Seedance Tutorial: Make AI Videos Step by Step (Text-to-Video and Image-to-Video)

If you searched for a Seedance tutorial, you probably already saw a clip that looked too smooth to be AI. Seedance is known for two things: believable motion and physics, and native multi-shot output, which means one prompt can produce a clip that cuts between angles like a real edit. That second part is what makes it useful for ads and social content, not just demos.

This guide is written for a first-time user. You will learn where Seedance actually lives (it is not a standalone website you sign up for), how to run a text-to-video generation, how to animate a still image, how to write prompts that control the shots and the camera, and how to turn the raw clip into a finished ad with voice and captions. Every step uses menu names and prompts you can copy.

Seedance is also one of the video tools taught hands-on inside the AI Mastery course, where the video module goes from prompt to published ad. This article covers the core workflow so you can start today.

What Seedance Is and Where to Access It

Seedance is a family of video generation models built by ByteDance, the company behind TikTok and CapCut. Seedance 1.0 arrived in 2025 with native multi-shot video, strong motion and physics, and both text-to-video and image-to-video. The latest Seedance generation, released in 2026, adds better audio and longer, more controllable shots. Do not worry about version numbers; the interfaces below always expose the current model.

There is no "seedance.com" account. You reach the model through products that embed it:

• Dreamina: ByteDance's creative web app (from the CapCut family). Go to the site, sign in with a CapCut or email account, and open AI Video. This is the cleanest place to learn, and it is what the steps in this tutorial use.
• CapCut: the desktop and mobile editor exposes Seedance inside its AI video features, so you can generate a clip and drop it straight onto the timeline. Handy once you already edit in CapCut.
• Jimeng (即梦): the Chinese-market version of Dreamina. Same model, Chinese interface, Chinese phone sign-up. Most readers outside China should use Dreamina instead.
• API partners: Seedance is offered through ByteDance's cloud platform and several third-party model aggregators. If you want to generate clips from a script or an automation like n8n, this is the route. Check each provider's docs for the exact model name and parameters.

For this tutorial, open Dreamina in a desktop browser. Mobile works, but prompts are easier to edit on a keyboard, and you will be editing prompts a lot.

Text-to-Video Step by Step

Text-to-video means you describe the scene and Seedance invents everything: subject, setting, lighting and motion. Follow these steps in Dreamina.

1. Open AI Video. From the Dreamina home screen click AI Video in the left sidebar. Make sure the mode toggle at the top of the panel is on Text to Video.
2. Pick the model. Open the model dropdown and select the Seedance option. If you see more than one Seedance entry, the higher one is the latest generation; the "Lite" or "Fast" entry is cheaper and quicker, good for drafts.
3. Set aspect ratio and duration. Choose 9:16 for TikTok, Reels and Shorts, 16:9 for YouTube and landing pages, 1:1 for feed posts. Start with the shortest duration offered while you test prompts.
4. Write the prompt. Paste something like: A barista pours latte art in a sunlit cafe, close-up on the cup, steam rising, slow push-in, warm morning light, shallow depth of field.
5. Generate. Click Generate. A job card appears with a progress bar. Short clips usually finish in a couple of minutes, but wait times depend on server load.
6. Review. Play the clip full screen. Watch hands, text, and anything that moves fast. Those are the areas where video models slip.
7. Download or extend. Hover the clip and click the download icon for an MP4. Use Regenerate to try the same prompt again, or edit the prompt and generate a new variant.

If the first result looks flat, do not rewrite everything. Check three things in order: is there a physical action in the prompt, is there exactly one camera movement, and is the lighting named. Nine times out of ten one of those slots is empty, and filling it changes the clip more than any style word will.

One habit to build now: keep a plain text file of every prompt and what it produced. Video generation is not deterministic, so your notes become your real learning curve.

Image-to-Video Step by Step

Image-to-video is the mode most ad creators actually use. You supply a still frame (a product photo, a Midjourney or Nano Banana render, a screenshot of your app) and Seedance animates it. The look of your brand stays locked because the first frame is yours.

1. Prepare the image. Export a clean image in the same aspect ratio you want for the video. A 9:16 source for a vertical clip avoids cropping. Avoid tiny text on the image; it will smear once it moves.
2. Switch modes. In AI Video, flip the toggle to Image to Video. An upload box replaces the empty canvas.
3. Upload the frame. Drag your image into the box. Some builds also offer a Last frame slot; use it when you want the clip to end on a specific image, such as a logo card.
4. Describe the motion, not the scene. The image already tells the model what things look like. Your prompt should say what moves and how the camera behaves: The bottle slowly rotates on the marble counter, water droplets slide down the glass, camera orbits left, soft studio lighting stays constant.
5. Generate and compare. Run it twice with the same prompt. Pick the take where the product stays undistorted for the full clip.
6. Download the MP4. Save it with a name that includes the prompt version, for example bottle-orbit-v2.mp4.

Which source images animate best? Photos with a single clear subject, a plain or softly blurred background and even lighting give the model less to guess. Busy scenes with many overlapping objects tend to produce ghosting when the camera moves. If your product photo has a cluttered background, cut it out and place it on a simple gradient before uploading. Also keep the subject away from the very edge of the frame, because any camera movement will push edge content out of view and the model has to invent what replaces it.

A useful pairing: generate your first frame with an image model that follows text well, then animate it here. The Seedance vs Veo 3 vs Sora 2 comparison covers which model wins for which kind of footage if you are still choosing.

