AI Daily: JD.com Opens Source Real-Time Video Editing Model; Qwen-Image-3.0 Launched; Tencent Hunyuan Launches Hy ASR 3.0 Preview

Welcome to the "AI Daily" section! 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 technology trends and learn about innovative AI product applications.

Fresh AI products Click to learn more:https://app.aibase.com/zh

1. JD Open-Sources Real-Time Video Editing Model: 30 Frames per Second Inference, Edit as You Watch Redefining Video Creation

JD has open-sourced its self-developed real-time streaming video editing model JoyAI-Video-Edit. This model has made breakthroughs in processing speed and video length, enabling the effect of editing videos while watching. At the same time, it also provides a new technical path for the synthesis of robot training data.

AiBase Summary:

🎥 Achieve real-time video editing, support editing as you watch

🚀 30 frames per second inference at 720P resolution

🤖 Support for robot training data synthesis

2. Alibaba Qwen Image Generation Model Qwen-Image-3.0 Launches, Open API and 0.18 Yuan per Image

The Alibaba Qwen image generation model Qwen-Image-3.0 has officially launched and is fully open to users. The model ranks first in the domestic text-to-image list, supports ultra-long text understanding and precise rendering, covers various art styles and languages, and has reasonable pricing, suitable for commercial workflows.

AiBase Summary:

🧠 Supports ultra-long text understanding, up to 4.5k token instructions input.

🎨 Covers 100+ art styles and 12 languages for native rendering.

💰 Reasonable pricing, starting from 0.18 yuan per image for the Standard version.

Details link: https://platform.qianwenai.com/try-ai?scene=image&models=qwen-image-3.0-pro

3. Tencent Hunyuan Releases Hy ASR 3.0 Preview: Speech Recognition No Longer Hard-Translates Word by Word, but Truly Understands Context

Tencent Hunyuan's Hy ASR 3.0 preview has achieved significant breakthroughs in speech recognition and semantic understanding, improving multi-language recognition accuracy and optimizing adaptability for professional scenarios.

AiBase Summary:

🧠 Deep integration of speech recognition and semantic understanding improves recognition accuracy.

📊 Significantly reduces word error rate (WER) across multiple languages.

💼 Optimizes professional scenario adaptation, supports hotword injection enhancement.

4. ByteDance Seed Unveils SeedRealtime: Full-Duplex Large Model for Audio and Video Goes into Doubao, Real-Time Interaction Becomes Smooth Without Lag

ByteDance Seed team's SeedRealtime is a full-duplex large model that naturally integrates audio, video, and text, and has been launched in the Doubao App, significantly improving the naturalness and smoothness of real-time interaction.

AiBase Summary:

🧠 SeedRealtime is a unified architecture full-duplex large model that can process audio, video, and text simultaneously.

🔄 Compared to traditional cascading solutions, SeedRealtime's end-to-end model reduces speaking rhythm issues and improves interaction naturalness.

💡 SeedRealtime shows good etiquette in real-time scenarios, can proactively remind and naturally pause, enhancing user experience.

5. Musk Announces Grok 4.6 Next Week, SpaceX Historical Data Will Be Injected into Model Training

Musk revealed the latest progress of Grok during the first earnings call of SpaceX, including the release dates of Grok 4.6 and Grok 4.7, and emphasized that SpaceX's historical data will be integrated into Grok's training to enhance the model's capabilities.

AiBase Summary:

🚀 Musk announced that Grok 4.6 will be released next week, followed by Grok 4.7 in the coming weeks.

📊 SpaceX will provide all historical data for Grok's training, enhancing the model's capabilities.

🧠 The next step of Grok will rely on SpaceX's private data, not just public corpora.

6. Insta360 AI Hardware Partners with Alibaba Qwen: Go Ultra Pocket Camera First Equipped with Voice Assistant Kira

Insta360 is integrating AI technology into hardware products, especially the Go Ultra series pocket camera. By combining self-developed technology with Alibaba Qwen's large model, the AI voice assistant Kira will provide users with a smarter interactive experience while considering regional differences.

AiBase Summary:

Insta360 introduces the AI voice assistant Kira in the Go Ultra series, enhancing user interaction experience.

Kira combines self-developed technology and Alibaba Qwen's multimodal model to achieve voiceprint recognition and cloud-based Q&A functions.

Different large models are used for different regions to ensure the accuracy of translation and broadcasting.

7. Mistral Launches Shieldstral: 3B Small Model Runs on a Single 16GB GPU for Multimodal Review, Claims to Win Open-Source SOTA

Mistral AI launched the Shieldstral content review model, which has 3 billion parameters and is released under the Apache 2.0 license with open weights. It supports 12 languages and can run on a single 16GB GPU. The unique feature of Shieldstral is that it writes review policies into the input, achieving a flexible review mechanism. Additionally, it can cover various review tasks, including prompt classification, answer review, and toxicity detection, and has reached the current state-of-the-art level in multimodal content review.

AiBase Summary:

🧠 Shieldstral is a content review model that supports 12 languages and can run on a single 16GB GPU.

🔒 It realizes a flexible review mechanism by writing review policies into the input.

📊 The model can cover various review tasks, including prompt classification, answer review, and toxicity detection.

Details link: https://huggingface.co/mistralai/Shieldstral-1.0-3B

8. Signed a Contract Worth 52,000 Yuan! Robot Unicorn Yushu Technology Strives for Sci-Tech Innovation Board, Estimated Issue Price Exceeds 100 Yuan

Yushu Technology, a leading company in human-like robots in China, has officially started its IPO on the Sci-Tech Innovation Board, expecting to raise over 4 billion yuan, with an estimated issue price of 104 yuan per share, marking the comprehensive acceleration of the capitalization process of embodied intelligence and the robot industry chain.

AiBase Summary:

🤖 Yushu Technology starts the preliminary inquiry for the Sci-Tech Innovation Board IPO, time from 9:30 to 15:00 on August 5.

💰 This time, it plans to issue 40.4464 million new shares, aiming to raise 4.202 billion yuan, with an estimated issue price of about 104 yuan per share.

🚀 As the "first stock of humanoid robots in A-shares," Yushu Technology marks the comprehensive acceleration of the capitalization process of embodied intelligence and the robot industry chain.