Morning Brief

1. Canva AI 2.0 launched
2. Anthropic releases Claude Opus 4.7
3. OpenAI upgrades Codex with multiple practical features
4. DJI launches Osmo Pocket 4 gimbal camera
5. Amazon introduces the thinnest Fire TV Stick HD
6. Adobe releases Firefly AI assistant
7. Tencent unveils Hunyuan 3D World Model 2.0
8. Mastercard enables cross-border Apple Pay support for Chinese cardholders
9. Apple Wallet now supports NFC transit cards via Alipay
10. News Worth a Quick Look

---

Canva AI 2.0 launched

On April 16, Canva unveiled Canva AI 2.0, announcing its shift toward an “integrated productivity system.”

This update introduces a new underlying architecture, with core capabilities including: conversational design, allowing users to generate fully editable designs through natural language or voice; agent orchestration, automatically coordinating multiple tools to accomplish complex tasks (such as generating full multi-channel marketing campaigns); intelligent object editing, enabling precise adjustments to individual elements without affecting the overall design; and persistent memory, which learns user workflows and automatically applies brand styles.

In terms of intelligent workflows, Canva AI 2.0 expands its application scenarios into everyday office tasks. The new system integrates with commonly used tools such as Slack, Gmail, Zoom, and Google Drive, enabling direct extraction of key points from audio, video, or chat logs to generate documents. It also introduces automated planning with background offline operation, web-wide research capabilities, Canva Code 2.0 with support for importing and editing HTML files, and Sheets AI, which can generate structured tables with a single prompt.

Canva AI 2.0 will begin rolling out in limited regions worldwide for early access. Source

---

Anthropic releases Claude Opus 4.7

Anthropic has released the Claude Opus 4.7 model. Based on Opus 4.6, the new version focuses on improving performance in complex software engineering tasks, reducing reliance on manual intervention. It also enhances capabilities in image analysis, instruction following, and the generation of documents and presentations, and is considered to demonstrate stronger creativity.

However, Anthropic noted that Opus 4.7 does not push the company’s capability boundaries further, and its overall performance still falls short of the previously released Claude Mythos Preview model, which outperforms it across multiple benchmarks.

For safety reasons, Anthropic is currently providing Claude Mythos Preview only to selected partners, including NVIDIA, JPMorgan Chase, Google, Apple, and Microsoft. Opus 4.7, as a public model, is being used to test new cybersecurity defense mechanisms and includes additional safety restrictions. The company has also introduced a Cyber Verification Program, allowing security researchers to conduct vulnerability research under specific conditions.

Pricing remains unchanged at $5 per million input tokens and $25 per million output tokens. However, Anthropic noted that Opus 4.7 features an upgraded tokenizer, meaning the same text may consume 1.0 to 1.35 times more tokens than before; under high-reasoning settings, especially in multi-turn interactions, the model produces deeper reasoning and longer outputs. Source

---

OpenAI upgrades Codex with multiple practical features

On April 17, OpenAI announced an upgrade to Codex, further enhancing its agent-based development capabilities. The new Codex can directly operate desktop applications, execute tasks in the background, and allow multiple agents to work in parallel, making it suitable for front-end debugging, application testing, and development workflows without APIs. It also introduces an in-app browser, image generation and editing capabilities, memory features, and support for plugins such as GitLab Issues, Atlassian Rovo, and Microsoft Suite.

According to OpenAI, the new “computer operation” capability enables agents to perform actions such as clicking and typing on the user’s computer without interfering with other applications. The built-in browser supports web browsing and page annotation, making it easier for developers to provide precise instructions. The image capabilities, powered by gpt-image-1.5, can be used for generating and iterating on product prototypes, UI designs, and game assets.

In addition to feature expansion, Codex is beginning to introduce “memory,” which can store user preferences, historical modifications, and common workflows to improve efficiency in future tasks. These personalization features will be gradually rolled out.

