Morning Brief

1. Google releases Gemini Embedding 2 model
2. MediaTek unveils multiple IoT SoC platforms
3. ComfyUI adds support for app-style workflows
4. RØDE launches Video Core
5. iReader releases iReader Neo 3 Ultra

---

Google releases Gemini Embedding 2 model

On March 11, Google announced the new Gemini Embedding 2 model. It is Google’s first native multimodal embedding model, capable of mapping text, images, videos, and documents into the same embedding space.

Gemini Embedding 2 supports text, images, video, audio, and documents, and can recognize semantic intent across 100 languages. The new model simplifies complex data processing pipelines while enhancing multimodal application capabilities. Use cases include retrieval-augmented generation (RAG), semantic search, sentiment analysis, and data clustering. Gemini Embedding 2 can also accept multiple input types—such as “image plus text”—within a single request, enabling analysis of relationships across different media formats.

The model is currently available in public preview through the Gemini API and Vertex AI. Source

---

MediaTek unveils multiple IoT SoC platforms

On March 9, MediaTek unveiled a series of IoT SoC platforms at Embedded World 2026 in Nuremberg, Germany, including the high-end Genio Pro solution as well as the lower-positioned Genio 420 and Genio 360.

Genio Pro is built on TSMC’s advanced 3nm process and features CPU and GPU specifications similar to the mobile Dimensity 9400. It is designed to operate under industrial wide-temperature environments. The platform supports up to 16 cameras and as many as three 4K displays, and is compatible with a range of Linux distributions as well as the open-source ROS robotics operating system.

Genio 420 uses a 6nm process and provides 7.2 TOPS of system AI computing power while integrating 16GB LPDDR5X memory. It remains pin-compatible with the existing Genio 720 and Genio 520, allowing users to migrate between chip models more easily.

The lower-tier Genio 360 and Genio 360P deliver system AI performance of 6 TOPS and 8.5 TOPS respectively. They support up to 8GB LPDDR4X-3733 memory and can meet the requirements for running 2B parameter models on-device. Source

---

ComfyUI adds support for app-style workflows

On March 10, ComfyUI introduced three new features: App Mode, App Builder, and ComfyHub. The goal is to transform the platform’s node-based workflows—originally designed for advanced users—into shareable, runnable application interfaces, allowing people unfamiliar with node graphs to use ComfyUI workflows.

App Mode enables any ComfyUI workflow to switch into a simplified interface with one click. The underlying node graph remains hidden, leaving only the inputs and outputs that end users need to operate. The workflow itself does not change and continues running on the same ComfyUI instance, backend, and task queue. As a result, existing nodes, model updates, and custom extensions can be inherited directly in App Mode.

The newly introduced App Builder allows creators to configure which node parameters are exposed to users. Developers can decide which inputs and outputs appear in the interface, while more complex parameters remain inside the node graph. ComfyUI explains that a typical text-to-image workflow may include numerous parameters such as samplers, steps, CFG, schedulers, LoRA weights, seeds, and resolution, but only a small subset may need to be exposed in practical applications.

ComfyUI also launched a preview version of ComfyHub, a public distribution hub for workflows and applications where users can browse and run community creations directly. These features are currently available on Comfy Cloud and Comfy Local. Source

---

RØDE Launches Video Core

On March 10, RØDE introduced the Caster Video Core, an integrated audio and video production solution designed for livestreaming and video capture scenarios.

The device features an octa-core CPU and can switch between up to four video sources and five custom scenes. It includes three 1080p HDMI inputs, supports automatic frame rate switching, and provides two USB-C ports for connecting video or audio devices. It also comes with two Neutrik combo inputs, allowing users to connect XLR microphones, musical instruments, or other audio equipment.

The product supports audio adjustments such as EQ, compression, filtering, and de-essing. It can also connect to mixing consoles such as the Caster Pro II or Duo, enabling additional audio and video input and output options. The USB-C ports can connect two computers simultaneously, or a portable storage drive to save recorded footage directly, and support dual-system livestreaming. Source

---

iReader Launches iReader Neo 3 Ultra

On March 10, iReader introduced its new flagship e-reader, the iReader Neo 3 Ultra. The device is powered by an octa-core processor, delivering a 200% overall performance improvement compared with previous quad-core chips.

In terms of storage, the device comes with 4GB of RAM and 64GB of internal storage. The iReader Neo 3 Ultra features a new “Ultra Refresh” technology that increases frame rate while maintaining similar power consumption, enabling smoother page transitions and stable video playback. Combined with iClean 2.0 technology, the device uses AI-based image recognition algorithms to introduce an improved partial refresh mechanism, effectively reducing ghosting and flickering while enhancing reading comfort.

For the display, the Neo 3 Ultra is equipped with a 300 PPI recessed screen made with Carta 1300 material, offering sharper text edges and clearer image details. The device measures just 7.5mm thick and weighs about 150g. It also includes a front light with 30 levels of brightness and color temperature adjustment. With ComfortLight Pro+ technology, the lighting remains soft and evenly distributed, allowing fine tuning based on ambient light conditions and personal preference.

The device includes a 2000mAh battery, rated for up to 53 hours of reading or 18 days of standby. The product is now available on e-commerce platforms with a price of 1,499 yuan. Source