How to Use Seedance 2.0 the Right Way
2026.04.13 · 149 readersHey, it's the Carat team. 🤗
Today we're breaking down the official Seedance 2.0 prompting guide, a character-swap feature Grok is developing, and a 3D balloon typography prompt.
Today we're breaking down the official Seedance 2.0 prompting guide, a character-swap feature Grok is developing, and a 3D balloon typography prompt.
🔥 How to Use Seedance 2.0 the Right Way
fal released an official Seedance 2.0 prompting guide. Here are the key takeaways.
❌ First, don't use image generation prompts as-is. Keywords like "cinematic shot, 4K, beautiful lighting" are meaningless for a video model. There's no motion information, so you get flat results. Think of a Seedance 2.0 prompt as a shot list you'd hand to a director of photography.
✅ Good prompts follow this order:
① Subject + action: Start with "what moves and how."
② Camera movement: Use cinematography terms like tracking shot, dolly zoom. The model understands them directly.
③ Sound: Write sounds specifically, like "crack of thunder" or "rain on a tin roof." The more specific you are, the better the audio gets.
④ Multi-shot labels: If you want multiple scenes, label them "Shot 1:", "Shot 2:". Without labels, the model renders one continuous long take.
② Camera movement: Use cinematography terms like tracking shot, dolly zoom. The model understands them directly.
③ Sound: Write sounds specifically, like "crack of thunder" or "rain on a tin roof." The more specific you are, the better the audio gets.
④ Multi-shot labels: If you want multiple scenes, label them "Shot 1:", "Shot 2:". Without labels, the model renders one continuous long take.
💡 Extra tip: Stick to one action per shot. If you cram running + camera pan + lightning into one shot, the model gets confused about what to prioritize. Ideal length is 2–4 sentences for a single shot, 4–8 for multi-shot.
Seedance 2.0 is available on Carat right now, along with a Seedance 2.0 Prompt Generator that has all these best practices built in. Just describe what you want in plain language and it creates the optimal prompt for your video.
📌 Three Things Today
1️⃣ Grok Is Building a Character Swap for Video
xAI is working on a new feature for Grok. It swaps out a person in a video with a character image you provide. For example, you could replace someone's face in a video with an AI-generated character.
It's called Remix Character and it's still in development. You feed it a character image and it replaces the person in the video while keeping their movements intact.
You can already do this on Carat. Just tell Carat: "Replace the person in this video with the person in this photo" and it handles the rest.
2️⃣ 3D Balloon Letters with Nano Banana
AI creator AmirMushich shared a prompt for making 3D balloon typography with Nano Banana. The puffy, inflated 3D letters filling the frame got nearly 1,000 bookmarks from people wanting to try it themselves.
You can make these on Carat right now. Feed Nano Banana the typography style you want along with your text, and it generates the 3D lettering. Great for brand logos or event title designs.
3️⃣ Meta Muse Spark Is Getting an API
Meta's new AI model Muse Spark, which we covered recently, got two updates. First, an API is on the way. This means developers will be able to plug Muse Spark directly into their own apps and services.
Second, a Contemplating Mode is now available. When you ask a question, multiple agents work on it in parallel and merge their answers into one consolidated response.
🧪 Prompt Tip of the Day
A Museum Painting That Becomes Reality
AI video creator umesh_ai shared a video made with Seedance 2.0. A museum painting of the sea transforms into real ocean, then a hurricane, then a battlefield, then a sun-drenched village. It's one continuous shot with no cuts, and a bewildered museum guard appears throughout.
The key technique is "shared-shape transitions." A ship's sail becomes a desert tent, the tent becomes a war banner, the banner becomes a laundry line. Objects with similar shapes flow naturally into each other. The prompt specifies these transition rules explicitly.
A cinematic one-shot begins inside a silent grand museum at night. The camera moves toward a massive painting of a storm at sea. As it nears the canvas, the painted waves begin moving, and the frame transforms continuously without cuts: the painted ocean becomes real water flooding the gallery floor, the gallery becomes a ship deck in a hurricane, the sailcloth becomes desert tents, the tents become war banners on a battlefield, the banners become laundry lines in a sunlit village, the village walls become canyon cliffs, the cliffs become bookshelves in an endless library. Every transition must come from shared shapes and motion. A museum guard walks through all of it in disbelief, but the camera never leaves him. Frames become windows, windows become doorways, doorways become portals into new realities. Final transformation: the guard realizes he is now inside a painting, frozen in brushstroke texture, while museum visitors admire him from outside.
Here's what we made on Carat with Seedance 2.0, minus the guard. ⬇️
Look closely at the prompt and you'll see it follows every rule from today's hero section. If writing prompts like this feels daunting, try Carat's Seedance 2.0 Prompt Generator. Just describe what you want in everyday language and it converts it into the perfect prompt.
That's all for today. We'll be back tomorrow with more useful updates. ☺️