English translation is being prepared.
Matched fieldsTitle、Summary
←Search results
Search “Video”
English translation is being prepared.
Matched fieldsTitle、Summary
NVIDIA Releases Four Open-Weight World Models in the Cosmos 3 Series
Under the Cosmos world model platform, NVIDIA has added four Cosmos 3 open-weight checkpoints, all post-trained versions of Cosmos 3 Nano, containing 16B trainable parameters and released under the OpenMDW1.1 license, which permits commercial and non-commercial use. The four models respectively target multi-view driving data generation, synchronized streaming simulation of cameras and LiDAR, simulator-controlled generation of photorealistic video, and video prediction from dual-arm robot actions. The original text also provides a four-stage training recipe covering bidirectional adaptation, teacher forcing, consistency distillation, and self-forcing.
Under the Cosmos world model platform, NVIDIA has added four Cosmos 3 open-weight checkpoints, all post-trained versions of Cosmos 3 Nano, containing 16B trainable parameters and released under the OpenMDW1.1 license, which permits commercial and non-commercial use. The four models respectively target multi-view driving data generation, synchronized streaming simulation of cameras and LiDAR, simulator-controlled generation of photorealistic video, and video prediction from dual-arm robot actions
Matched fieldsSummary、Body
Manus Studio now supports export to editing software including Premiere Pro
Manus announced that Manus Studio projects can now be exported to Premiere Pro, DaVinci Resolve, CapCut, or Final Cut Pro. Users can either create and edit directly in Manus with AI, or export videos to their preferred tools for editing.
Matched fieldsSummary、Body
Generalist demonstrates GEN-1.5's fine manipulation capabilities
Generalist released a video showing GEN-1.5 completing a thinner O-ring manipulation task, stating that such tasks test the model's dexterity and its ability to perceive millimeter-level details. The post also directs users to read the GEN-1.5 details and join the waitlist.
Matched fieldsSummary
MiniMax H3 reenacts Chinese classical allusions with faux archival footage
MiniMax Design (H3) posted that ancient Chinese stories and idioms are often hard to explain, but H3 has found "archival footage." The post cited user 0xMomo's tutorial on making a "Sima Guang smashing the vat surveillance recording" with MiniMax H3 (ComfyUI) on RunningHub, which covered generating the video in segments, describing a fixed camera position, and the workflow of creating low-resolution dot-matrix timestamps with VCR OSD Mono and GNU Unifont, then overlaying them with FFmpeg.
Matched fieldsSummary
MiniMax H3 scores full marks in both speaking and writing on language tests
MiniMax Design (H3) posted that H3 achieved full marks in both speaking and writing on language tests. The post cited a case in which a user created an original shopping street video using Blender previsualization combined with H3, noting that H3 had the strongest ability to preserve Japanese signage.
Matched fieldsSummary
Krea Releases Krea 3 Technical Report
Krea releases the Krea 3 technical report, providing an in-depth introduction to the data, architecture, and training techniques used by the model. The official tweet provides a link to the report and includes a video.
Matched fieldsSummary
Hao AI Lab integrates MiniMax H3 Flex into FastVideo, delivering up to 7x inference speedup
Hao AI Lab integrated Pika Labs' open-source MiniMax H3 Flex into FastVideo, achieving inference speeds up to 7 times that of the official open-source inference code. On a single GB200, generating a 5-second 480p video with audio using the same prompt and random seed took 59.7 seconds with the official code, 26.6 seconds with FastVideo BF16, 16.8 seconds with NVFP4 QAD, and 15.2 seconds with FP4 attention. The official code only supports a single GPU; after FastVideo added multi-GPU support, generating a 15-second video on 4 GPUs was 7 times faster; generating a 5-second 480p first-and-last-frame video on a single DGX Spark took 59.5 seconds. The project also releases the weights, inference code, and complete training recipe, and supports the 2K model of Open-H3-Regenerate-2K.
Hao AI Lab integrated Pika Labs' open-source MiniMax H3 Flex into FastVideo, achieving inference speeds up to 7 times that of the official open-source inference code. On a single GB200, generating a 5-second 480p video with audio using the same prompt and random seed took 59.7 seconds with the official code, 26.6 seconds with FastVideo BF16, 16.8 seconds with NVFP4 QAD, and 15.2 seconds with FP4 attention. The official code only supports a single GPU; after FastVideo added multi-GPU support, gen
Matched fieldsSummary、Body
Hao AI Lab integrates MiniMax H3 Flex into FastVideo, single...
