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 · 2026-09-09

AWS Announces Major August Updates: Expanding Amazon Bedrock Features and AI Agent Applications

In August 2026, AWS introduced significant updates to Amazon Bedrock, AgentCore, and Strands Robots, including support for million-token context windows for OpenAI models, cross-Region inference, dedicated compute support for AgentCore, and physical deployment capabilities for Strands Robots. These updates aim to enhance the reliability and efficiency of AI systems in enterprise applications while providing developers with more powerful tools to facilitate AI adoption in complex scenarios.

News · 2026-09-09

Google DeepMind and Filmmakers Collaborate: AI Recreates 70-Year Love Story

Google DeepMind has partnered with filmmakers to use AI technology to recreate the unrecorded past of a couple in the short film 'Love, Rendered.' This breakthrough demonstrates AI's potential in visual content generation and emotional storytelling, offering new tools and methods for filmmaking and historical recreation.

News · 2026-09-09

NVIDIA Launches AI for Media and Sports Intelligence Playbooks, Empowering Real-Time Content Creation and Multimodal AI

At IBC 2026 in Amsterdam, NVIDIA announced an expansion of its AI for Media solutions and introduced the NVIDIA Sports Intelligence Playbooks. These innovations leverage GPU-accelerated SDKs, microservices, and blueprints to empower the media and entertainment industry with real-time video analysis, content verification, personalization, and localization capabilities. Additionally, the NVIDIA Sports Intelligence Playbooks provide sports organizations with a framework to fine-tune AI models on pr

Info · 2026-09-09

Caption.me Achieves Autonomous Improvement: AI Image Captioning Enters a New Era

Caption.me has announced that its AI image captioning system has achieved autonomous improvement capabilities, allowing it to continuously optimize its performance without human intervention. This breakthrough marks a new era for AI image captioning technology, as the system can now adapt more efficiently to diverse scenarios and enhance the quality of generated captions through automated learning and iterative mechanisms. This innovation not only improves AI performance in image understanding a

News · 2026-09-09

Suno Releases Suno v6: New AI Model Trained on Licensed Music Amid Rising Copyright Lawsuits

Suno has announced the release of its latest AI music generation model, Suno v6, in response to a growing number of copyright lawsuits. Unlike its predecessors, Suno v6 is trained exclusively on licensed music, avoiding the use of potentially infringing content. This strategic shift underscores Suno's commitment to legal compliance and innovation in the AI music generation space while mitigating legal risks.

News · 2026-09-08

AuK Open-Sourced: A Foundational AI Model for Unified Speech Generation and Editing

AuK is an open-source foundational model designed to unify speech generation and editing through a common interface of natural language instructions and audio context. Built on approximately 3.03 billion instruction-audio instances and 1.95 million hours of effective supervision, AuK integrates a multimodal large language model, a jointly trained VAE, and a hybrid rectified-flow Transformer. It excels in zero-shot and instruction-controlled speech generation and general instruction-guided editin

News · 2026-09-08

Hugging Face Proposes Mask Forcing: Enhancing Autoregressive Video Diffusion Distillation

Hugging Face's research team introduces Mask Forcing, a novel Dual-Noise Masking Rollout strategy for enhancing autoregressive video diffusion distillation. By injecting cleaner signals through random masks during the self-rollout process, this method mitigates mode collapse induced by reverse-KL mode seeking, leading to improved visual quality and realism in generated videos. Extensive experiments demonstrate that Mask Forcing effectively enhances multiple autoregressive video diffusion distill

News · 2026-09-08

NeoHorse-1 Released: Exploring Recursive Self-Improvement via Agentic Post-Training

NeoHorse-1 is a novel family of agent-native models designed to explore recursive self-improvement (RSI) through agentic post-training. The system combines a heterogeneous model pool with intelligent routing, converting user interaction data into training examples optimized through structural validation and multi-dimensional semantic evaluation. NeoHorse-1 demonstrates significant performance improvements across multiple benchmarks, showcasing a closed-loop learning path from evaluation to updat

