A prompt formula for 30-second videos with Seedance 2.5

A 30-second video is a small film, not a moving picture. A prompt that describes a still scene tends to give you 30 seconds of the same action on repeat, or a video that comes back short. The formula below comes from real runs on Dola Seedance 2.5. It can't guarantee Dola does exactly what you want, because Dola is a third-party service that decides its own output, but it helps cut down on re-runs.

30-second videos exist only on Dreamina Seedance 2.5 and need a paid plan.

Four hard rules

Break one of these and you'll almost certainly be running it again.

1. One prompt is one line. No line breaks inside a prompt. When you paste in bulk, the app splits prompts at every blank line by default. A prompt with a blank line inside it, or with any line break when you choose One prompt per line, is cut into two shots in the wrong place.
2. 1500–2000 characters. Long enough to tell 30 seconds, not so long that Dola drops parts.
3. Don't write "30s", a resolution or the model name at the start. Spider Hub adds a short instruction line (length and resolution) itself before sending. Your prompt is only the creative part.
4. Don't name famous buildings, landmarks, characters or brands. Dola may accept the prompt, finish the video and only then block it at review, so you wait for nothing. Describe instead: "a suspension bridge over cloud-covered mountains" rather than naming the bridge.

Writing in Chinese often gets closer to what you mean

In our experience, prompts written directly in Chinese often get action, pacing and transitions closer to the intent than a line-by-line translation from another language. One obvious reason: each Chinese character carries more meaning, so 1500–2000 characters of Chinese tell far more than 1500 characters of English.

If you can't write Chinese, draft the idea in your own language following the structure below, then have a translation tool translate the whole passage in one go rather than sentence by sentence.

The structure: 3–4 scenes, three things each

Treat 30 seconds as 3 or 4 small scenes in a row. Each scene needs:

1. A concrete action by the characters, not just scenery.
2. How it transitions to the next scene.
3. The pace of that part: slow, fast or normal.

Common pace and transition phrases:

Always close the prompt with one phrase about lighting, colour and shooting style for the whole piece. It keeps the colour consistent across scenes.

With reference images: name them in the first sentence

When you attach images, Dola numbers them in upload order: image 1, image 2… The first sentence of the prompt must say who each image is, with a name:

图1中的角色是一位名叫小陈的年轻女背包客，图2中的角色是一只名叫阿福的橘色小狗。

After that, just use the names ("小陈", "阿福"); no need to repeat the image numbers. More in Reference images: when and how.

A template to fill in

图1中的角色是[character 1: age, clothes, recognisable details]，图2中的角色是[character 2, if any]。[Scene 1: concrete action, generic setting with no real landmark names]，[pace of part 1, e.g. 手持跟拍，正常速度]，甩镜头切至[Scene 2: a different action]，[pace of part 2]，硬切至[Scene 3: climax or ending]，[pace of part 3, often 慢动作 for the final moment]，[lighting, colour and shooting style for the whole piece]。

The template is one line, as rule 1 requires.

Example: two characters on the road

The idea, so you can picture it:

• Scene 1 (handheld, normal speed): Xiao Chen (image 1) and her dog A Fu (image 2) stroll down an old town street. A sudden downpour; she scoops up the dog and runs under a porch, and they look at each other and burst out laughing.
• Scene 2 (whip pan in, quick cuts): the rain stops, a ball rolls down a slope covered in yellow leaves, A Fu bolts after it and Xiao Chen chases, calling out. Many short shots: feet, splashing water, flapping ears.
• Scene 3 (hard cut, slow motion): they tumble into a pile of dry leaves that float up slowly as they laugh. The camera pulls back and rises, ending on their silhouettes in the late sun.
• Close: warm sunset light, soft cinematic lighting, light film grain.

Written out in Chinese, each scene is told in much more detail: clothes, background sounds, small gestures, until the prompt reaches about 1500–2000 characters. To make it longer, add small details to the existing scenes rather than adding new ones.

When Dola asks a question

For a 30-second shot Dola sometimes asks whether to make it shorter, or offers options to split it. Spider Hub never answers that for a 30-second shot: experience shows that saying yes usually gives a short video. The app gives the run back, leaves the chat and opens a new chat on the same account. If Dola says it will make something shorter but then starts by itself, the app waits, because the result is often still a full 30 seconds.

A clear, long enough prompt with a scene structure makes these questions less frequent.

Checklist before you press Generate

• One line, no breaks in the middle.
• 1500–2000 characters.
• Written in Chinese (recommended).
• With images, the first sentence says who image 1, image 2 are.
• No names of copyrighted buildings, landmarks, characters, brands or songs.
• Each scene has an action, a transition and a pace.
• One closing phrase for lighting, colour and style.
• Model Dreamina Seedance 2.5, length 30 seconds.

Summary

• A 30-second video = 3–4 scenes, each with action, transition and pace.
• One line, 1500–2000 characters, no length or model written in.
• Chinese usually lands closer to your intent.
• With images: name each one in the first sentence, then just use the names.
• Avoid copyrighted names; close with lighting and colour.
• A 30-second shot never answers Dola's "make it shorter" question; the app opens a new chat.