Welcome to the "AI Daily" column! This is your guide to exploring the world of artificial intelligence every day. Every day, we present you with the latest content in the AI field, focusing on developers to help you understand technical trends and innovative AI product applications.
New AI products Click to learn more: https://app.aibase.com/zh
1. Keling AI Launches O1 Video Large Model Today: Unified Multimodal Architecture Supports Generating Videos with a Single Sentence
Keling AI announced that its self-developed O1 video large model has been fully released. The model uses an MVL unified interaction architecture, supporting three types of input instructions: text, image, and video, and can complete tasks such as text-to-video, image-to-video, local editing, and lens extension in one go. In addition, the model solves the "feature drift" problem during lens switching through multi-viewpoint subject construction technology, ensuring smooth transitions. The O1 model is now available for experience on Keling App and the official website, and will later open API interfaces for third-party platforms to integrate.
【AiBase Highlights:】
🧠 The O1 video large model uses the MVL unified interaction architecture and supports three types of input instructions: text, image, and video.
🎬 Completes text-to-video, image-to-video, local editing, and lens extension in one go.
📊 Keling AI plans to open API interfaces for third-party platform integration.
2. Qwen App Integrates Wan2.5, Video Capabilities Upgraded
The Qwen App has integrated the Wan2.5 model, significantly enhancing video creation capabilities. It supports synchronized audio and video output and allows users to generate high-quality dynamic video content by customizing images and text, further lowering the creation barrier and stimulating user creativity.
【AiBase Highlights:】
🎥 Qwen App integrates the Wan2.5 model, upgrading video creation capabilities comprehensively.
🎙️ Supports synchronized audio and video output, generating high-quality videos with sound and visuals aligned.
🔄 Users can upload photos and text to create dynamic dance and sing videos, reducing the creation barrier.
3. PixVerse V5.5 Released: Supports "Director-Level" Audio-Visual Synchronization
The release of PixVerse V5.5 brings significant changes to video production. Users need only input a sentence to generate high-definition videos with synchronized sound and lip movements, and it supports automatic switching between multiple lenses, greatly improving the convenience of video creation.
【AiBase Highlights:】
🎥 Support generating high-definition videos with a single sentence, achieving audio-visual synchronization.
🔄 Automatic switching between multiple lenses enhances video narrative logic.
🚀 Self-developed architecture improves video generation speed and quality, offering a one-stop service.
Details: https://pai.video
4. DeepSeek-V3.2 Launched: Introduces Innovative Sparse Attention Architecture, API Cost Halved, Performance Comparable to Top Closed-Source Models
Chinese AI startup DeepSeek AI released the DeepSeek-V3.2 series model, including DeepSeek-V3.2 and its high-computation-enhanced version DeepSeek-V3.2-Special. The new model introduces an innovative sparse attention mechanism (DSA), improving efficiency in long-text tasks and reducing API costs. The DeepSeek-V3.2-Special version performs exceptionally well in high-difficulty reasoning tasks, even surpassing GPT-5. In addition, the model provides open-source kernels and demonstration code, supporting researchers and enterprises for commercial deployment.
【AiBase Highlights:】
🧠 Introduces an innovative sparse attention mechanism (DSA) to improve the efficiency of long-text tasks.
🚀 DeepSeek-V3.2-Special performs excellently in high-difficulty reasoning tasks, surpassing GPT-5.
📊 API cost reduced by 50%, supporting researchers and enterprises for commercial deployment.
Details: https://huggingface.co/deepseek-ai/DeepSeek-V3.2-Exp
5. Runway Unveils New Gen-4.5 Video Generation Model, Enhancing Creativity and Visual Quality
Runway launched its latest video generation model, Gen-4.5, significantly improving visual accuracy and creative control, suitable for social media short video creation. Despite competition, Gen-4.5 excels in object and character consistency, but still faces issues with causal reasoning and temporal coherence. At the same time, the authenticity of AI-generated content has sparked industry discussions, and it is recommended to add disclaimers to distinguish real from fake content.
【AiBase Highlights:】
🎥 Gen-4.5 model from Runway makes video generation more creative and visually consistent.
📱 The model is mainly targeted at social media short videos, differing from competitors' focus on long-form videos.
⚖️ Authenticity of AI-generated content has sparked internal discussions in the industry, and it is recommended to add disclaimers to differentiate between real and fake content.
6. Google AI Search Experience "Speeds Up": New Design Enables Seamless Conversations, Gemini3Pro Enters 120 Countries!
Google is taking a series of measures to make its AI models and features more accessible and user-friendly. On one hand, Google is testing a new design to optimize the transition experience from AI overview to AI mode; on the other hand, the Gemini3Pro model is undergoing large-scale international expansion.
【AiBase Highlights:】
💡 Optimizing AI experience for seamless conversations.
🌍 Gemini3Pro/Nano Banana Pro expands to 120 countries and regions.
🚀 Enhancing AI functions in Google search.
7. Lovart Touch Edit Launched: Tap and Edit, AI Image Editing Enters the "Zero Mask" Era
Lovart's Touch Edit feature enables efficient and convenient image editing through natural language instructions and intelligent recognition technology. Users no longer need manual operations to complete complex image modification tasks, significantly improving design efficiency.
【AiBase Highlights:】
✨ Core function of Touch Edit: Automatically identify and edit image elements through natural language instructions.
🧩 Select & Remix supports mixing multiple images, allowing users to drag and rearrange different image elements.
⚙️ Integrated technology: Combines GPT-4o, Flux Pro, and Sora models to provide an efficient AI workflow.
8. Ant Digital Agentar Selected as "Top Tier" in China's Intelligent Body Development Track
Ant Digital's Agentar platform successfully entered the "top tier" of China's intelligent body development track due to its comprehensive technical architecture, mature product iteration, and years of accumulation in the financial field, demonstrating its leading position in AI intelligent body development.
【AiBase Highlights:】
🧠 Ant Digital's Agentar platform demonstrates leading strength in intelligent body development.
💼 The platform has rich technical experience and large-scale implementation results in the financial field.
📈 Agentar-Fin-R1 inference large model ranks first in three financial benchmark tests.
