Expanding AI Horizons: Open AI Launches in Japan

Blackmagic’s DaVinci Resolve 19 arrives with AI-powered motion tracking and colour grading

AI News Blackmagic Design unveiled a series of new products at NAB 2024, headlined by an updated version of the widely-used DaVinci Resolve editing suite. Highlights from the release include the Micro Color Panel for iPad integration, a groundbreaking 17K 65mm camera, and the versatile Pyxis 6K cube camera.

Davinci Resolve 19

AI News Blackmagic Design has introduced version 19 of its DaVinci Resolve editing suite, loaded with innovative AI-powered features, marking its latest update at NAB 2024. The new release offers editors a subscription-free alternative to Adobe Premiere Pro, with advanced capabilities in effects, colour correction, audio, and more.
In the Edit module, a standout feature allows video clips to be edited through text transcriptions. This feature displays detected text from multiple speakers, enabling easy section removal, text searches, and more. Enhancements include a new trim window, a fixed play head to minimise zooming and scrolling, and a streamlined window for adjusting audio attributes.
The Color tool now includes “Color Slice,” which adjusts images based on six color vectors—red, green, blue, yellow, cyan, and magenta—plus a specialised skin tone slider for precise skin adjustments. The “IntelliTrack” feature, powered by a neural engine AI, enables precise point tracking for effect application or image stabilization.
Noise reduction has also been redefined with an AI-driven tool that removes digital noise or film grain effortlessly. The “Film Look Creator” module offers over 60 filmic parameters for customizable color grading, starting with presets like default 65mm and cinematic looks, allowing for fine-tuning to achieve desired effects.
Additionally, the “Defocus Background” feature allows for realistic depth of focus simulations, and Face Refinement tracks facial features to adjust brightness, color, and detail. The Fusion FX editor has been enhanced for simpler 3D object manipulation.
On the audio side, the Fairlight module’s new “Dialogue Separator FX” separates dialogue from background noise. DaVinci Resolve 19 is now available in open beta, with the Studio version priced at $295 and a free main version, continuing Blackmagic’s approach of making professional-grade editing tools accessible to a broader audience.

Micro Color Panel

AI News BMD’s DaVinci Resolve for iPad proved to be a popular option for editors on the go, and now the company has introduced a dedicated control surface with the new Micro Color Panel. It’ll offer editors control that goes well beyond the already decent Pencil and multitouch input, while keeping a relatively low profile at 7.18 x 14.33 inches.
A slot at the top front lets you slide in your iPad, and from there you can connect via Bluetooth or USB-C. The company promises a “professional” feel to the controls, which consist of three weighted trackballs, 12 control dials and 27 buttons. With those, you can perform editors, tweak parameters like shadows, hues and highlights, and even do wipes and other effects.
“The old DaVinci Resolve Micro Panel model has been popular with customers wanting a compact grading panel, but we wanted to design an even more portable and affordable solution,” said Blackmagic Design President Grant Petty. It’s now on pre-order for $509.

Pyxis 6K camera

AI News Blackmagic Design is set to challenge industry giants like RED, Sony, and Panasonic with its innovative Pyxis 6K full-frame box-style camera. Designed for versatility, the camera features a modular “brain” which includes controls, a display, CFexpress media, and a 36 x 24mm sensor. Users can customize their setup by attaching various accessories like handles, microphones, and SSDs using side plates or mounting screws.
The Pyxis 6K builds on the success of the Blackmagic Cinema Camera 6K, offering similar specs including a 24-megapixel resolution that supports up to 13 stops of dynamic range and dual native ISO up to 25,600. It records in 12-bit Blackmagic RAW (BRAW) directly onto CFexpress Type B cards or an external SSD.
A standout feature is its built-in capability for direct streaming to platforms like YouTube, Facebook, and Twitch via RTMP and SRT, either through Ethernet or a cellular connection. This integration allows users to monitor stream status and data rates directly from the camera’s viewfinder or LCD.
Coming in June at a price of $2,995, the Pyxis 6K offers three lens mount options: Canon EF, Leica L, and Arri PL. It also pairs with Blackmagic’s URSA Cine EVF, an optional electronic viewfinder, adding $1,695 to the total cost.

Blackmagic URSA Cine 12K and 17K

AI News Blackmagic has expanded its lineup with two new cinema cameras, the URSA Cine 12K and 17K models, showcasing groundbreaking resolutions and advanced features. The URSA Cine 12K features a full-frame 36 x 24mm sensor, delivering 12K resolution (12,888 x 6,480 in a 17:9 aspect ratio) at an impressive 100 frames per second. The URSA Cine 17K steps it up with a 65mm sensor (50.8 x 23.3mm), boasting a resolution of 17,520 x 8,040 and up to 16 stops of dynamic range.
Both cameras are equipped with professional-grade features including built-in ND filters, an optical low pass filter, and the latest generation 5.0 color science from Blackmagic Design. The URSA Cine 12K offers 8TB of internal storage and supports additional storage via CFexpress media. Additional features include live streaming capabilities, a high-resolution electronic viewfinder (EVF), support for V-mount batteries, and wireless Bluetooth for remote camera control.
The URSA Cine 12K is currently available for pre-order at $14,995, or $16,495 with the URSA Cine EVF, and is set to ship in April. The URSA Cine 17K is still under development, with its pricing and release date yet to be announced. SOURCE 

.

