Skip to main content
ZICQ

Topics Multimodal & Video Generation

Multimodal & Video Generation

Image, video, and speech generation plus production practice: Sora, Veo, Kling, Suno, and the open-source multimodal stack.

· Sep 15, 2026 #Multimodal AI#video generation#Multimodal#image generation#speech generation#voice cloning#Sora#Veo#Kling#Suno · 30 / 638 articles

News · 2026-08-07

PRISM: Priority-Aware Rubric Internalization via Structured Multimodal Data Synthesis

A new arXiv paper introduces PRISM, a four-stage data synthesis framework that enables multimodal large language models (MLLMs) to follow prioritized rubrics. With only 10K synthetic samples, PRISM boosts Qwen3-VL-4B's strict accuracy on PRISM-Eval from 9.5% to 30.1% while preserving general performance. The gains transfer to four other open-source MLLMs, suggesting a scalable path toward multi-rule, priority-aware multimodal instruction following.

News · 2026-08-07

CGTime: Decoupling Perception from Description, 4B Computation-Grounded Time-Series Language Model Outperforms GPT-4o-mi

A new arXiv study introduces CGTime, a 4B-parameter computation-grounded time-series language model that decouples perception from description. By using deterministic code to compute statistics from multivariate time series and letting an LLM verbalize these facts, it overcomes the self-supervision trap and single-variable limitations. On multivariate understanding tasks, CGTime achieves a multivariate fact score of 0.283, outperforming GPT-4o-mini (0.173) and GPT-5.4-nano (0.203), with signific

News · 2026-08-07

RIG-RoPE: Relation- and Instance-Gated Rotary Positional Encoding for Multimodal LLMs

A new arXiv paper introduces RIG-RoPE, a relation- and instance-gated rotary positional encoding mechanism for multimodal LLMs. It mitigates spatial interference by applying spatial rotations only within the same visual instance, and uses duration-aware temporal coordinates to account for varying information density across modalities. RIG-RoPE adds no learned parameters and can be implemented in tiled attention kernels, offering a theoretically grounded improvement for multimodal position encodi

News · 2026-08-07

SKILLTRACE: Multi-Trace Provenance Auditing for LLM-Agent Skill Reuse

A new arXiv study presents SKILLTRACE, a multi-trace provenance auditing framework for LLM-agent skill reuse. It extracts three provenance traces: Expression, Implementation, and Operational, with the Operational Trace represented as a Skill Operational Graph (SOG). On SKILLTRACE-BENCH, it achieves AUROC 0.938 and F1 0.898, and a wild audit of 36,446 skills surfaces actionable review queues beyond repository-level baselines. This work offers a new paradigm for auditing skill reuse in agent ecosy

News · 2026-08-06

Suno Launches AI Music Watermarking to Combat Junk Music

Suno has announced the implementation of a new AI music watermarking technology aimed at combating the spread of junk AI music and increasing transparency. This move comes as the company faces legal challenges and seeks to provide a more reliable solution for music creation and distribution. The watermarking technology will help identify AI-generated music and prevent unauthorized copying and misuse, thereby enhancing trust in the AI music industry.

News · 2026-08-06

Suno Unveils AI Music Watermarking to Combat Spam and Enhance Transparency

Suno has announced plans to implement a new watermarking technology and download policy aimed at curbing the spread of spammy AI-generated music and increasing transparency. In a detailed blog post, CEO Mikey Shulman outlined the company's principles and next steps, including partnering with distribution platforms to combat fraud and misuse, and rolling out new transparency tools aligned with emerging industry standards.

News · 2026-08-06

OncoTriad-QA: A Patient-Level Radiology-Pathology-Genomics Benchmark for Pan-Cancer Reasoning

The arXiv paper introduces OncoTriad-QA, a patient-level radiology-pathology-genomics benchmark for pan-cancer question answering. It contains 86.1k semantic questions across 9,281 TCGA patient cases from 32 cancer cohorts, aligning CT/MRI, whole-slide histopathology, somatic mutations, copy-number alterations, DNA methylation, RNA-seq, and clinical metadata. The authors also present OncoVLM, a reference multimodal model that maps modality-native evidence into an LLM interface via learned projec

News · 2026-08-06

Suno Announces Watermarking and Download Policies to Curb Spammy AI Music

AI music generation company Suno announced plans to implement new watermarking technology and download policies to limit the spread of spammy AI tracks and increase transparency. CEO Mikey Shulman outlined the company's principles and next steps in a blog post, including new transparency tools, watermarking and fingerprinting tech, and partnerships with distribution platforms to combat fraud and misuse. This move aims to address the flood of AI music on streaming platforms and aligns with emergi

News · 2026-08-06

Suno Launches Song Watermarking Amid Legal Battles to Strengthen AI Music Provenance

AI music generation company Suno has announced it will add watermarks to its generated songs, responding to multiple copyright lawsuits. The feature aims to provide content provenance and help identify AI-generated music, though technical details remain undisclosed. This move is a significant step for the AI music industry seeking compliance amid copyright disputes, but may spark debates over watermark effectiveness and privacy.