News · 2026-09-08

LlamaIndex Releases LlamaParse v3: Enhanced AI-Driven Document Parsing and Intelligent Processing

LlamaIndex has released LlamaParse v3, an innovative tool for document parsing and intelligent processing. The new version introduces AI-driven OCR technology based on plan-act-verify loops, supporting visual grounding and dynamic table processing to handle various complex document types without the need for template maintenance or training phases. Additionally, v3 enhances capabilities for creating audio knowledge bases, semantic search, and cross-platform integration, providing more efficient

Info · 2026-09-08

Inception Labs Releases Mercury 2.5: Advancing the Intelligence of Diffusion Models

Inception Labs has released Mercury 2.5, an intelligent system designed for diffusion models (Diffusion LLMs), aimed at enhancing the intelligence of generative AI. Mercury 2.5 achieves significant breakthroughs in generation quality, efficiency, and controllability by optimizing the model architecture and inference process. This version supports more complex generation tasks and introduces advanced context understanding and multimodal processing capabilities, providing a more powerful tool for

News · 2026-09-08

OpenAI Releases ChatGPT Images 2.5 with Sketch Feature for Enhanced Image Generation

OpenAI has announced the release of ChatGPT Images 2.5, introducing a new feature called Sketch that allows users to create AI-generated images by drawing sketches directly within the chat interface. This feature enhances the personalization and precision of image generation, offering a more intuitive way for users to interact with AI for creative design and content creation purposes.

Info · 2026-09-08

inclusionAI Releases Ling-3.0-flash-VL: Multimodal AI Model for Efficient Fusion of Vision and Text

inclusionAI has released Ling-3.0-flash-VL, a multimodal AI model that integrates vision and text understanding capabilities. Built upon the Ling-3.0-flash architecture, it extends native image and video understanding with a total of 1240 billion parameters, activating only 55 billion parameters per token, and supports a context window of up to 1 million tokens. Key features include a ViT visual encoder, VideoRoPE for spatial-temporal encoding, and a sparse MoE mechanism, balancing strong multim

News · 2026-09-08

OpenAI Releases ChatGPT Images 2.5: Enhanced Personalization for Image Generation

OpenAI has launched ChatGPT Images 2.5, an upgraded version of its image generation tool designed to transform users' ideas, sketches, and reference photos into more personalized and polished images. This release focuses on enhancing the quality of image generation, personalization, and alignment with user intent, offering a more tailored and refined image creation experience. The update broadens ChatGPT's applicability in creative design, content creation, and other related fields.

Info · 2026-09-08

Qwen Releases Qwen-Drive-1.0-4B: A Vision-Language Foundation Model for Autonomous Driving

Qwen has released Qwen-Drive-1.0-4B, a vision-language foundation model tailored for autonomous driving. Built upon a pretrained VLM, it integrates 3D perception, visual question answering, and motion planning within a unified framework. Its bird’s-eye-view (BEV) perception head performs 3D object detection, semantic occupancy prediction, and BEV map segmentation, while a Planning Expert generates future ego trajectories. Experiments demonstrate strong performance in 3D perception and driving sc

Info · 2026-09-08

Inception Launches Mercury 2.5: A New Tier of Intelligence for Diffusion LLMs

Inception has launched Mercury 2.5, an advanced system for diffusion-based large language models (LLMs) that aims to elevate the intelligence of generative AI. This release features significant improvements in model architecture and inference processes, enhancing generation quality, efficiency, and controllability. Mercury 2.5 supports more complex generative tasks and introduces advanced contextual understanding and multimodal processing capabilities, providing a powerful tool for AI applicatio

Info · 2026-09-08

AI Conversations Analysis: 71% of 22GB Data Revealed as Screenshots

Mint DZGAPP's research team conducted an in-depth analysis of 22GB of AI conversation data and found that 71% of the content consisted of screenshots. This discovery highlights the significance and prevalence of non-text data in AI interactions, providing new insights for optimizing AI systems. The study suggests that the prevalence of screenshots in AI conversations indicates a user preference for visual communication, which has significant implications for AI model training and interaction des

