pass without separate tools.
Google is adding a privacy toggle to disable Gemini Live's camera vision feature, giving users direct control over when AI can see their surroundings.
Image generators score as low as 21 out of 100 on real-world knowledge, so researchers built a search-aware framework that teaches them when to look.
AI models count objects in long videos with just 24% accuracy versus 83% for humans, revealing deep flaws in how they search and track video evidence.
Multimodal AI is reshaping healthcare and entertainment, from detecting Parkinson's via speech patterns to powering a $26 billion interactive microdrama.
AirflowAttack fools infrared vision language models with fake thermal noise, hitting a 48.5% success rate and transferring between AI systems at up to.
Dentists are ditching generic AI chatbots over HIPAA risks and accuracy gaps, as dental-specific models outperform tools ten times their size.
Vision language models hallucinate physics, but VAORA fixes this with dual reward signals that ground reasoning in reality, achieving zero-shot transfer.
AI video calls now feel natural: Wan-Streamer v0.2 delivers sharper real-time video under 550ms latency, solving audio-visual AI's toughest trade-off.