Introducing Idefics2: A Powerful 8B Vision-Language Model for the community

AI News We’re thrilled to announce the launch of Idefics2, our latest multimodal model designed to handle diverse inputs of text and images seamlessly, producing text responses. This advanced model is capable of interpreting images, answering visual questions, describing content, crafting stories from multiple images, extracting document information, and even performing simple arithmetic.
Building on the success of Idefics1, Idefics2 features 8 billion parameters, is released under the open Apache 2.0 licence, and boasts improved Optical Character Recognition (OCR) capabilities, making it a robust tool for researchers and developers in the field of multimodal AI. It delivers exceptional performance on Visual Question Answering benchmarks, rivalling larger models like LLava-Next-34B and MM1-30B-chat.
Fully integrated with the 🤗 Transformers library, Idefics2 is ready for immediate fine-tuning across various multimodal tasks. Explore and experiment with the models today on the Hub!

Training Data

AI News Idefics2 underwent pretraining using a diverse array of openly accessible datasets, including web documents from Wikipedia and OBELICS, image-caption pairs from the Public Multimodal Dataset and LAION-COCO, OCR datasets like PDFA (en), IDL, and Rendered-text, as well as image-to-code data from WebSight. Users can also explore the OBELICS dataset through an interactive visualization tool.
In line with standard practices within the foundation model community, we further enhanced the base model by training it on task-specific data that often comes in varied formats and from multiple sources, presenting a significant challenge in terms of data aggregation. To overcome this, we have developed and are now releasing The Cauldron, an open resource compiling 50 meticulously curated datasets designed for multi-turn conversations. Idefics2 was fine-tuned on a combination of The Cauldron and several text-only instructional datasets, enriching its multimodal capabilities.

Improvements over Idefics1

AI News Our approach utilizes images in their original resolutions (up to 980 x 980) and aspect ratios, adopting the NaViT strategy that eliminates the traditional requirement of resizing images into fixed-size squares. We also incorporate techniques from SPHINX, which optionally allows for sub-image splitting and handling of very high-resolution images.
We’ve greatly improved OCR capabilities by integrating data that demands transcription of text from images or documents. This enhancement extends to more accurately answering questions about charts, figures, and documents using specifically tailored training data.
Moving away from the architecture of Idefics1, which used gated cross-attentions, we’ve streamlined the integration of visual features into the language backbone. Images first pass through a vision encoder, followed by learned Perceiver pooling and an MLP modality projection, creating a pooled sequence. This sequence is then interleaved with text embeddings to form a combined stream of image and text data.
These advancements, along with superior pre-trained backbones, have led to a significant performance improvement over Idefics1, despite the model being ten times smaller. SOURCE 

Introducing Open AI Japan

AI News We’re thrilled to announce the opening of our first office in Asia, located in Tokyo, Japan. This expansion marks a significant step for Open AI as we grow our international presence. Tokyo was chosen for its global leadership in technology and a culture that embraces innovation, making it the ideal location for our first Asian hub.
To lead our operations in Japan, we are pleased to welcome Tadao Nagasaki as the new President of OpenAI Japan. Under his guidance, we aim to forge strong partnerships with the Japanese government, local businesses, and research institutions to create AI tools tailored to Japan’s unique needs and opportunities.
In conjunction with our new office, we are excited to offer early access to our GPT-4 custom model, specially optimized for the Japanese language. This model enhances the efficiency and effectiveness of translating and summarizing Japanese text, offering up to three times the speed of its predecessor at a lower cost.
Our initial applications have been promising. For example, Speak, a leading English learning app in Japan, has reported a 2.8x increase in the speed of tutor explanations and a 47% reduction in token costs, significantly enhancing the learning experience.
Moreover, our presence in Japan has enabled us to collaborate closely with major companies like Daikin, Rakuten, and TOYOTA Connected, who are utilizing ChatGPT Enterprise to streamline complex business processes and improve data analysis and internal reporting. Local governments, including Yokosuka City, are also adopting our technology to boost public service efficiency, with 80% of city employees noting a productivity increase.
This expansion aligns with our commitment to contribute to the global discourse on AI policy, supporting Japan’s efforts to address societal challenges such as rural depopulation and labour shortages through AI innovations.
As we continue to grow globally, we remain dedicated to learning from diverse perspectives to ensure that AGI benefits all of humanity. For those interested in joining our expanding team, please visit our Careers page for current opportunities. SOURCE 

Conclusion:

Blackmagic Design’s unveiling of DaVinci Resolve 19 at NAB 2024 signifies a major leap forward in video editing technology. With its host of AI-powered features and innovative hardware additions, Blackmagic continues to set the standard for affordable, high-quality film production tools. The introduction of advanced motion tracking, color grading capabilities, and seamless iPad integration via the Micro Color Panel highlights their commitment to enhancing the flexibility and efficiency of the editing process. Additionally, the new Pyxis 6K camera and updates to the URSA Cine series underscore Blackmagic’s dedication to pushing the boundaries of camera technology, offering creators powerful tools that rival those of industry giants. This blend of cutting-edge technology and accessible pricing ensures that Blackmagic Design remains at the forefront of the digital cinema industry, democratizing professional-grade filmmaking and continuing to innovate in ways that empower filmmakers worldwide.
Stay connected with Arcot Group for more AI News into how such collaborations are reshaping the tech landscape and paving the way for future innovations. For further reading on similar breakthroughs and the impact of AI and robotics, explore our blog.