News · 2026-08-04

Mistral AI Releases Shieldstral: A 3B Multimodal Safety Classifier Outperforming Models 7x Its Size

Mistral AI has introduced Shieldstral, an open-source multimodal safety classifier with a mere 3 billion parameters that outperforms models up to seven times its size. Designed to enhance AI system safety, Shieldstral is particularly suited for applications requiring real-time processing, such as autonomous driving and medical diagnostics. This release marks a significant advancement in AI safety, offering a new solution for building more reliable and efficient AI systems.

News · 2026-08-02

Hugging Face Releases FactorJEPA: Enhancing World Modeling for Complex Urban Environments

Hugging Face has introduced FactorJEPA, a novel world model architecture designed to address the limitations of existing models in handling complex, crowded, and chaotic urban environments, such as those in the Global South. By decomposing the world structure into layout, entities, and interactions, and introducing visibility gating and separate subspaces to preserve partially observed agent information, FactorJEPA significantly improves future prediction accuracy, causal reasoning, and robustne

News · 2026-07-31

xAI Releases Imagine Video 1.5: High-Definition Video Generation with Multi-Modal References

xAI has launched Imagine Video 1.5, an advanced video generation model that now supports text, image, and voice references, capable of producing videos up to 1080p resolution. This enhancement signifies a significant advancement in AI-driven multimodal video generation, empowering creative professionals with more powerful tools and improving the quality and flexibility of video content creation.

News · 2026-07-31

MiniMax Launches H3: A Groundbreaking Omni-Modal Generation Model Bridging Task and Modality Boundaries

MiniMax has officially launched MiniMax H3, a next-generation general-purpose omni-modal generation model. The model can jointly understand multimodal contexts, including text, images, video, and audio, and generate videos with native stereo audio at up to 2K resolution and 15 seconds in length. This represents a significant advancement in AI's ability to bridge the gap between different modalities and modalities and tasks, offering powerful new tools for creative content generation and cross-mo

News · 2026-07-30

Google DeepMind Unveils Gemini Robotics ER 2: Advancing Video Understanding, Task Orchestration, and Multi-Robot Collabo

Google DeepMind has officially released Gemini Robotics ER 2, a multimodal AI model designed for robotic applications, aiming to enhance robots' reasoning, collaboration, and task-solving capabilities in the real world. The model represents a step change in video understanding, tool orchestration, and multi-robot collaboration, providing a stronger closed loop from perception to action. This article analyzes its technical highlights, application scenarios, and potential impact on the robotics fi

News · 2026-07-29

xAI Releases Grok Voice Think Fast 2.0: Next-Gen Speech-to-Speech Model Launched

xAI has launched Grok Voice Think Fast 2.0, a next-generation speech-to-speech model designed to deliver enhanced voice generation capabilities. The new model features significant improvements in voice quality, generation efficiency, and naturalness compared to its predecessor, offering a superior solution for voice interaction applications.

News · 2026-07-28

Google DeepMind Unveils Gemini Robotics 2: Whole-Body Intelligence for Robots

Google DeepMind has officially released Gemini Robotics 2, the latest iteration of its robotics AI model, designed to endow robots with 'whole-body intelligence.' The model achieves significant breakthroughs in visual understanding, motor control, dexterous manipulation, and task generalization, enabling robots to interact with the physical world more naturally and safely. Built on the Gemini family of large models, Gemini Robotics 2 supports multimodal inputs, processing visual, linguistic, and

News · 2026-07-21

Google DeepMind Unveils Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber

Google DeepMind has officially introduced three new Gemini models: Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber. Gemini 3.6 Flash, the flagship flash model, delivers significant improvements in inference speed and multimodal capabilities; 3.5 Flash-Lite focuses on lightweight efficiency for edge devices and low-cost scenarios; 3.5 Flash Cyber is tailored for cybersecurity applications, enhancing threat detection and defense. This release enriches the Gemini ecosystem, offering developer

News · 2026-06-16

xAI Releases Grok Imagine Video 1.5: Enhanced Quality and Speed in Video Generation

xAI has announced the release of Grok Imagine Video 1.5, a new version that significantly enhances the quality of video generation while also improving processing speed. This update demonstrates the continuous advancement of AI in the video generation domain, providing users with a more efficient and higher-quality AI video generation tool.

News · 2026-06-03

xAI Releases Grok Imagine 1.5: Preview of Image-to-Video Model Now Available

xAI has launched Grok Imagine 1.5, its latest image-to-video model, now available in preview via the xAI API. This model is designed to transform static images into dynamic videos, showcasing AI's continuous advancements in video generation. Grok Imagine 1.5 features significant improvements in both video quality and processing speed, providing users with a more efficient and higher-quality AI video generation tool.

News · 2026-05-18

xAI Releases Grok Skills: Empowering Document Generation and Workflow Automation

xAI has announced new skills for Grok, including document, presentation, and spreadsheet generation capabilities, as well as workflow automation features. These skills aim to enhance user productivity and allow developers to build and share custom skills, further expanding Grok's application scenarios.

