This AI news roundup for August 3-4, 2026, covers significant model releases from Alibaba Qwen, NVIDIA, Mistral AI, and Black Forest Labs, alongside developments in agent routing, infrastructure, and cybersecurity.
MiniMax has released MiniMax H3, a general-purpose multimodal generation model that unifies text, image, video, and audio into a single context to produce 2K video clips up to 15 seconds long with native stereo sound. Unlike previous stacks that rely on separate expert models for specific tasks, H3 allows users to express reference and editing relationships through natural language prompts.
xAI has updated its Imagine Video 1.5 model to support text, image, and voice references, enabling users to generate videos up to 1080p resolution. The update allows for character consistency by locking faces and voices across scenes, with support for up to seven references per generation.
MiniMax has launched MiniMax H3, a general-purpose multimodal generation model that understands unified context across text, images, video, and audio. The model generates video with native stereo sound at up to 15 seconds duration and 2K resolution.
NVIDIA has introduced Cosmos-H-Dreams, a real-time, action-conditioned generative simulator for surgical robotics that distills the capabilities of its Cosmos-H-Surgical-Simulator into a causal student model. Served through NVIDIA's FlashDreams accelerated streaming-inference library, the system enables interactive control by a person or learned policy in a closed loop.
Black Forest Labs has released FLUX 3, a multimodal foundation model that learns from images, videos, and audio within a single architecture. It is the first FLUX model to ship video, audio, and action prediction from one set of weights.
Black Forest Labs has launched FLUX 3, a multimodal flow model family that includes video generation capabilities with native audio. The release also introduces FLUX-mimic, a variant designed for video-action robotics applications.
NVIDIA and Hugging Face have integrated NVIDIA NeMo Automodel with the 🤗 Diffusers library to provide production-grade, distributed training for open-source diffusion models. This collaboration allows users to fine-tune any Diffusers-format model on the Hugging Face Hub without requiring checkpoint conversion or model code rewrites.
Google has rolled out two new updates to Google Vids: the integration of Gemini Omni for text-based video generation and editing, and a feature allowing users to create personal digital avatars. These tools aim to simplify video creation by enabling users to generate clips from natural language prompts and image references, as well as star in videos without physical recording.
A viral video generated using Kimi K3 as the model provider has been shared, showcasing its capabilities in creative tasks. The creator notes that the output quality feels significantly better than previous examples using GLM 5.2.
Researchers introduce a structure-aware framework for music-driven dance generation that models choreography as a sequence of atomic movements rather than continuous signals.
A video analytics AI agent capable of perceiving, reasoning, and acting on massive amounts of video footage must be integrated with existing workflows to be useful. These workflows include content management systems, messaging platforms, databases, ticket queues, and escalation paths.
Researchers present Text2Sign, a text-conditioned diffusion model designed to generate short sign-language clips on a single NVIDIA L4 GPU, addressing the high costs associated with training and evaluating video diffusion models. The system combines a frozen vision-language text encoder with a 3D encoder-decoder and factorized spatiotemporal attention to reduce computational requirements while preserving motion coherence.
Developers building video analytics applications across large spaces must track the same object as it moves between camera views. Single-camera 2D tracking lacks reliable depth information and typically loses track of the object when it leaves the frame, limiting applications such as warehouse safety, retail analytics, and smart-building monitoring.
Researchers identify the "seriality gap" in video diffusion models, where performance degrades as causal chains lengthen during multi-ball hard-sphere dynamics simulations. Controlled experiments show that this degradation is caused by dependent-event structures rather than video length, as it disappears in single-ball controls without interactions.
NVIDIA demonstrates that using autonomous coding AI agents to post-train the Cosmos 3 vision reasoning model can push accuracy above 90% with minimal manual effort.
Researchers have introduced Wan-Dancer, a hierarchical framework that generates long-duration, high-definition dance videos synchronized with music. The system overcomes the typical 20-second limitation of diffusion models by decoupling generation into global keyframe planning and local temporal refinement.
Developer MorphLand has released a free Steam title called "Simulation Simulator," which integrates a local large language model to serve as an in-game NPC. The game is a freeform conversation experience where players attempt to convince their AI companion that reality is a simulation.