Google Drops Gemini 3.1, AI Music & PhotoShoots (Plus: Magnific Video & Midjourney v8!)

Overview of Recent AI Developments

Google’s New Releases

• The video discusses significant updates from Google, including Gemini 3.1, LIIA 3 (a music generator), and a new tool called Photoshoot.

• While many anticipated the release of V4, it has not yet been announced; however, there may be a target date in the future.

Gemini 3.1 Pro Features

• Google recently launched Gemini 3.1 Pro, which builds on the previous deep think update and enhances core reasoning capabilities.

• The model achieved a score of 77.1 on the ARC AGI 2 benchmark, indicating superior fluid intelligence compared to average human test scores (60-66%).

Performance Metrics

• In "humanity's last exam," Gemini 3.1 Pro scored 44.4% without tools and 51.4% with search and code features, showing significant improvement over its predecessor.

• Comparatively, Claude's Opus scored lower at 40% without tools and 53% with them.

Exploring Gemini’s Capabilities

Accessing Gemini

• Users can access Gemini through the app for pro and ultra plan subscribers; however, version details are often vague.

Practical Applications

• A demonstration involved building a Missile Command clone using simple prompts; the model suggested enhancements like sound effects during development.

Updates on LIIA Music Generator

Music Generation Features

• LIIA currently generates songs that are limited to 30 seconds but is expected to expand in length in future updates.

Multimodal Composition

• Unique to LIIA is its ability to compose music based on images; users can prompt compositions by describing visual elements.

Creative Examples from LIIA

Song Creation Demonstrations

• An example song was generated about a man in a blue suit jaywalking while avoiding police tickets, showcasing playful lyrics reminiscent of late '90s hip-hop.

Image-Based Song Generation

• Another example involved generating a song based solely on an image titled "Flamethrower Girl," illustrating how LIIA interprets visuals into musical themes.

What’s New in Music and AI Tools?

Updates on Liia and Personal Music Aspirations

• The speaker expresses interest in Liia, hinting at future developments and personal aspirations to engage with music again, noting that guitars have been unused for too long.

Introduction of Photo Shoot Feature in Pali

• A new feature called "photo shoot" is introduced, aimed primarily at branding and marketing professionals. It allows users to create product photo shoots from basic images.

Demonstration of Photo Shoot Capabilities

• The speaker uploads a poor-quality image of a USB hub as a test case. After using the default templates, the tool generates four improved images from the original.

Potential Applications for Selling Products

• Users can leverage this tool to enhance images of items they wish to sell online, suggesting it could lead to successful sales on platforms like Facebook Marketplace.

Campaign Creation with Pomelo

• The speaker demonstrates how Pomelo can generate an entire ad campaign based on the enhanced images created from the initial upload, emphasizing its utility for e-commerce.

Upcoming Features and Enhancements

Anticipation for V4 Release

• Speculation arises regarding the release date of V4 during Google IO on May 19th. The speaker notes this event typically features significant announcements from Google.

Magnific's Video Upscaling Feature

• Magnific introduces video upscaling—a highly requested feature since its launch—allowing users to upscale videos alongside existing image upscaling capabilities.

User Interface Overview for Magnific

• The layout includes options for natural or vivid upscaling, resolution choices (1K, 2K, 4K), premium quality settings, FPS boost options, and sliders for sharpening and grain adjustments.

Performance Observations

• Initial tests show effective frame rate increases; however, there are concerns about heavy contrast effects when using certain settings.

Creative Outputs and Quirks

• While Magnific performs well with CGI content, it occasionally produces unexpected results such as misinterpreting elements within frames or generating faces where none exist.

Witcher 4 and Upscaling Techniques

Overview of CG Characters in Witcher 4

• The speaker discusses the effectiveness of upscaling techniques for computer-generated (CG) characters, specifically referencing a screen grab from the upcoming Witcher 4 trailer.

• Notably, the character Siri shows minimal face changes, indicating that high creativity settings can still yield satisfactory results.

Exploring Different Presets for Animation

• The discussion shifts to various presets available for animation, including options like 3D, realistic, artistic, and custom settings.

• An example is provided where footage captured in World Labs is processed through an upscaler called Magnific, demonstrating significant improvement in quality.

Utilizing Marble as a Virtual Set

• The speaker highlights using Marble as a virtual set to capture different angles and integrate characters into scenes effectively.

• With the anticipated release of Cance 2 at only 720p resolution, having an upscaler is emphasized as essential for enhancing visual quality.

MidJourney V8 Release Insights

Anticipation for MidJourney V8 Features

• MidJourney is nearing its V8 release; early previews suggest improvements aimed at community feedback and expectations.

• Key features include corrected text generation and enhancements in prompt understanding and coherence—common themes across model updates.

New Editing Tools and Video Model Developments

• A new editor aligned with modern image editing styles is expected alongside improved image reference capabilities.

• Additionally, a new video model (V2), larger than its predecessor V1, will be introduced later this year.

Conclusion on Timing Expectations

• While MidJourney V8 could launch soon (potentially next week), the speaker notes that timing may vary based on internal schedules.