News · 2026-05-15

Nous Research Open-Sources Hermes Agent with Grok Integration for Enhanced AI Interaction

Nous Research has open-sourced its self-improving Hermes agent, integrating xAI's Grok subscription service. This integration aims to leverage Grok's advanced AI capabilities within Hermes' adaptive learning framework, providing developers with a more powerful AI interaction solution. Hermes supports multimodal task processing and continuous learning, offering a flexible and efficient agent platform for complex AI applications.

News · 2026-04-30

xAI Releases Grok Custom Voices: Personalized Voice Cloning API with Short Recording Support

xAI has launched Grok Custom Voices, a feature that enables users to clone personalized voices from short recordings and apply them across Grok's Text-to-Speech (TTS) and Voice Agent APIs. This tool empowers developers with enhanced capabilities for creating customized voice interactions, improving the user experience and brand consistency in AI-driven applications.

News · 2026-04-08

Stability AI Launches Brand Studio: A Creative Production Platform for Brand-Centric Workflows

Stability AI has launched Brand Studio, a creative production platform designed to offer end-to-end, brand-specific solutions. The platform features Brand Central for training custom brand models and deep customization, Producer Mode for automated production planning, and Curated Model Routing to select the best models for optimal outputs. Additionally, it includes precision editing tools like Precision Inpainting and Product Insertion, catering to the needs of professional creative teams.

News · 2026-03-18

MiniMax Releases M2.7: The First AI Model to Participate in Its Own Evolution

MiniMax has unveiled the M2.7 model, the first AI system deeply engaged in its own evolution. Equipped with native Agent Teams and self-evolving capabilities, M2.7 excels in software engineering, professional tasks, and entertainment. This release highlights the potential of AI in handling complex tasks and continuous optimization, paving the way for new advancements in AI agent development.

News · 2026-01-28

xAI Releases Grok Imagine API: Next-Gen Video Generation Optimizing Quality, Cost, and Latency

xAI has launched the Grok Imagine API, a state-of-the-art video generation service that optimizes quality, cost, and latency. This API leverages advanced AI technologies to support video content creation across various applications, such as entertainment, advertising, and automated task processing. The release of Grok Imagine API signifies a significant advancement in AI video generation by xAI, providing developers with a more efficient and flexible solution for video content creation.

News · 2025-10-25

Hugging Face Releases LongCat-Video: A High-Efficiency Long Video Generation Model

Hugging Face has introduced LongCat-Video, a 13.6B parameter video generation model based on the Diffusion Transformer (DiT) framework, which excels in efficient and high-quality long video generation. It supports text-to-video, image-to-video, and video-continuation tasks with a unified architecture. Key features include a coarse-to-fine generation strategy and Block Sparse Attention for enhanced efficiency, as well as multi-reward RLHF training to achieve performance on par with leading models

News · 2025-09-18

Stability AI Launches Image Services on Amazon Bedrock, Empowering Enterprise-Grade Creative Workflows

Stability AI has announced the launch of its Image Services on Amazon Bedrock, providing professional-grade image editing tools such as inpainting, recoloring, and background removal. These tools are designed to support granular control over multi-step creative workflows and integrate seamlessly with existing image generation capabilities on Bedrock, offering enterprises an end-to-end AI solution for image generation and editing.

News · 2025-08-18

Qwen Launches Qwen-Image-Edit: A New Tool for Efficient and Precise Image Editing

Qwen has introduced Qwen-Image-Edit, an image editing tool built upon the 20B parameter Qwen-Image model. This tool extends Qwen-Image's unique text rendering capabilities to enable precise text editing. Additionally, by leveraging Qwen2.5-VL for visual semantic control and a VAE Encoder for visual appearance control, Qwen-Image-Edit achieves dual capabilities in semantic and appearance editing, offering a more efficient and intelligent solution for image editing tasks.

News · 2025-08-04

Qwen Releases Qwen-Image: A Breakthrough in Multimodal Image Generation and Text Rendering

Qwen has officially released Qwen-Image, a 20B parameter multimodal image generation model (MMDiT) that achieves significant advancements in complex text rendering and precise image editing. The model excels in handling multi-line text layouts, paragraph-level semantics, and fine-grained details, supporting alphabetic languages and intricate text formatting. Its key features include superior text rendering, image editing, and semantic understanding, providing a more efficient and intelligent sol

News · 2025-06-27

Qwen-TTS Update: Enhanced Multilingual Support and AI Voice Generation Breakthrough

Qwen has released an updated version of Qwen-TTS, featuring human-level naturalness and expressiveness in speech generation. The update introduces support for three Chinese dialects—Pekingese, Shanghainese, and Sichuanese—along with seven bilingual Chinese-English voices. Trained on a massive dataset, Qwen-TTS can automatically adjust prosody, pacing, and emotional inflections, providing users with more natural and intelligent voice generation capabilities. This advancement strengthens Qwen's po