Ideogram 3.0 Arrives With New Graphic Design Capacities

⚡️ Headlines

🤖 AI

Amazon unveils new AI features to enhance shopping discovery and personalized browsing - The company is rolling out AI tools like "Browse with Amazon" to tailor product recommendations and enrich customer experiences. [About Amazon]

OpenAI's Studio Ghibli-style image generator sparks AI copyright debate - A viral image showing AI mimicking Studio Ghibli’s animation style reignites questions around fair use and derivative content. [TechCrunch]

BMW and Alibaba partner to power smart in-car systems with Qwen’s AI in China - The collaboration will integrate Qwen’s AI to enhance voice recognition and digital assistant functions in BMW vehicles. [Alizila]

Qwen 2.5 Omni launches with multilingual capabilities and strong benchmark performance - The latest Qwen model supports over 30 languages and excels at reasoning, vision-language tasks, and tool use. [Qwen Blog]

Garmin introduces Connect Plus subscription with AI training plans and recovery tracking - The new $10/month service adds advanced fitness insights and generative AI for personalized coaching. [The Verge]

🦾 Emerging Tech

Dubai's VARA prioritizes consumer protection in tokenization push - Regulators emphasize safeguards as Dubai advances its framework for asset tokenization. [CoinDesk]

🤳 Social Media

Utah enforces law requiring app stores to verify users' ages - App platforms must now confirm user age or risk delisting under the state’s new online child safety law. [Social Media Today]

YouTube updates Shorts view count rules to address creator complaints - The platform will discount replays in view totals, aiming to better reflect original engagement. [TechCrunch]

🎱 Random

JPMorgan claims quantum experiment produced truly random numbers - The breakthrough could enhance encryption and financial modeling through improved randomness. [Bloomberg]

Krisp AI adds real-time accent conversion to voice calls - The tool lets users modify their accents live during conversations while preserving identity. [The Verge]

🔌 Plug-Into-This

Ideogram 3.0

Ideogram 3.0 introduces a new generation of text-to-image capabilities focused on photorealism, precise layout, and stylistic control. Now available via its web platform and iOS app, the model demonstrates major improvements in prompt alignment and legible, artistic text rendering—making it suitable for both creative and professional design use cases.

• A major new feature is Style References, which lets users upload up to three example images to guide generations toward a consistent aesthetic, enabling visual continuity across designs.

• The platform includes a “Random style” tool that draws from over 4.3 billion style presets, allowing for discovery and reuse of unique looks via persistent Style Codes.

• Ideogram 3.0 excels at accurate, layout-aware text rendering, handling complex typography and positioning that previous models typically failed at, particularly in marketing, editorial, and cinematic poster formats.

• Human evaluations placed it at the top of the ELO rating system across varied prompts and subjects, outperforming other leading text-to-image models in overall quality and versatility.

• Real-world examples showcased stylized fashion posters, cinematic layouts, and typographic book covers—underscoring the model’s readiness for use cases in publishing, advertising, and branding.

🎨 Ideogram 3.0 positions itself not just as a generator of beautiful images but as a design-native tool—bridging generative AI with the workflows of visual professionals who demand both creativity and consistency.

Putting Gemini 2.5 Pro through its paces

In his latest blog post, Simon Willison shares hands-on experiments with Google’s Gemini 2.5 Pro, evaluating its performance across a range of tasks from image generation to audio transcription. Willison integrates the model with his LLM command-line tool and explores its reasoning, visual, and coding abilities, concluding it’s a remarkably capable release that may justify its top spot on the LM Arena leaderboard.

• Willison tested image generation using his benchmark “pelican riding a bicycle” prompt, finding Gemini’s SVG output amusingly effective, arguably better than Claude 3.7 Sonnet’s previous best.

• The model's transcription capabilities impressed him, especially on multilingual audio featuring both English and Spanish, generating structured JSON with accurate timestamps and language metadata.

• Gemini 2.5 Pro handled custom schema formats with precision, extracting speaker names and aligning them correctly to transcribed podcast dialogue, a task many models fumble.

• Willison appreciated the ease of use and stability when running the model via llm-gemini, which he updated to support this new version, offering reproducible CLI workflows for deep testing.

• Though he didn’t conduct formal benchmarks, his anecdotal findings across varied modalities led him to endorse Gemini 2.5 Pro as “a very strong new model” with meaningful real-world utility.

🏁 Willison’s evaluation reinforces Gemini 2.5 Pro’s position as a credible frontrunner in the current AI model race—not just matching competitors in benchmarks but offering fluid multimodal performance, strong tooling integration, and developer-accessible reliability that few frontier models balance this well.

Introducing Researcher and Analyst in Microsoft 365 Copilot

Microsoft has unveiled two new reasoning agents—Researcher and Analyst—within Microsoft 365 Copilot, aimed at transforming how professionals engage with information and data. These agents are powered by OpenAI models and integrated directly into Microsoft’s productivity ecosystem, delivering advanced research and analytical workflows grounded in a user’s enterprise data.

• Researcher synthesizes complex information across internal files, emails, meetings, and external web sources to assist with strategic tasks such as market analysis, whitespace identification, and report creation.

• It supports third-party connectors (like Salesforce and ServiceNow), extending its data reach and enriching outputs with broader context and competitive insights.

• Analyst uses OpenAI’s o3-mini model and chain-of-thought reasoning to simulate data science workflows, capable of writing and executing Python code for advanced queries and visualizations.

• Both agents are part of a new “Frontier” program rolling out in April, giving early access to innovations still in development within Microsoft 365 Copilot.

• Microsoft Copilot Studio is also expanding with deep reasoning capabilities and autonomous agent flows, enabling businesses to build task-specific agents that operate independently on enterprise data.

🧩 These additions elevate Microsoft 365 Copilot from a productivity assistant to a modular reasoning platform, signaling Microsoft’s ambition to dominate enterprise AI by embedding domain-specific intelligence directly into everyday business workflows.

 🆕 Updates

📽️ Daily Demo

🗣️ Discourse