Info · 2026-09-08

DeepSeek Releases Flash 4.1: Enhanced Multimodal Support, Performance, and Cost Efficiency

DeepSeek has announced the internal beta release of its latest version, DeepSeek Flash 4.1. This version features a new architecture with native multimodal support, enhanced capabilities, faster speeds, and lower costs. Users can access the model via API by setting the model name to 'deepseek-v4.1-flash-expires-on-0910', with pricing identical to DeepSeek Flash 4 and a rate limit of 20 concurrent requests per account. This release signifies DeepSeek's continued advancement in the field of multim

News · 2026-09-08

OpenAI Explores AI Capabilities and Economics: Enabling More Efficient and Affordable Growth

OpenAI has released a research-oriented article exploring how more capable and affordable AI can expand the work capabilities of individuals and businesses, driving more efficient and economical growth. The article highlights that advancements in AI technology can enhance productivity while reducing barriers to access, allowing more businesses and individuals to benefit. This trend is expected to reshape multiple industries and have a profound impact on the global economy.

Info · 2026-09-08

DeepSeek Releases V4.1 Flash: Enhanced Multimodal Support, Performance, and Cost Efficiency

DeepSeek has announced the internal beta release of its latest version, DeepSeek V4.1 Flash. This new iteration features a redesigned architecture with native multimodal support, delivering enhanced capabilities, faster processing speeds, and improved cost efficiency. Users can access the beta by setting the model name to 'deepseek-v4.1-flash-expires-on-0910' while maintaining the same pricing as DeepSeek V4 Flash, with a limit of 20 concurrent requests per account. This release signifies DeepSe

News · 2026-09-08

LlamaIndex Releases LlamaParse and LiteParse: Revolutionizing Document Parsing and Agent Understanding

LlamaIndex has launched LlamaParse and LiteParse, two groundbreaking tools designed to enhance AI agents' document understanding capabilities. LlamaParse leverages a cloud-based API to deliver structured, multimodal parsing for complex documents, while LiteParse offers a local-first approach optimized for speed and privacy, preserving document layout and spatial relationships. These tools empower agents to process intricate documents in fields like finance, law, and healthcare with greater accur

Info · 2026-09-08

Reddit User Proposes Method to Reduce Image-Processing Token Usage by 95%

A Reddit user, angelinusbread, has proposed a novel approach for reducing the cost of image-based LLM inference. The method reportedly achieves approximately 95% lower token usage compared to GPT-4o's direct image processing while maintaining similar accuracy. The approach was evaluated on the MOMA Graph benchmark using 1,315 questions. While implementation details remain undisclosed, the result has sparked significant interest in the AI community regarding multimodal AI efficiency improvements.

Info · 2026-09-08

Generating Full 'Bad Apple' Video from Single Initial State: Open-Sourcing a Tiny Recurrent Dynamical System with 417k P

Reddit user SEBADA321 has open-sourced BadAppleRNN, a tiny recurrent dynamical system capable of autonomously generating the full 'Bad Apple' video (6,500 frames) from a single initial state (h_0, c_0). The system employs an LSTM-style recurrent neural network and a bilinear upsampling decoder, with a total of 417k parameters, achieving over 200 FPS on an RTX 4080. The project demonstrates techniques like noise injection and acceleration regularization to ensure dynamical stability over long seq

Info · 2026-09-07

DeepSeek Releases DeepSeek-V4-Flash-Vision-Exp: Revolutionizing Game Development and Visual AI Applications

DeepSeek has unveiled its latest multimodal AI model, DeepSeek-V4-Flash-Vision-Exp, which integrates advanced vision processing capabilities to efficiently generate and optimize game worlds. The model can automatically generate and correct game models, fix visual artifacts, write animation scripts, and perform game testing, enabling developers to create high-quality game content in just a couple of days. This release marks a significant breakthrough for DeepSeek in the field of multimodal AI, pr