Hao AI Lab has integrated Pika Labs' open-source MiniMax H3 Flex into FastVideo. The official open-source inference code takes 59.7 seconds to generate a 5-second 480p video with audio on a single GB200, while FastVideo BF16 takes 26.6 seconds, NVFP4 QAD takes 16.8 seconds, and FP4 attention takes 15.2 seconds. The team claims it is up to 7 times faster than the official code, and has added multi-GPU support, generating a 15-second clip 7 times faster on 4 GPUs. NVFP4 QAD distills the BF16 Flex model, coming 14-17% closer to BF16 than ordinary quantization; FP4 attention reduces a 768p 15-second clip from 101.8 seconds to 85.3 seconds. This clip takes 59.5 seconds on a single DGX Spark. FastVideo also supports the 2K model of Open-H3-Regenerate-2K, and the weights, inference code, and complete training recipe have all been released.
Hao AI Lab has integrated Pika Labs' open-source MiniMax H3 Flex into FastVideo. The official open-source inference code takes 59.7 seconds to generate a 5-second 480p video with audio on a single GB200, while FastVideo BF16 takes 26.6 seconds, NVFP4 QAD takes 16.8 seconds, and FP4 attention takes 15.2 seconds. The team claims it is up to 7 times faster than the official code, and has added multi-GPU support, generating a 15-second clip 7 times faster on 4 GPUs. NVFP4 QAD distills the BF16 Flex
Matched fieldsSummary
FastVideo Supports the Open-H3-Regenerate-2K Model
Hao AI Lab announced that FastVideo now supports the Open-H3-Regenerate-2K model from HaoZhe65347 and NVIDIA AI, and provided a link to the 2K model. The series previously covered FastVideo's inference optimization of MiniMax H3 Flex, open-sourced by Pika Labs: on a single GB200, generating a 5-second 480p video with audio takes 5.8 seconds, up to 7 times faster than the official open-source inference code, with newly added multi-GPU support.
Matched fieldsSummary
Hao AI Lab Integrates MiniMax H3 Flex into FastVideo, Single-GPU…
Hao AI Lab announced that Pika Labs' open-source MiniMax H3 Flex has been integrated into FastVideo. The team claims its inference speed is up to 7 times faster than the official open-source inference code, generating a 5-second 480p video with audio in 5.8 seconds on a single NVIDIA GB200. Under the same prompt and random seed, the official code takes 59.7 seconds for a 5-second clip with audio, while FastVideo BF16 takes 26.6 seconds, NVFP4 QAD takes 16.8 seconds, and FP4 attention takes 15.2 seconds; the official code only supports a single GPU, whereas FastVideo adds multi-GPU support, making a 15-second clip 7 times faster on 4 GPUs. NVFP4 QAD uses BF16 Flex as its distillation target, coming 14-17% closer to BF16 than ordinary quantization; FP4 attention reduces a 768p 15-second clip from 101.8 seconds to 85.3 seconds. A 5-second 480p first-and-last-frame clip takes 59.5 seconds on a single DGX Spark, and it supports the Open-H3-Regenerate-2K model from Hao AI Lab and NVIDIA. The weights, inference code, and complete training recipe have all been released.
Hao AI Lab announced that Pika Labs' open-source MiniMax H3 Flex has been integrated into FastVideo. The team claims its inference speed is up to 7 times faster than the official open-source inference code, generating a 5-second 480p video with audio in 5.8 seconds on a single NVIDIA GB200. Under the same prompt and random seed, the official code takes 59.7 seconds for a 5-second clip with audio, while FastVideo BF16 takes 26.6 seconds, NVFP4 QAD takes 16.8 seconds, and FP4 attention takes 15.2
Matched fieldsSummary
Apple's redesigned Health app expands language support
Apple's redesigned Health app expands language support to global English, Simplified and Traditional Chinese, French, Japanese, Korean, Norwegian, Portuguese, Spanish, Turkish, and Vietnamese, with Apple Intelligence at its core. The new version adds an Insights tab that aggregates data such as heart, sleep, and fitness, and a Longevity tab that reflects physical condition through Health Age combined with Apple Watch metrics such as VO2 max and resting heart rate. The update also introduces at-home movement assessments, using the iPhone camera and Apple Watch to measure flexibility, strength, balance, and VO2 max on-device, without recording or storing video. This version enters the iOS 27.3 developer beta today and will be released for free later this year.
