Former Google Team Launches AI Photo Party App "Mixup": "Nano Banana+" Fill-in-the-Blank Recipe iOS Limited Invite Codes Now Available

Things, Inc., the company that created the 3D design tool Rooms, has launched an AI photo editor called Mixup today. It is available first on iOS and uses an invitation system. TechCrunch readers can experience it early with the code TCHCRH. The app is based on Google's Nano Banana model and features an innovative "Recipe" fill-in-the-blank prompt system. Users simply upload a photo or doodle, and they can instantly generate fun remixes like "Renaissance Self-Portrait" or "Halloween Pet Costume." They can also share the recipe and the final result on a public Feed, allowing friends to use the materials for their own secondary creations.

The Recipe mechanism solves the problem of "not knowing where to start when facing a blank text box": each recipe leaves blanks for "character/style/scenario" in a Mad Libs format, and users just need to fill them in to generate high-quality images. It also provides a Before/After switch to reduce the "slot machine" anxiety caused by random generation. Mixup also introduces a social feature called "Mixables": uploaded selfies are defaultly available for followers to use, and friends can creatively imagine each other being transported to funny scenarios like "Italy Brain Fog." If you don't want to be remixed, you can choose not to follow or not to upload.

The economic model uses a "Credit Point" billing system, offering 100 free credits (≈ $4) and charging about $0.04 per image; subscription tiers provide monthly packages of 100/250/500 credits. In terms of content review, besides using the OpenAI classifier, it mainly relies on the built-in restriction strategies of Nano Banana to block violent and sexual content. Jason Toff, founder of Things, Inc., said that if the iOS popularity meets the target, the team will soon launch Web and Android versions and explore video recipe features.

Mixup has started global pre-orders on the App Store. The official version will be released at midnight on November 21st, with limited initial invitation codes available.

MOSS-Speech Open Source: China's First Speech-to-Speech Large Model, Bypassing Text Intermediate

The MOSS team from Fudan University released MOSS-Speech, which realizes end-to-end speech dialogue for the first time. The model is now available and open-sourced on Hugging Face. It adopts a 'layer splitting' architecture, freezing the original text model and adding new layers for speech understanding, semantic alignment, and vocoder. It can complete speech Q&A, emotional imitation, and laughter generation in one step, without the traditional three-step process. Evaluation results show that the word error rate has been reduced to 4.1% in the ZeroSpeech2025 task, and the emotion recognition accuracy reached 91.2%.

Poe App from Quora Launches AI Group Chat Feature Supporting Up to 200 People

Quora's AI platform Poe has launched a group chat feature that allows up to 200 people to interact with various AI models simultaneously, covering text, image, video, and audio generation. This move coincides with OpenAI's ChatGPT group chat pilot, driving a transformation in AI interaction and enhancing collaboration and communication between users and their friends, family, or colleagues.

The AI Entrepreneurship Trend in County Towns: Training Classes for Workers and Photo Studios on the Street Are Popular

AI applications are spreading from big cities to county towns: free AI portrait stalls, low-cost ChatGPT account resales, and hundred-yuan AI hardware are selling well. County-level AI training has grown rapidly, with two-week intensive courses costing a thousand yuan, becoming a new choice for part-time jobs for workers. In half a year, more than 300 institutions have been added in Jiangsu, Shandong, and Hunan provinces, focusing on basic skills training.

Kuaiz AI Glasses S1 Launch Event Scheduled for November 27

Alibaba's Kuaiz will launch its first self-developed AI glasses, S1, on November 27. Pre-sales have already started, and the product won the top three sales positions in the smart glasses category on Tmall during the Double 11 shopping festival. The product is equipped with dual chips and a dual-eye display system, and it features proprietary patent technology, marking Alibaba's official entry into the smart wearable market.

New Open-Source Speech Model Maya1: Achieving Real-Time, Expressive Text-to-Speech

Maya Research launches the Maya1 text-to-speech model, with 3 billion parameters, which can run in real-time on a single GPU. The model generates controllable and expressive speech through natural language descriptions and text input, accurately simulating human emotions and voice details, such as specifying age, accent, or character traits.