Google's Nano Banana Pro Solves AI's Biggest Text Problem 🤯

In partnership with

Welcome to another edition of Horizon AI,

Google's upgraded Nano Banana Pro image model is being praised for its realism and control, particularly for its breakthrough ability to generate high-quality, perfectly legible text within the image output.

Let’s jump into it!

Read Time: 4.5 min

Here's what's new today in the Horizon AI

• Google's Nano Banana Pro Solves Image-Text Problem 🤯

• Samsung's Push for True On-Device AI Intelligence

• AI Tutorial: How to Create a Professional Email Template Using AI

• AI Tools to check out

• AI Findings/Resources

• The Latest in AI and Tech 💡

AI News

GOOGLE

Google's Nano Banana Pro Solves Image-Text Problem

Google has launched its highly anticipated image generation and editing model, Nano Banana Pro, built on Gemini 3 Pro's advanced reasoning, and it's being hailed as a monumental leap forward for AI imagery. The model's key breakthrough addresses the long-standing industry challenge of generating accurate, legible text directly within image outputs.

Details:

• Nano Banana Pro is designed to be the best model for creating images with correctly rendered and legible text (including long paragraphs) directly embedded in the visual.

• Due to its foundation in Gemini 3, the model connects to Google Search's knowledge base, allowing it to generate context-rich content like accurate infographics and diagrams.

• Users gain studio-quality control, able to adjust elements like camera angles, focus, scene lighting (day to night), and output final results in high-resolution 2K and 4K.

• The model supports blending up to 14 reference images to maintain a consistent style, branding, or character likeness across an entire composition or campaign.

• A new verification tool in the Gemini app lets users upload an image and ask whether it was generated by Google AI, thanks to an embedded, imperceptible SynthID digital watermark.

Source

TOGETHER WITH NEBIUS TOKEN FACTORY

Cut Your LLM Inference Costs by 70% Without Losing Performance

Running large language models in production is draining budgets—but it doesn't have to.

This on-demand webinar from Nebius reveals how model distillation can slash inference costs by up to 70% while maintaining enterprise-grade accuracy.

Watch a live demo covering the complete workflow: data prep, distillation, verification, and production deployment with SLAs and compliance guarantees.

Learn distillation vs. fine-tuning, see real benchmarks, and get a step-by-step playbook you can apply immediately.

SAMSUNG

Samsung's Push for True On-Device AI Intelligence

Samsung Research is leading efforts in model compression and runtime optimization to deliver cloud-level AI performance directly to devices like smartphones. The goal is to maximize efficiency and security within the constraints of limited hardware.

Details:

• Quantization is the primary method for reducing the size of massive LLMs to run efficiently on devices. This involves simplifying complex calculations, such as converting 32-bit floating-point values to more efficient 8-bit or 4-bit integers, to reduce memory use and computational load significantly.

• The "engine control unit" optimizes how a device's resources (CPU, GPU, NPU) are used during execution.

• It intelligently assigns operations to the optimal chip and minimizes memory access to boost overall AI performance and enable larger models to run at the same speed.

• Researchers are focused on redesigning the fundamental blueprints of AI systems to build device-friendly structures from the ground up, maximizing hardware efficiency.

• This work explores new methodologies beyond the standard transformer architecture to effectively handle the constraints of on-device environments.

• The ultimate goal is to achieve cloud-level performance directly on the device, ensuring the AI is fast, accurate, and power-efficient while enhancing real-time learning and user data privacy.

Source

AI Tutorial

How to Create a Professional Email Template Using AI

1. Go to MailUI.co, Log in and open the Email Builder.

2. Pick a ready-made template OR Start from a Blank Email.

3. Use drag-and-drop blocks to add structure for the  Header, Body, and Footer.

4. Customize the Design by setting your brand colors & fonts, adding images, and styling your CTA button. Keep the layout clean and simple.

5. Write Your Content with a short subject line and preheader, a clear headline, and brief, scannable text. You can also add a single main call to action.

6. Add Personalization with merge tags, like {first_name} If your ESP supports it.

7. Check desktop + mobile views. Send a test email and verify that the links and images work.

8. Then export the HTML and upload it to your email platform.

AI Tools to check out

🎬 ElevenLabs Image & Video: Generate visuals, voiceovers, music, AI sound effects, and captions.

🛎 Delight.ai: Branded AI concierge that makes every experience personal and delightful for customers.

🧭 Guideflow: AI demo automation platform to create interactive demos & guides for SaaS.

🤝 Dimension: AI coworker that learns about you and your business to get work done.

🎻 Spine Canvas: Visual AI workspace for orchestrating models.

AI Findings/Resources

🧪 AI-powered "virtual human" to predict drug effects systemically and transform pharmaceutical discovery

🦟 AI and public photos detected a malaria mosquito in Madagascar.

📉 Firms report that AI investment may cause initial productivity losses.

🎓 A university launched a private, academic version of ChatGPT.

The latest in AI and Tech

The AI Loop: Is Generative Content Eating Itself?

There is a rising concern about the "loopification" trend, in which generated AI content is fed back into the training data of new models. This self-referential cycle creates risk regarding the future reliability and quality of AI outputs.

Grok's Partisan Personality Sparks Bias Debate

Grok generated a highly controversial, partisan response that suggested Elon Musk's superiority, highlighting the ongoing public debate over AI bias and the reliability of models trained on real-time public data.

Brandwatch Boosts Consumer Insight with Iris AI Upgrade

Brandwatch launched a major update to its Iris AI layer, introducing features like a quick chat assistant and expanding its data coverage to include newer social platforms like Threads and TikTok for marketers.

ScaleOps Cuts GPU Costs for Self-Hosted AI

ScaleOps launched a new AI infrastructure product that uses smart orchestration and resource management to drastically reduce the high GPU cloud costs for enterprises running their own AI models.

Grok 4.1 Agent API Technical Leap Overshadowed

XAI's release of the Grok-4.1 API, which includes advanced agent tools and enhanced developer access, was largely overshadowed by the model's public controversies, underscoring the challenge of balancing technical innovation and public perception.

That’s a wrap!

Thanks for sticking with us to the end! Let’s stay connected on LinkedIn and Twitter.

Not subscribed yet? Sign up here and send it to a colleague or friend!

See you in our next edition!

Gina 👩🏻‍💻