EastPrompt

AI video prompt generator for Seedance, Kling, Wan and MiniMax H3

Describe your scene once. EastPrompt writes it the way each model’s own documentation asks for, so timestamps, sound tags and “no music” lines end up where the model looks for them.

Sound and dialogue
Shots (optional)
Leave out
Prompt for Seedance 2.5

Last updated: 2026-10-03

Same scene, five formats

The leading Chinese video models all read English prompts, but each vendor documents a different structure. These are the rules the generator applies, taken from the official prompt guides and API references.

ModelRecommended orderSeveral shots in one clipThings to leave outPrompt limit
Seedance 2.5Subject, action, scene, visual style, camera, soundWhole-second timestamps: “0-3s: … 3-6s: …”No negative field. “No subtitles” and audio instructions such as “No BGM” are followedAbout 1,000 English words (recommended)
Seedance 2.0Subject and action, scene, atmosphere, camera, then style and constraints“Shot 1: … Shot 2: …”; ByteDance says its exact-timing control is unstableNo negative field. Constraint phrases such as “keep no subtitles”About 1,000 English words (recommended)
Kling 3.0Subject, movement, scene, then camera language, lighting, atmosphere“shot 1, 3, …; shot 2, 3, …;” with up to 6 shots; seconds must add up to the clip lengthNo negative field in the current API; write negative sentences in the prompt3,072 characters
Wan 3.0Overall description, timed shots, dialogue, sound, style, negative list“Shot 1 [0-3s] …”; “Generate single shot.” for one takeNo negative field; end with a negative list20,000 characters
MiniMax H3One core-concept sentence, then a shot-by-shot timeline“0–3 seconds: …”; cuts between shots by defaultNo negative field; end with exclusions such as “Non-diegetic music: N/A”7,000 characters

Why the format matters

These models read structure, not just words. ByteDance’s Chinese guide says Seedance 2.0 responds to shot numbers rather than timestamps, while Seedance 2.5 follows whole-second timestamps. Kling 3.0’s multi-shot format requires the seconds of each shot to add up to the clip’s duration. Wan 3.0 decides by itself whether to add dialogue unless you write “No dialogue.”, and Alibaba’s guide says to write “No background music.” if you want none. MiniMax H3 cuts between shots by default, so a single take needs the shot list removed and the words “one continuous take”.

The generator applies these rules for you, so your attempts go into the idea rather than the syntax.

How to use the AI video prompt generator

  • Pick a model. Fill in a subject, an action and a scene; every other field is optional.
  • Add shots if you want cuts. Give each shot a length in seconds and the tool turns them into that model’s timeline syntax and adds up the duration.
  • Tick the boxes for anything to leave out, such as subtitles or background music. Each model gets the wording its guide recommends.
  • Copy the prompt. Set duration and aspect ratio in the generator you use; they are settings, not prompt text.

The models at a glance

ModelMakerClip lengthResolutionWhere to use it
Seedance 2.5ByteDance4–30 s480p, 720p, 1080pDreamina by CapCut (its Seedance 2.5 rollout list doesn’t include the US); Jimeng AI and Doubao in China
Seedance 2.0ByteDance4–15 s480p, 720p, 1080p, 4KBytePlus and Volcano Engine APIs; Replicate, OpenRouter
Kling 3.0Kuaishou3–15 s720p, 1080p, 4KKling AI (kling.ai), 可灵AI in China
Wan 3.0Alibaba2–30 s480P, 720P, 1080PAlibaba Cloud Model Studio API; fal, Replicate, Runware and other API platforms
MiniMax H3MiniMax4–15 s768P, 2KHailuo AI (hailuoai.video)

Frequently asked questions

Which of these models is cheapest per second?

At official list prices checked on October 3, 2026, the cheapest are Wan 3.0 and MiniMax H3 Max at 480P, both $0.05 per second with audio. At 1080p, Kling 3.0 without audio is cheapest at $0.112, ahead of Wan 3.0 at $0.20 and Seedance 2.5 at about $0.569. The price calculator compares every provider at each resolution.

Do these models support negative prompts?

Not as a separate field in their current APIs. Seedance 2.5, Kling 3.0, Wan 3.0 and MiniMax H3 all expect exclusions written into the prompt itself. The legacy Kling API and Wan 2.x still have a negative_prompt parameter.

Should I write prompts in English or Chinese?

All of them accept English, and none of the official guides says one language works better. Seedance 2.5 also accepts nine more languages, including Spanish, Japanese and Korean. Keep each line of dialogue in a single language.

Is the generator free?

Yes. It runs in your browser, needs no account and stores nothing. It writes prompts; it does not generate video.

Where do the rules come from?

From each vendor’s own prompt guide and API reference, checked on October 3, 2026. The links are listed below.

Sources: Seedance 2.5 prompt guide · Seedance 2.0 prompt guide · Kling text-to-video prompt guide · Kling 3.0 API reference · Wan 3.0 prompt guide · MiniMax H3 cookbook

All tools