Latest Update
Nano Banana 2 Lite: an image in 4 seconds
2026.07.02Hey, it's the Carat team. 🤗 Today we've got Google's new model that fires off an image in about 4 seconds, a trick to stop your AI video characters from changing faces, and a no-code way to build a talking voice AI. We saved a fun one for last: a prompt for video-calling your fave.
Nano Banana 2 Lite: an image in 4 seconds
Google just dropped a new image model, Nano Banana 2 Lite. It's the lightweight version of Nano Banana 2, and it spits out a single image in about 4 seconds.
It's 5x faster than the full Nano Banana 2 and about half the price. The trade-off: the full version still nails complex compositions better. It's built for when you have a pile of ideas and want quick mockups to compare before you commit to a video.
Both the fresh Nano Banana 2 Lite and Gemini Omni Flash, the video-editing one, are already on Carat, so you can try them right now. 👉 Create on Carat
📌 Today's 3 stories
1️⃣ Does a storyboard ruin your character?
A creator put a popular rumor to the test: that starting an AI video with a storyboard wrecks character consistency. The verdict? Kind of yes, kind of no.
They sketched a storyboard in ChatGPT Image 2, then animated it with Seedance 2.0. The storyboard wasn't the problem. What mattered was whether they kept the same character image as the anchor. Generate a fresh one each cut and the face drifts. Lock one image and reuse it, and the face stays put.
So the move is to nail the character image first, then go to video. Don't know the workflow? On Carat you just drop in a photo and get this kind of cinematic video anyway. 👉 Create on Carat
2️⃣ Build a talking AI with zero code
xAI just launched Voice Agent Builder, a no-code way to make a voice AI. With Grok Voice, you can spin up a voice agent that talks like a real person in a few clicks.
It starts at $0.05 a minute. Hook up a phone number and it's another $0.01 a minute. Pick from 80-plus built-in voices, or record about 2 minutes of audio to get an agent that talks in your own voice.
It cleared 5,500 likes in a day. People like that you only pay for the minutes you actually use, which feels lighter than other voice AIs.
3️⃣ Fable 5 tops a real-work benchmark
Anthropic's top model Fable 5 came back online worldwide yesterday. It got briefly blocked outside the US over a safety issue, then returned after a fix, and it made noise right away with its benchmark score.
On a test that measures how much real freelance work AI can actually finish, it scored 16.1% for first place. That's nearly double the runner-up, Opus 4.8 at 8.33%. The benchmark uses 240 real freelance projects across 23 fields. It's the first score in double digits, though at one or two jobs out of ten, there's still a long way to go.
🧪 Prompt of the day
Video-call your fave
Here's a prompt won_3232 shared on Threads. It builds a scene that looks like you're on a video call with your favorite star, as an image or a video.
Here's what we made on Carat.
The image prompt. Since there's a person in it, spelling out the exact look you want helps.
Realistic Korean young woman taking a casual selfie during a video call in a cozy dim bedroom at night, wearing a simple white t-shirt, long natural black hair, soft warm lighting, natural bare makeup, holding phone naturally, slight smile, realistic skin texture, relaxed and candid atmosphere, FaceTime feeling, emotional fan communication mood, soft shadows, cinematic realism, natural eye contact, handheld smartphone camera aesthetic, ultra realistic, Korean influencer vibe, subtle flash reflection, calm and warm mood.
And here's the video prompt.
Realistic video call with a Korean young woman in a white t-shirt, cozy night room lighting, natural smartphone camera feeling.
The woman gently waves first, smiles softly, fixes her hair naturally, and talks to fans calmly.
Afterward, the fan reacts separately with excited eyes, covering her mouth, smiling brightly, softly cheering and saying thank you emotionally.
Important: both motions must NOT happen at the same time or mirror each other.
Make it on Carat right now. 👉 Create on Carat
That's it for today. From an image model that fires in 4 seconds to a prompt for video-calling your fave, there's plenty to play with. Catch you tomorrow. 😊
Was this helpful? Share with your friends
Share
Get daily AI trend updates