These updates have begun rolling out to users of the Codex desktop app within ChatGPT. At the initial stage, desktop operation features are only available on macOS. Support for Enterprise, Edu, as well as users in the EU and UK, will be added in future updates. Source

---

DJI launches Osmo Pocket 4 gimbal camera

On April 16, DJI introduced the new-generation pocket gimbal camera, Osmo Pocket 4.

The device features a new 1-inch CMOS sensor combined with an ƒ/2.0 large aperture, achieving 14 stops of dynamic range and supporting 10-bit D-Log professional color mode. In terms of video capabilities, it supports up to 4K recording and 240 fps slow motion, while also adding spatial audio recording and zoom-based audio pickup. It supports direct connection to DJI microphone transmitters via OsmoAudio, enabling four-channel audio recording.

On the system side, Osmo Pocket 4 is equipped with the Intelligent Tracking 7.0 system, supporting up to 4× distance tracking. The autofocus system has been upgraded with new “Subject Lock Tracking” and “Registered Subject Priority” modes, along with gesture control support. For imaging, the device includes optimized skin tone rendering and adjustable beauty filters. It also supports external fill lights with adjustable color temperature and brightness.

The product is available in Standard and Creator Combo versions, priced at RMB 2,999 and RMB 3,799 respectively. The Creator Combo additionally includes a DJI Mic 3 transmitter, fill light, wide-angle lens, and mini tripod. DJI has also introduced the DJI Care Refresh service, priced at RMB 219 for one year (including two replacement claims) and RMB 349 for two years (including four replacement claims). Source

At the same time, DJI announced plans to release a dual-camera version, Pocket 4P. Source

---

Amazon introduces the thinnest Fire TV Stick HD

On April 15, Amazon officially launched the new-generation Fire TV Stick HD streaming device, emphasizing a slimmer design and improved performance. The device supports 1080p video output, is priced at $34.99, and is now available for pre-order, with shipments expected to begin on April 29 across multiple markets.

According to official information, the new Fire TV Stick HD features a significantly optimized design, with overall thickness reduced by about 30% compared to the previous generation. Amazon describes it as the “thinnest streaming device ever.” It can be powered directly via a TV’s USB port, or through a USB-C cable with a power adapter for TVs without USB ports.

In terms of performance, the new device delivers an average improvement of over 30% compared to the previous HD model, with optimizations in startup speed and app loading times. It also supports Wi-Fi 6 and Bluetooth 5.3 standards. On the software side, the device runs Vega OS, based on the Linux kernel, and integrates the Alexa+ voice assistant for more natural voice interactions. The interface adopts Amazon’s updated content layout, categorizing movies, live TV, sports, and news into distinct sections.

The device will initially launch in the United States, United Kingdom, Canada, Mexico, Japan, Australia, and New Zealand, with expansion to additional European markets planned. Source

Adobe releases Firefly AI assistant

On April 15, Adobe announced the Firefly AI Assistant. According to the company, this is an AI creative assistant with agent capabilities, capable of executing multi-step tasks across various Creative Cloud applications.

Unlike traditional AI tools that rely on step-by-step instructions, Firefly AI Assistant adopts a goal-oriented task execution model. After users describe their needs in natural language, the system can automatically plan workflows and perform multi-step operations across applications such as Adobe Photoshop, Adobe Premiere Pro, Adobe Lightroom, Adobe Illustrator, and Adobe Express.

Adobe stated that the assistant will provide a unified conversational interface to manage task context and synchronize generated results across applications. It also includes preset creative functions, such as completing image style adjustments with a single prompt, simplifying common workflows.

Firefly AI Assistant also offers a degree of personalization, gradually adapting output styles based on user history. Additionally, it integrates Frame.io’s review features, enabling users to organize project feedback and share it with collaborators, while allowing external participants to submit actionable revision suggestions.

Firefly AI Assistant has not yet been officially released. Adobe plans to roll out a public beta to test users in the coming weeks. Source

---

Tencent unveils Hunyuan 3D World Model 2.0

On April 16, Tencent announced the official release and open-sourcing of the Hunyuan 3D World Model 2.0 (HY-World 2.0). This model is a multimodal world model capable of automatically generating, reconstructing, and simulating 3D scenes based on text, images, and video inputs.