Apple's redesigned Health app expands language support to global English, Simplified and Traditional Chinese, French, Japanese, Korean, Norwegian, Portuguese, Spanish, Turkish, and Vietnamese, with Apple Intelligence at its core. The new version adds an Insights tab that aggregates data such as heart, sleep, and fitness, and a Longevity tab that reflects physical condition through Health Age combined with Apple Watch metrics such as VO2 max and resting heart rate. The update also introduces at-h
Matched fieldsSummary、Body
What Is Call Transcription? A Practical Guide for Businesses
Call transcription is the process of converting phone or video calls into written records, in most cases done automatically by AI models, either transcribed in real time during the call or batch-transcribed after recording. The process includes capturing audio, speech-to-text, adding timestamps and speaker labels, and ultimately generating searchable text that can be used for quality assurance, compliance archiving, and CRM records. Voice AI agents built with ElevenAgents save a transcript with every conversation.
Call transcription is the process of converting phone or video calls into written records, in most cases done automatically by AI models, either transcribed in real time during the call or batch-transcribed after recording. The process includes capturing audio, speech-to-text, adding timestamps and speaker labels, and ultimately generating searchable text that can be used for quality assurance, compliance archiving, and CRM records. Voice AI agents built with ElevenAgents save a transcript with
Matched fieldsSummary、Body
NVIDIA Releases Cosmos Framework and Cosmos NIM, Supporting World Model Training and Deployment
NVIDIA has released the Cosmos Framework and Cosmos NIM for physical AI. The former provides modular tools and agent skills for training, reinforcement learning, optimization, and inference, while the latter provides accelerated inference services in container form. The Cosmos 3 Nano simulation recipe can generate synchronized camera video and LiDAR observations from released checkpoints, covering autonomous driving, robotics, and surgical scenarios. NIM uses quantization, speculative decoding, and KV-cache-aware routing to achieve a measured speedup of about 2–3x on H200, and Cosmos3-Nano generation tasks require 58 GiB of VRAM under BF16.
NVIDIA has released the Cosmos Framework and Cosmos NIM for physical AI. The former provides modular tools and agent skills for training, reinforcement learning, optimization, and inference, while the latter provides accelerated inference services in container form. The Cosmos 3 Nano simulation recipe can generate synchronized camera video and LiDAR observations from released checkpoints, covering autonomous driving, robotics, and surgical scenarios. NIM uses quantization, speculative decoding,
Matched fieldsSummary、Body
MiniMax H3 and ORB360 LoRA Enable Interactive Rotation
MiniMax Design reposted user monoradio's demo: using MiniMax H3 plus ORB360 LoRA, a 360° rotation video is generated from a single image, and then frames are switched back and forth as the phone tilts, creating a stereoscopic effect. MiniMax Design says this combination has already attracted attention, and now goes a step further to enable interaction, hinting that VR is the next direction.
Matched fieldsSummary
Meshy generates the 3D model, Blender sets the scene, Houdini sets it on fire
The official Meshy account posted a video showcasing a 3D artist's Halloween farewell piece: the model was generated in Meshy, the scene was set in Blender, and it was set on fire in Houdini.
Matched fieldsSummary
Krea plugin enables turning sketches into cartoon animations within Grok
Krea announced that with the Krea plugin, users can turn sketches into cartoon animations directly in Grok simply by making a request to Grok. Previously, Krea had launched a plugin for the Grok bot that can generate images and videos within Grok.
Matched fieldsSummary
State of AI 2026: Nathan Benaich on AGI in 2027, Scientific Superintelligence, and World Models
Nathan Benaich of Air Street Press discussed the State of AI Report 2026 with Matt Turck on the MAD Podcast, reiterating his prediction that AGI will be achieved in 2027, based on AI's ability to approach his own level in knowledge work such as investing, AI research, finance, and law. He also discussed how images, video, and audio are converging into world models, as well as the cybersecurity risks posed by agents; ElevenLabs' agents already handle more than 15 million conversations per week, three times the volume in February.
Nathan Benaich of Air Street Press discussed the State of AI Report 2026 with Matt Turck on the MAD Podcast, reiterating his prediction that AGI will be achieved in 2027, based on AI's ability to approach his own level in knowledge work such as investing, AI research, finance, and law. He also discussed how images, video, and audio are converging into world models, as well as the cybersecurity risks posed by agents; ElevenLabs' agents already handle more than 15 million conversations per week, t
Matched fieldsSummary、Body
SenseTime's Sekor Generates Halloween Short Film, Creator's Works Surpass 74 Million Views
SenseTime used its AI video agent Sekor to produce a Halloween-themed short film. The creator is 22-year-old CHunye, whose short drama series has accumulated over 74 million views. The post also provides the accounts of CHunye and Sekor on Douyin, YouTube, IG, and X, along with a link to create with Sekor.
Matched fieldsSummary、Body