News · 2026-09-07

LlamaIndex Launches Deep Extraction: Revolutionizing Document Processing with Agentic Verification

LlamaIndex has launched Deep Extraction, a novel approach to document processing that leverages agentic verification to achieve unprecedented accuracy in handling complex documents. Unlike traditional single-pass extraction methods that lack error detection, Deep Extraction employs an iterative process of extraction, verification, and correction. This results in field-level accuracy improvements from 10-20% to 99-100%, making it ideal for high-stakes documents. The technology utilizes Vision Lan

News · 2026-09-07

Meta Releases Muse Spark 1.1: Significant Advances in Multimodal Reasoning and Agentic Task Handling

Meta AI has launched Muse Spark 1.1, the latest version of its multimodal reasoning model designed for agentic tasks. This release brings significant performance improvements in tool use, coding, and multimodal understanding, excelling in handling complex projects, coordinating workflows across applications, and managing long contexts of up to one million tokens. Muse Spark 1.1 also demonstrates strong capabilities in coding tasks and agentic interactions, showcasing its potential for real-world

News · 2026-09-07

Meta Launches Muse Image and Muse Video: AI-Powered Multimodal Generation Tools Revolutionizing Creativity

Meta AI has launched two groundbreaking multimodal AI generation tools: Muse Image and Muse Video. Muse Image excels in high-quality image generation through precise instruction following, multi-reference composition, and intelligent tool invocation, and integrates a Content Seal watermark system for provenance tracking. Muse Video delivers exceptional visual fidelity with native audio support, focusing on enhancing video generation experiences. These tools mark a significant advancement in AI g

Info · 2026-09-07

Tencent Releases EVIE-8B and EVIE-4.5B: Breakthrough Models in Visual Document Retrieval

Tencent has released two high-performance visual document retrieval models: EVIE-8B and EVIE-4.5B. The EVIE-8B model sets a new industry benchmark with a 66.75 nDCG@10 score on the ViDoRe V3 benchmark, showcasing its leading retrieval performance. It employs high-capacity 4096D representations to capture fine-grained document structures, including layouts, typography, charts, and tables. The lightweight EVIE-4.5B variant leverages Prefix-MRL technology for efficient single-projection inference a

News · 2026-09-07

Kimi Launches Kimi K3: World's First Open 3T-Class Multimodal Model

Kimi has unveiled Kimi K3, the world's first open 3T-class model, featuring 2.8 trillion parameters, native multimodal capabilities, and a 1-million-token context window. Designed for frontier intelligence in coding, knowledge work, and reasoning, Kimi K3 demonstrates strong performance in long-horizon tasks and scientific computations. While it lags behind the most advanced proprietary models, its open architecture and efficiency improvements make it a significant advancement in the AI landscap

News · 2026-09-07

LlamaIndex Releases Agentic Document Processing: Revolutionizing Document Workflows

LlamaIndex has released Agentic Document Processing (ADP), an AI agent-based document processing technology that aims to achieve precise understanding and automation of complex documents. This technology significantly enhances the accuracy and efficiency of document processing by understanding document context, inferring intent, and self-validating results. It is particularly suitable for high-stakes scenarios such as healthcare and finance. ADP employs visual localization and bounding box techn

News · 2026-09-07

Kimi Releases PerceptionBench: A Novel Benchmark for Atomic Visual Perception in Multimodal Models

Kimi has released PerceptionBench, a novel benchmark designed to evaluate the atomic visual perception capabilities of multimodal large language models (MLLMs). By isolating 10 atomic perceptual capabilities derived from model failures and constructing 3,000 verified questions, PerceptionBench measures what models actually perceive rather than infer. The benchmark reveals that current models struggle with perception-related hallucinations, and its release aims to drive advancements in faithful a