According to Tencent, HY-World 2.0 supports exporting multiple 3D asset formats, including Mesh, 3D Gaussian Splatting (3DGS), and point clouds. It can also integrate with existing game development workflows, enabling rapid generation of maps and level prototypes. In terms of performance, Tencent claims improvements in scene completeness (such as generating object sides and backs) and fidelity to input content.

Additionally, the model adopts a hybrid representation combining 3DGS and Mesh, allowing generated scenes to support realistic collision-based interactions. The upgraded HY-Pano 2.0 model introduces an end-to-end implicit learning approach, capable of converting standard images into 360-degree panoramas without requiring camera parameters.

For training, Tencent stated that the model combines real panoramic images with synthetic data generated via Unreal Engine (UE) to enhance quality and generalization. The model and related technical materials have now been open-sourced on GitHub. Source

---

Mastercard enables cross-border Apple Pay support for Chinese cardholders

On April 16, Mastercard announced progress in its collaboration with NetsUnion Clearing Corporation, a domestic bank card clearing institution in China: Mastercard-branded bank cards issued within mainland China now support cross-border payments via Apple Pay.

Currently, Mastercard single-branded or dual-branded credit cards issued by Bank of China, Agricultural Bank of China, China CITIC Bank, and Shanghai Pudong Development Bank, as well as Mastercard debit cards issued by China CITIC Bank, are supported for Apple Pay binding. In terms of usage, users can add their bank cards to Apple Wallet (Apple Pay) through the latest version of their bank’s app, or directly tap the “+” button in the iPhone Wallet app and select a credit card to complete the setup. Once added, users can make payments using Apple Pay on iPhone, Apple Watch, or iPad. Source

---

Apple Wallet now supports NFC transit cards via Alipay

On April 14, Alipay announced that Apple Wallet now supports adding NFC transit cards via Alipay, covering cities including Beijing, Shanghai, Nanjing, Changsha, Xiamen, Suzhou, Kunming, Qingdao, Shijiazhuang, and Tianjin. Users can tap the “+” button in Apple Wallet to add a transit card, select “Transit Card” – “Alipay,” confirm opening Alipay, and complete the setup. The corresponding city transit card will then appear in Apple Wallet. Source

News Worth a Quick Look

• Google Quick Share has recently encountered a series of cross-platform transfer issues.
  • Some Pixel 10 users reported that upon opening the Quick Share interface, their devices immediately disconnect from Wi-Fi, making it impossible to display available networks. According to user feedback, uninstalling related extension updates can temporarily alleviate the issue. The problem has also appeared on Google Issue Tracker, but the related entry was quickly closed, and users were directed to continue reporting through official support forums. As of now, no clear timeline for a fix has been announced. Source
  • Samsung Galaxy users have reported that when transferring photos to iPhone via Quick Share, EXIF metadata such as location information is not fully preserved. A moderator on Samsung’s forum has confirmed the issue and stated that a fix is currently in development, expected to be resolved in a future software update. Source
• Canon has introduced new additions to its Cinema Servo lens lineup: the CN30×40 IAS J/R1 and CN30×40 IAS J/P1, equipped with RF and PL mounts respectively. While maintaining portability, both lenses achieve the longest focal length in the series, offering 1200 mm ultra-telephoto reach and up to 30× optical zoom, covering a range from 40 mm to 1200 mm. With the built-in 1.5× extender enabled, the focal length can be extended to 1800 mm, and the lenses support cameras equipped with full-frame 35 mm sensors. In terms of features, the RF mount version supports Dual Pixel CMOS autofocus and focus guide functions, reducing operational complexity; when paired with the EOS C400 camera, it also supports automatic exposure ramping compensation to minimize brightness shifts during zooming. Additionally, the lenses feature focus breathing correction and support virtual production systems, making them suitable for various professional filmmaking scenarios. Both models adopt newly developed drive units and include multifunction USB Type-C interfaces to enhance control and expandability. The new lenses are scheduled for release in late September 2026. Source