Latest Update

AI-Made Content Is Starting to Get Hidden Labels

2026.05.27
Hi, it's the Carat team. 🤗
Today, we're covering hidden labels for AI-made content, safety guardrails for Claude agents, Google’s personal morning briefing, and Microsoft’s new image model.

AI-Made Content Is Starting to Get Hidden Labels

© Google DeepMind
© Google DeepMind
Google DeepMind is expanding a technology that leaves a hidden label on content made by AI. People may not see it with their eyes, but later, it can help identify whether something was made by AI.
The technology is called SynthID. Google says it has already added these labels to more than 100 billion pieces of AI content, and OpenAI, ElevenLabs, and Kakao are joining the same movement. As AI content grows, the ability to check whether something was AI-made is becoming just as important as the ability to make it well.
As AI content blends into everyday media, people will care more about where a result came from. In the future, standards for recognizing AI-made content may become just as important as generation quality.

📌 Today's Top 3

1️⃣ Claude Agents Working Inside a Safety Fence

© Anthropic
© Anthropic
Anthropic shared a sandbox approach for using AI agents like Claude more safely. When an agent handles files or runs commands, its actions can be kept inside a limited workspace.
Once AI moves beyond writing answers and starts reading files, editing code, and running tasks for us, permission control becomes more important. Smarter agents also need clearer safety guardrails.

2️⃣ Google Wants to Make Your Morning AI Briefing

© Google
Google introduced Daily Brief inside the Gemini app. Based on connected apps like Gmail and Calendar, it can summarize your schedule, to-dos, and follow-up actions every morning.
AI assistants are moving beyond waiting for a command. They are starting to prepare the information you need before your day begins. The smartphone AI race may ultimately come down to how well an assistant understands your day.

3️⃣ Microsoft Joins the Image Model Race?

© Microsoft AI
© Microsoft AI
Microsoft AI introduced a new image generation model called MAI-Image-2.5. The company says it ranked third on a public image model evaluation site and emphasized that image model competition is spreading across OpenAI, Google, and Microsoft.

🧪 Today's Prompt Tip

Turn Your Fandom Photo into a Cute Character

This prompt from hi.se_hee on Threads reached about 100,000 views. Upload a baseball game photo, and it turns the outfit, cheering stick, hair accessories, and fan mood into a chibi-style merchandise character.
© @hi.se_hee, @hye2ove
© @hi.se_hee, @hye2ove
This one was made on Carat.
Generated on Carat with ChatGPT Images 2.0
Generated on Carat with ChatGPT Images 2.0
Paste the prompt below into Carat and upload your own photo.
Use the ChatGPT Images 2.0 Medium model with the prompt below to generate an image.

Using the uploaded photo as reference, turn the person in the photo into a cute, simple character illustration for merchandise.
Important: do not reference only the face. Reference all visible elements in the photo.
That means reinterpreting the person’s facial impression, clothing, baseball uniform, hair accessories, ribbons, cheering tools, handheld props, accessories, and fan atmosphere into one character.
The final result must always follow this fixed character style:
* Very cute, rounded 2D character
* Simple illustration suitable for merchandise
* Shape suitable for stickers, keychains, and cheering goods
* Chibi proportions, about 2.5 to 3 heads tall
* Large head, small body
* Round face, short arms and legs
* Thick, clean outlines
* Soft, tidy colors
* Clean cel-shaded look with minimal shadows
* Overall soft and lovable mood

Face expression rules:
* Preserve the impression and traits of the uploaded person
* Do not copy realistically; simplify into a character
* Eyes should feel like small black dot eyes
* Nose should be very small and simple
* Mouth should be small and cute
* Add soft pink blush on the cheeks
* Expression should feel slightly confident and cute-neutral rather than a big smile
* Avoid mature or flashy looks
Hair rules:
* Simplify the hairstyle, bangs, tied hair, baby hairs, and decorations from the photo
* Emphasize headbands, ribbons, pins, scarves, or decorations as key points
Outfit rules:
* Base the outfit on the clothing in the photo
* If it is a baseball uniform, cutely simplify the colors, logo impression, sleeve details, and layered look
Cheering tools / props rules:
* Be sure to turn cheering tools and fan props from the photo into character elements
* Keep key elements such as cheering sticks, signs, slogans, towels, dolls, and hair accessories
Pose rules:
* Stable full-body pose that reads clearly as merchandise
* Front-facing or slightly angled
Background rules:
* White or very pale solid-color background
* Only small hearts, stars, or sparkles may be added

Most important rule:
No matter who uploads what photo, the result must always keep the same character language.
Avoid: realistic style, semi-realistic style, mature body proportions, flashy anime eyes, complex backgrounds, excessive detail, randomly added props, or outfits unrelated to the photo.
Try making it yourself. 👉 Create on Carat
That's all for today. See you tomorrow with more AI news. ☺️

Was this helpful? Share with your friends

Share

Get daily AI trend updates