Latest Update
Seed Audio 1.0: voice, music, and sound effects in one pass
2026.06.26Hey, it's the Carat team. ☺️ Today kicks off with a new audio AI that makes voice, music, and sound effects in one go. Then: a short anime you can make without knowing a single tool, a new Google feature that pulls real streets in as backdrops, and the AI agents that reshaped every team at OpenAI.
🔥 Seed Audio 1.0: voice, music, and sound effects in one pass
fal just dropped ByteDance's new audio AI, Seed Audio 1.0, as an exclusive. It generates voice, background music, and sound effects together in a single pass.
Feed it up to three reference clips to lock in a voice's tone, emotion, and character, and it can build scenes with several speakers at once. One generation runs up to two minutes long.
You can define the voice you want from just a text description, a photo of a person, or a short recording.
📌 Three things worth knowing today
1️⃣ A short anime, with zero tool knowledge?
A creator chained several AI tools together to make a short action animation. A swordfighter squares off against a giant beast, and the linework came out pretty refined.
This creator hand-picked tools like Midjourney, Nano Banana, and Seedance at each step. Carat 2.0 has a skill built for animation that figures out which model to use, and in what order, on its own. You don't need to know the tools; just say what you want and the workflow runs itself.
You can make it right in Carat. 👉 Create on Carat 2.0
2️⃣ Google Maps streets are becoming AI backdrops
Google's video tool Flow can now pull real street scenes in as backgrounds. It builds images and videos based on actual places from Google Maps Street View.
Turn on Agent mode, name a place or address in your prompt, and you can set a character in that neighborhood or restyle a city landmark. For now it only works for US locations.
3️⃣ Every team at OpenAI runs on AI, Research up 53x
OpenAI shared that every department inside the company now works through its Codex agents. Engineering, sure, but also non-dev teams like Legal and Recruiting lean on Codex as their main tool.
Compared to last November, the Research team's usage jumped 53x, Customer Support 32x, and Engineering 26x. More than 85% of what an average employee produces now runs through Codex.
🧪 Prompt of the day
A couple-mood 4-cut photo strip with your fave and you
youcantouchmylips posted this reel on Instagram, and it passed 130K views. Drop in two photos, one of your fave and one of you, and you get a Haru Film style 4-cut photo strip with four different poses.
Here's what we made on Carat with the same prompt, using a photo of Jang Won-young.
Make a 4-cut photo strip from two photos with ChatGPT Image 2 Medium. Prompt:
A man and a woman posed in a 4-cut photo booth strip
- Cut 1: hands crossed into an X pose
- Cut 2: one hand covering the mouth
- Cut 3: one hand raised to the forehead looking off somewhere, but still facing the camera
- Cut 4: both hands covering the ears. Pose them like an affectionate couple doing the same pose together
- Make it look like a Haru Film 4-cut photo
- The image doesn't need to be super sharp; a slightly rough, grainy texture is fine
- Keep the texture natural, like a shot from a Photogray photo booth
- Keep the man's and woman's facial features exactly as they are; gray background
- A faint Haru Film style blue tint in the background is okay
- Lay it out as two cuts on top and two cuts on the bottom
- Match the outfits to the man's
- Image ratio 9:16
- Never change the faces
- Vary the woman's expression naturally across the cuts, don't make them too similar
Make yours on Carat right now. 👉 Create on Carat
Today we ranged from audio to video to how we work. If one of these caught your eye, go make something with it on Carat. See you next week with more. 😊
Was this helpful? Share with your friends
Share
Get daily AI trend updates
