Google Vids, Sakana AI Orchestration, and Germany’s AI Regulation Set New Standards

Google Vids, Sakana AI Orchestration, and Germany’s AI Regulation Set New Standards

Alex Rivers

Alex Rivers
Senior AI Journalist

Google Vids Embraces Personalized AI Avatars to Challenge Video Creation Market

Google is making a bold move into the personalized AI video space with a major overhaul of its Vids platform. The tech giant announced on Thursday that users can now create custom digital avatars that look and sound like them, requiring only a selfie and voice recording. This advancement represents a significant step beyond the original role of Vids as a workplace presentation tool, positioning it as a full-fledged video creation platform that competes directly with specialized AI video startups.

The announcement comes despite OpenAI’s Sora shutting down earlier this year, suggesting that Google sees lasting demand in the personal AI video creation space. Beyond avatars, Google is integrating its multimodal Gemini Omni model into Vids, allowing users to create videos from written prompts combined with reference images. The system can intelligently mix these inputs to generate the desired video output, and now supports step-by-step edits so creators can refine their work iteratively rather than starting from scratch each time.

Google has implemented security measures to prevent misuse, including tying avatars to account holders’ likenesses and watermarking them invisibly with SynthID technology. The company notes that access to personal avatars is limited to users in certain regions who are 18 or older. These updates position Vids more competitively against alternatives like HeyGen, Synthesia, Captions, and D-ID, while leveraging Google’s existing Workspace integration for enterprise adoption.

Advertisement

Source: TechCrunch

Sakana AI Proves Orchestrated Open Models Can Match Frontier Performance

Tokyo-based startup Sakana AI is challenging the dominance of single large frontier models with an innovative orchestration approach. The company has added Nvidia’s open Nemotron models to its Fugu orchestrator, a system designed to prove that coordinated open models can deliver performance comparable to proprietary frontier systems. Fugu operates as a master language model that intelligently selects and combines other LLMs from a diverse agent pool, including instances of itself, to tackle complex tasks dynamically.

The Fugu system’s modularity is a key strength. New models can be added at any time without depending on any single provider’s strengths or worrying about service outages. In Sakana AI’s benchmarks, Fugu Ultra performed on par with Anthropic’s Fable 5 and Mythos Preview, though early independent tests showed mixed results regarding speed and cost efficiency. By integrating Nvidia’s Nemotron family—which specializes in coding, tool calling, and instruction following—Sakana AI fills specialist roles within Fugu’s orchestration layer rather than relying on frontier models exclusively.

Nvidia’s Nemotron 3 Ultra, featuring approximately 550 billion parameters with 55 billion active parameters, has become the smartest open U.S. model to date according to benchmark platform Artificial Analysis. It surpasses Google’s Gemma 4 31B and OpenAI’s gpt-oss-120b, though Chinese models like Kimi K2.6 still lead in some benchmarks. Sakana AI frames this partnership as part of a broader shift where progress in AI depends increasingly on how well models can be evaluated, combined, and integrated into real-world workflows. The startup, founded by former Google researchers including Llion Jones (co-author of the Transformer paper), positions open model orchestration as a hedge against regulatory restrictions and vendor lock-in.

Source: The Decoder

Germany Classifies AI Search Engines as Content Providers, Setting Global Precedent

German media regulators have issued historic rulings against Google and Perplexity, classifying AI search engines and chatbots as content providers subject to the country’s State Media Treaty. This marks the first time German regulators have applied media law to AI services, potentially setting a global precedent for how governments will regulate generative AI systems. The Commission for Licensing and Supervision (ZAK) determined that AI-generated responses constitute original content created by the service providers, not neutral search results protected by the Digital Services Act’s liability shield.

Regulators accuse Google of violating transparency rules and discriminating against traditional search links by placing AI summaries prominently above conventional results. This positioning pushes journalistic sources down the page, effectively burying them beneath Google’s own generated content. Studies indicate that users rarely click source links after receiving summarized answers, meaning repositioning links alone is unlikely to restore traffic to news outlets. Google includes source links in AI Overviews but disputes claims that this damages publisher traffic, though it has not released data to support its position. In Perplexity’s case, regulators have flagged missing transparency disclosures and lack of a designated German representative.

The rulings formally find violations of Section 109 of the State Media Treaty and are immediately enforceable, giving both companies one month to appeal. A related legal opinion from professors Jan Oster and Christoph Busch recommends creating a separate regulatory category for AI search engines under media law, with specific rules to protect media diversity. This German action may influence how other jurisdictions approach AI regulation, particularly in Europe where the Digital Markets Act and other regulations already constrain tech giant operations. Google has partly prepared with features like “Preferred Sources,” allowing users to customize which outlets appear in AI responses, though adoption is expected to remain low among regular users.

Source: The Decoder

What to Read Next

Bookmark aistackdigest.com for daily AI tools, reviews, and workflow guides.

This article was produced with the assistance of AI tools and reviewed by the AIStackDigest editorial team.

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top