Prompt Formula and 5 Copy-Paste Examples

Most good Seedance prompts follow the same order. Use this formula until it becomes automatic:

[Subject + one defining detail] + [Action] + [Setting] + [Lighting] + [Camera framing + movement] + [Style] + [Pace]

Keep it to two or three sentences per shot. Fill every slot, because a missing slot is a slot the model fills with its own guess. Here are five prompts you can paste as they are and then adapt.

1. Product ad (image-to-video)

The sneaker on the white podium slowly rotates 180 degrees, laces sway slightly, dust particles float in a beam of light. Studio softbox lighting, commercial product shot, camera orbits right at the same height, real time.

2. Lifestyle B-roll (text-to-video)

A young man in a grey hoodie opens a laptop at a kitchen table, morning sunlight through the window, coffee steam rising beside him.

Medium shot, slow push-in, shallow depth of field, documentary realism, real time.

3. Multi-shot story (text-to-video)

Shot 1 (wide): A golden retriever waits by a front door in a bright hallway, tail wagging. Shot 2 (close-up): The door handle turns. Shot 3 (medium): The dog leaps toward a woman in a blue coat as she walks in, both laughing.
Warm afternoon light, handheld, cinematic 35mm film, real time.

4. Food reel (text-to-video, vertical)

Chopsticks lift a glossy piece of teriyaki salmon from a black bowl, sauce drips slowly, sesame seeds scatter. Extreme close-up, tilt up to follow the salmon, overhead softbox lighting, commercial food shot, slow motion.

5. App demo (image-to-video)

The phone on the desk stays still while the screen content scrolls smoothly upward, a hand enters from the right and taps once, subtle screen glow.
Locked-off tripod, top-down angle, clean studio lighting, real time.

Notice that none of the prompts include on-screen text. Add text later in the editor. Video models still mangle letters, and a misspelled word is the fastest way to reveal a clip as AI.

Turning Clips Into Ads With Voice and Captions

A raw Seedance clip is B-roll. An ad needs a hook, a voice, captions and a call to action. Here is the workflow that goes from prompt to post.

1. Script first. Write the voiceover before you generate anything. A 15-second ad is roughly three sentences: hook, benefit, call to action. Every sentence becomes one shot.
2. Storyboard with stills. Generate one image per sentence with an image model. Approve the look here, where changes are cheap.
3. Animate with image-to-video. Run each still through Seedance with a motion prompt. You now have three clips that match your brand.
4. Record the voice. Paste the script into ElevenLabs, pick a voice, export MP3. Or record yourself; a real voice still tests well for founder-led ads.
5. Assemble in CapCut. Import clips and audio, trim each clip to its sentence, add a cut on every beat. Use Text then Auto captions to generate word-level subtitles, then choose a bold caption style with high contrast.
6. Add the CTA card. End on a still with your offer and a large button-style text. Keep it on screen for at least two seconds.
7. Export and disclose. Export at the platform's native resolution. Where a platform requires it, tick the AI-generated content label. It is required in several places now and it protects your account.

This exact script-to-storyboard-to-video-to-voice-to-caption chain is the backbone of the video module in the full AI course, where you build a complete ad end to end. If you plan to automate parts of it, the how to build an AI agent for beginners shows how to chain these tools with a simple loop, and the prompt engineering guide teaches the same one-variable-at-a-time method for text and image models.

Limits, Costs and Common Errors

Limits you should plan around

Each generation produces a short clip, and the maximum duration depends on the model entry and the app you use. Longer ads are made by generating several clips and stitching them, which is why the storyboard step above matters. Resolution and frame rate options also vary by app; check the settings panel before you export, and upscale in CapCut if you need a larger master.

Costs

Dreamina and CapCut sell credits; the full Seedance model costs more credits per clip than the Lite or Fast entry, and longer or higher-resolution clips cost more. API partners bill per second of generated video. Prices change often, so check the current pricing page of whichever route you use before you plan a campaign. A safe rule: assume every finished shot will take two to three generations, and budget credits accordingly.

One more cost that is easy to forget: your own time. A generation that takes three minutes and then needs three retries is ten minutes per shot, so a three-shot ad is a solid half hour of waiting before editing begins. Batch your generations. Queue all three shots at once, then step away, rather than generating, watching, and generating again. If you are on the API route, the same logic applies: submit jobs in parallel and poll for results instead of running them one at a time.

Common errors and fixes

• "Generation failed" or a job that never finishes. Usually server load or a prompt the safety filter rejected. Wait, then retry once. If it fails again, remove brand names, celebrity names and anything violent or explicit from the prompt.
• Image upload rejected. The file is too large, an unsupported format, or contains a real person's face that the policy blocks. Export a JPG or PNG under the size limit shown in the upload box, and avoid uploading photos of identifiable people you do not have rights to.
• Subject changes appearance between shots. Your shot lines describe the subject differently. Copy the exact same subject phrase into every shot.
• Camera ignores your instruction. You stacked several movements. Keep one movement per shot and place it at the end of the line.
• Text on screen is garbled. Expected. Remove text from the prompt and add it in the editor.
• Product warps during rotation. Reduce the movement (rotates 90 degrees instead of a full spin), or use a cleaner, higher-contrast first frame.
• Clip is beautiful but nothing happens. The prompt had no action verb. Add one physical action per shot.
• Output is the wrong shape. The aspect ratio setting defaulted back after a model switch. Re-check it every time you change models.
• Export looks blurry on the platform. You uploaded a low-resolution master. Export at the platform's recommended resolution and upscale first if needed.

Official documentation for the Dreamina app and the CapCut AI video features lives on their help centers: Dreamina and CapCut. For the API route, read the parameter list on your provider's page before writing a single line of code.