Hey, it's the Carat team. 🤗
Today: an AI that visualizes design concepts in real time, Google's new AI voice model, and an AI-generated person doing a food review ad.

Describe your design idea, and AI turns it into a polished visual in just one minute. Feed in a sketch or a rough concept memo, and out comes a preview that looks like a finished product.

Interior design, product design, architecture -- it doesn't matter the field. If you have an idea, you can now create professional-quality visuals. The post racked up over 1,000 bookmarks.

Creating a single design mockup used to require learning specialized software and spending hours on it. Now you describe your idea in words or draw a quick sketch, and AI shows you the finished version. You don't have to be a designer to see your ideas come to life.

Google released Gemini 3.1 Flash TTS. It converts text into natural-sounding speech, but this time you can control the voice style through text as well.

For example, adding an Audio Tag like "in a calm, warm voice" makes the output match that description. It supports over 70 languages including Korean. Useful for narration, audiobooks, voice assistants, and more.

OpenAI launched GPT-5.4-Cyber, an AI model dedicated to cybersecurity. It's a specialized version of GPT-5.4 focused on analyzing malicious code and finding vulnerabilities.

It's not open to everyone. Only verified security professionals get access, and dangerous requests are automatically blocked. Just like yesterday's AI store hacking incident, the more AI can do, the more security has to keep pace.

The US Treasury Secretary publicly mentioned Anthropic's Claude Mythos at a Wall Street Journal event, calling it a "step change in capability."

The backstory is fascinating. Last week, he convened an emergency meeting with the Fed Chair and Wall Street CEOs. The week before, the Vice President briefed major AI company CEOs on cyber risks. Meanwhile, the Department of Defense is pushing to blacklist Anthropic. The same AI is simultaneously a "national security threat" and a "national security asset" -- an ironic situation.

With Seedance 2.0, you can create an ad video of a non-existent person doing a food review. All you need is one person image, one food image, and a prompt.

The key is clearly assigning roles in the prompt: the person (@Image1) and the product (@Image2). Then add natural actions like "explaining the taste" and you get a video that looks like a real person reviewing.

Generate a short video for product promotion. A woman (@Image1), sitting at a restaurant table, eating [Product] (@Image2), explaining the taste of the product. Use colloquial language and a casual, conversational tone to share her experience. Simultaneously, she uses chopsticks to pick up and showcase special ingredients. Her expression is friendly and engaging.

Here's what we made on Carat with Seedance 2.0. ⬇️

Swap [Product] with whatever you want and upload your person and product images. Try it in Carat's Seedance 2.0 Prompt Generator.

That's all for today. We'll be back tomorrow with more useful updates. ☺️