AI Daily: ByteDance Launches StoryMem System; Moonshot AI Unveils New Multimodal Model; AI Glasses Pickle 1 Released

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 innovative AI product applications.

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

1. ByteDance Launches StoryMem System to Solve Character Consistency Issues in AI Video Generation

ByteDance, in collaboration with a research team from Nanyang Technological University, developed the StoryMem system to address the issue of character inconsistency in AI-generated videos across different scenes. The system stores key frames and references them when generating subsequent scenes, thereby maintaining consistency in characters and environments. Research results show that StoryMem improves cross-scene consistency by 28.7%, but it still faces challenges in complex scenarios and requires clearer character descriptions to enhance generation quality.

[AiBase Summary:]

🌟 The StoryMem system effectively solves the problem of inconsistent characters and environments in AI video generation.

📊 By storing key frames, StoryMem improves cross-scene consistency by 28.7% compared to existing models.

🛠️ The system still faces challenges in handling complex scenes and requires clearer character descriptions to improve generation quality.

More details: https://kevin-thu.github.io/StoryMem/

2. Moonshot AI Unveils New Multimodal Model, Kimi K2 Upgrade Set to Launch in First Quarter

Moonshot AI plans to launch its multimodal model K2.1/K2.5 in the first quarter of 2026. This model will be an upgrade based on its trillion-parameter open-source model Kimi K2, further enhancing multimodal processing and agent capabilities. The company currently has over 1 billion RMB in cash reserves, providing strong support for continued R&D.

[AiBase Summary:]

🚀 Moonshot AI plans to launch the multimodal model K2.1/K2.5 in Q1 2026, enhancing multimodal processing and agent capabilities.

🧠 The new model is upgraded from the trillion-parameter open-source model Kimi K2, supporting the "Thinking" model that allows "thinking and tool collaboration."

💰 Moonshot AI has over 1 billion RMB in cash reserves, providing sufficient support for continuous R&D.

3. A New King in AI Glasses! “Soul Computer” Pickle 1, Capable of Remembering Everything About You

Pickle 1 is a smart glasses that integrates AR display with advanced AI, positioned as a "soul computer." It continuously captures users' visual and audio context to achieve unlimited memory, emotional understanding, and proactive interaction.

[AiBase Summary:]

🧠 Pickle 1 actively learns user habits, converting daily experiences into searchable "memory bubbles."

👓 Pickle 1 features a lightweight design, supports all-day wear, and offers dual-eye full-color AR display and Qualcomm Snapdragon AI engine.

🔒 Pickle 1 emphasizes local data processing, using hardware isolation encryption to ensure user privacy and security.

More details: https://www.pickle.com/

4. Tsinghua University and OpenBMB Jointly Launch UltraEval-Audio: Open-Source Audio Model Evaluation Framework Released

UltraEval-Audio is an audio model evaluation framework jointly developed by Tsinghua University's NLP Lab, OpenBMB, and Miga Intelligence. The latest version v1.1.0 adds one-click reproduction functionality for popular audio models and expands support for TTS, ASR, and Codec models. The open-source release will significantly improve researchers' efficiency in audio model development and promote research progress in the field.

[AiBase Summary:]

🌟 UltraEval-Audio is a specialized evaluation framework for audio models, developed by multiple institutions.

🚀 The latest version v1.1.0 adds one-click reproduction function and supports more professional models for evaluation.

📈 The open-source release will improve the efficiency of researchers in developing audio models and promote the development of the audio model field.

More details: https://github.com/OpenBMB/UltraEval-Audio

5. OpenAI Bets on a "Voice-First" Future! Integrating Multiple Teams to Redesign Audio Models, First AI Audio Hardware May Launch Next Year

The article analyzes OpenAI's strategic layout in the voice interaction field, emphasizing its restructuring of the audio system to drive human-computer interaction into the post-screen era and explore voice-first hardware products to compete for user attention.

[AiBase Summary:]

🎙️ OpenAI restructures its audio strategy, expecting to launch voice-first personal devices in 2026.

🔊 The new audio model will enable more natural speech synthesis and real conversational interruptions.

📱 OpenAI plans to launch screenless smart speakers, AI glasses, or wearable devices, aiming to become the user's "smart companion."

6. Antigravity: The Ultimate Tool to Get Unlimited Gemini Quotas! Switch Accounts Instantly, Say Goodbye to AI Limitations

Antigravity Tools is an open-source desktop application that helps users expand the usage time of top models like Gemini and Claude through intelligent account management and seamless switching. It has become a hot topic in the AI community.

[AiBase Summary:]

🧠 Real-time quota monitoring: The app can monitor the remaining quotas and health status of multiple AI accounts globally.

🔄 Automatically recommends the best accounts: The system intelligently selects accounts with sufficient quotas based on real-time algorithms and supports seamless switching with one click.

🌐 Multi-protocol compatibility: Supports converting Web session into standardized API interfaces, solving protocol differences among different manufacturers.

More details: https://github.com/lbjlaq/Antigravity-Manager

7. Yuanxiang Opensources XVERSE-Ent Large Model! Focused on General Entertainment Scenarios, Bilingual Support, Filling Industry-Specific Model Gaps

Yuanxiang Technology has open-sourced the XVERSE-Ent large model tailored for the general entertainment industry. The model performs well in social interaction, game storytelling, and cultural creation, and provides multiple parameter versions to suit different needs.

[AiBase Summary:]

🎮 Optimized for general entertainment scenarios, supports social interaction, game storytelling, and cultural creation.

🧩 Provides multiple parameter versions, easy to deploy and compatible with open-source commercial use.

🌐 Bilingual support, integrating a lot of Chinese online literature and multilingual film and television texts.

8. Apple Responds to "AI Function Suspected of Being Limited": Do Not Use Third-Party Software to Bypass Restrictions, Be Cautious About Account Risks

Apple officially responded to rumors about the gray-scale testing of Apple Intelligence in the China version, clearly stating that it is not yet available and reminding users not to activate AI functions through third-party software, which may cause security risks.

[AiBase Summary:]

Apple confirmed that Apple Intelligence is not yet available in the China version, and all information should be based on the official website announcement.

Apple Intelligence requires high hardware performance and is only compatible with iPhone 15 Pro and newer models.

Apple warns users to avoid activating AI functions through third-party software to prevent account and financial security risks.