Latest Update
Google DiffusionGemma writes text 4x faster, in blocks
2026.06.11What's up, Carat crew here. 🙂
Today: Google DiffusionGemma writing text 4x faster, Pika swapping the language you speak on video, an HONOR phone bringing photos back to life as video, and an AI site builder one dev built in two weeks after getting laid off.
🔥 AI answers, 4x faster? Google DiffusionGemma
Google DeepMind unveiled a new AI model called 'DiffusionGemma.' Where most AI writes text one word at a time in order, DiffusionGemma bundles words into blocks and generates them all at once, so it delivers the same answer up to 4x faster.
It borrows the trick image AIs use, refining a blurry picture into a sharp one, and applies it to writing. The upside is answers come with almost no wait. In return, the text quality is a bit lower than the standard model, traded off for speed. It reportedly supports over 140 languages.
📌 3 more things today
1️⃣ One tool and you're fluent in any language? Pika
Pika unveiled 'Language Swap,' which changes the language someone speaks inside a video. Take a video of yourself speaking Korean and switch it to English or Japanese, and even your mouth movements and voice match the new language. It looks like you spoke it all along.
Shoot a video once and you can spin out versions in several languages. Even if you don't speak the language, you get a video of yourself speaking it fluently.
2️⃣ AI video generation, built right into your phone?
Chinese smartphone brand HONOR built a feature into its new phone that brings photos back to life as video. Pick a photo and that moment starts moving like a short clip.
Turning photos into video isn't new, but the point here is you can use it as a built-in phone feature, no separate app to install. You can replay a goal you scored or a travel photo as video.
3️⃣ He got laid off and built 'this' in two weeks?
A developer got laid off from the company he worked at, and in two weeks he built an AI site builder called 'Open Canvas.' Unlike the usual fill-in-a-template approach, you design right on the screen and an AI agent at your cursor jumps in to help.
It's a multiplayer setup where several people edit on one screen together. Watch the video and a few clicks spin up a page in seconds.
🧪 Prompt of the day
Today's prompt makes a profile pic that looks like you're at an amusement park in a school uniform. With just one photo of yourself, you get a school-uniform profile shot without going anywhere.
Transform my photo with the following prompt using ChatGPT Image 2.
Prompt: Use the uploaded model image as the strongest identity reference. Preserve the person's facial identity, face shape, eyes, nose, lips, skin tone, hairstyle impression, and natural expression.
Transform the person into a trendy Korean amusement park school-uniform photoshoot. The person is wearing a stylish rental school uniform: fitted brown or navy blazer, white dress shirt, striped necktie or ribbon tie, neat preppy styling, and coordinated school-look bottoms. Create a bright daytime profile-photo style image in front of a dreamy castle-inspired amusement park backdrop with pastel balloons and soft crowd blur. Add a small cute amusement park accessory only if it fits naturally.
Camera style: high-quality smartphone photo taken by a friend, Korean SNS profile photo aesthetic, bright clean skin tone, soft focus, shallow depth of field, polished but realistic. Natural gentle smile, slightly angled body, youthful lovely high-teen mood. No text, no captions, no watermark, no readable logos, no duplicated face, no extra merged people.
Find more prompts in the prompt gallery. 👉 See the amusement-park prompt
That's it for today. We'll be back tomorrow with fresh AI news. ☺️
Was this helpful? Share with your friends
Share
Get daily AI trend updates
