AI Daily: Qwen-Image-i2L Enables Zero-Threshold Training of LoRA; Zhipu AI Keyboard Launches New Version; Lingguang Introduces a Feature for Generating Animated Explainers

Welcome to the "AI Daily" column! This is your guide to exploring the world of artificial intelligence every day. Each day, we present you with the latest content in the AI field, focusing on developers to help you understand technology trends and innovative AI product applications.

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

1. OpenAI's next-generation image generation model appears to have surfaced, and blind testing has already begun on Design Arena and LM Arena

OpenAI is conducting small-scale blind tests on two new image generation models codenamed "Chestnut" and "Hazelnut," marking a significant advancement in the text-to-image domain. The new model shows significant improvements in several key dimensions, including world knowledge understanding, image realism, and code embedding capabilities.

AiBase Summary:

🧠 The new model has nearly reached the level of Google's latest Nano Banana Pro in terms of world knowledge understanding.

📷 The new model can generate celebrity-style selfies with near-photographic realism, with significantly improved facial details, facial proportions, and lighting treatment compared to gpt-image-1.

💻 The new model excels in embedding readable code within images, accurately rendering complex code snippets, flowchart labels, and mathematical formulas.

2. One image becomes an AI style tool! Qwen-Image-i2L revolutionizes image generation, and zero-barrier LoRA training is coming

Qwen-Image-i2L is a revolutionary technology that can transform any single image into a trainable LoRA model, greatly lowering the barrier for personalized style transfer. It uses a multimodal feature extraction system to intelligently decompose the input image into core visual elements such as "style, content, composition, and tone," forming a compact LoRA module. The article also introduces four exclusive model styles, along with its technical foundation and potential challenges.

AiBase Summary:

🖼️ Core Innovation: From one image to LoRA, unlock personalized generation with one click.

🎨 Four model variants, precisely adapted to diverse scenarios.

🚀 Technical Foundation and Potential Challenges: Efficient but beware of overfitting.

More details: https://modelscope.cn/models/DiffSynth-Studio/Qwen-Image-i2L/summary

3. Zhipu AI Input Method Released New Version, Voice Recognition Technology Also Open Sourced!

Zhipu AI has launched the GLM-ASR series of voice recognition models and open-sourced the related technologies, aiming to enhance users' voice interaction experience. At the same time, they released the desktop version of the Zhipu AI Input Method, supporting functions such as voice-to-text conversion, translation, and text rewriting, providing users with a convenient PC voice interaction experience. New users can also get 2000 points and enjoy 28 days of free usage time.

AiBase Summary:

🌟 Launch of the GLM-ASR series model, including globally leading cloud-based and edge voice recognition models with excellent recognition accuracy.

🛠️ The new Zhipu AI Input Method is launched, supporting voice-to-text, translation, and rewriting, offering a convenient PC voice interaction experience.

🎁 New users can get 2000 points for free and enjoy up to 28 days of usage, encouraging more users to experience the intelligent input method.

More details: https://huggingface.co/zai-org/GLM-ASR-Nano-2512

4. Lingguang Launches New Features: Generate Educational Animations with One Click, Visualize Complex Principles

The general AI assistant "Lingguang" has launched a new feature for generating educational animations, aiming to explain complex scientific principles through intuitive and vivid animation forms to improve learning outcomes. This feature covers multiple core science and engineering fields and has been integrated into the conversation function, allowing users to trigger dynamic explanations instantly.

AiBase Summary:

🧪 The general AI assistant "Lingguang" has launched the educational animation generation feature, visually presenting abstract scientific principles.

🌍 Lingguang can automatically generate simulation animations to help users understand complex concepts in astronomy, physics, and other fields.

🔄 This feature has been integrated into the conversation function, allowing users to trigger and receive dynamic explanations immediately.

5. McDonald's AI-generated Christmas Ad Removed Due to Mockery

McDonald's Netherlands branch launched an AI-generated Christmas ad that was strongly criticized for its unconventional viewpoint and poor visual effects, ultimately being withdrawn. Although the production team claimed to have invested a lot of time and effort, public disapproval remained evident.

AiBase Summary:

🌟 McDonald's Netherlands branch launched an AI-generated Christmas ad, which was criticized by netizens and withdrawn due to unconventional viewpoints and poor visual effects.

🎬 The ad's rapid scene transitions and strange characters caused audience dissatisfaction, with poor video quality.

💼 The production company defended the ad, claiming it had invested a lot of manpower, but the public reaction remained negative.

6. Musk's xAI Launches New Tool, Real-Time AI Ads Can Be Inserted Into Movies

Musk's xAI launched a new tool called Halftime, which can insert AI-generated advertisements into movies and TV shows in real time, making ads blend with the plot. However, this technology has sparked widespread discussions about the integrity of film art and copyright issues.

AiBase Summary:

🌟 Halftime is a tool that can insert AI-generated ads into movies and TV shows in real time.

🎥 The tool can interrupt a character's dialogue and display the product, aiming to make the advertisement look like part of the storyline.

📜 The controversy raised concerns about the potential impact of this technology on the integrity of film art.

7. Alibaba Establishes Qwen C-end Business Group: Integrates Quark, UC, Challenging the Super App Status

Alibaba Group announced the establishment of the "Qwen C-end Business Group," integrating the core resources of the original Smart Information and Smart Interconnection business groups, with the goal of making "Qwen" a super app in the AI era, becoming the first entry point for users to access digital life. At the same time, it plans to expand to multi-terminal scenarios such as glasses, PCs, and cars, building an omnipresent AI network to enhance user experience and strengthen its core competitiveness in the AI era.

AiBase Summary:

🚀 Alibaba established the Qwen C-end Business Group, integrating Quark, UC, and other core apps.

📱 Aimed at building a super app in the AI era, becoming the first entry point for users' digital life.

🌐 Expanding to multi-terminal scenarios such as glasses, PCs, and cars, building an omnipresent AI network.

8. Microsoft Excel Web Version Launches New "Agent Mode," AI Helps Efficient Data Processing

Microsoft launched the "Agent Mode" for the web version of Excel, enabling AI to participate deeply in complex data processing tasks, improving efficiency and enhancing transparency.

AiBase Summary:

🧠 Agent mode makes Excel a smarter digital assistant, improving data processing efficiency.

📊 AI can handle complex workflows independently and update spreadsheet data in real time.

🔍 Provides AI transparency, allowing users to review the AI's decision-making process and reasoning logic.