Write a video prompt that actually holds
Fill in the subject, one camera move, one lighting setup and a look, and the builder assembles a structured prompt with the aspect ratio and duration already attached. Copy it, or send it into the workspace with one click. Building a prompt is free and needs no account.
Esta página está en inglés. El espacio de trabajo, los precios y las preguntas frecuentes están disponibles en español.
A matte ceramic coffee mug with a matte black lid rotates slowly to reveal the label and texture. Camera: slow dolly push-in. Lighting: soft studio key light with gentle fill and a clean falloff. Keep the product label, proportions and colour accurate throughout. Look: clean commercial look, high-key lighting, accurate product colour. No captions, no overlaid text, no watermarks. Aspect ratio 9:16. Duration 8 seconds.
- Seedance 2.5 outputs 480p or 720p only. For 4K, switch to Seedance 2.0 Pro or Veo 3.1 — Veo 3.1 renders 5 second clips.
- Text on packaging is the first thing to break. Shoot the source photo flat, evenly lit and high resolution, keep camera movement slow, and check the label frame by frame before a paid campaign.
- A working video prompt names the subject and the action, then one camera move, one lighting setup, one look, an aspect ratio and a duration.
- This builder assembles that structure for you and states the model limits next to the result.
- Building and copying a prompt is free; generating the clip needs an account and credits.
- The prompt is text, not magic: it cannot add capabilities the selected model does not have.
What the builder covers
What makes a video prompt work?
Structure beats adjectives. The model needs to know what is in frame, what moves, how it is lit and how long it lasts — in that order.
- Subject and action first: one sentence, one subject, one beat.
- One camera move: two moves in one clip usually produces neither.
- One lighting setup: mixed lighting is the most common cause of a muddy result.
- An explicit look: photoreal, cinematic, commercial, 3D render, anime or documentary.
- Aspect ratio and duration last, so the model treats them as constraints rather than decoration.
What does each field change?
Each field maps to a phrase in the output, and the phrase is what the model reads.
| Field | Effect on the output | Example phrase |
|---|---|---|
| Subject | What is in frame; usually supplied by the image in image-to-video | a matte black ceramic mug |
| Action | The beat inside the shot | rotates slowly to reveal the label |
| Camera | Where the viewer stands and how they move | slow dolly push-in |
| Lighting | Direction and quality of light | soft studio key with gentle fill |
| Look | Grade, texture and rendering style | clean commercial look, accurate colour |
| Aspect + duration | Delivery format and runtime | 9:16, 6 seconds |
How do you use the prompt?
Build, copy, generate — the handoff keeps the text intact.
- 01
Describe the subject in plain words
Avoid brand names and abstract adjectives. 'A stainless steel water bottle with a black cap' works; 'premium hydration experience' does not.
- 02
Pick one camera move and one lighting setup
The builder writes them as separate clauses so the model does not have to infer them from atmosphere words.
- 03
Copy the prompt, or open it in the workspace
The Generate button passes the prompt as a URL parameter, so the workspace opens with the text already filled in and Seedance 2.5 selected.
Where prompts still fail
A good prompt cannot fix the wrong model, the wrong source image or an impossible request.
- Asking for 4K from Seedance 2.5 — it outputs 480p or 720p only.
- Asking for 60 seconds in one generation — the cap is 30 seconds on Seedance 2.5.
- Expecting readable packaging text; generated typography distorts first.
- Expecting the same face from text alone; that needs reference images.
- Very long prompts: past roughly 150 words, later clauses get less attention, not more.
Prompt questions
Is the prompt generator free?
Yes. Building and copying a prompt costs nothing and needs no account. Only generating the clip consumes credits.
Do I need to log in to use it?
Not to build or copy a prompt. Generating requires an account because the job runs against your credit balance.
How long should a prompt be?
One tight paragraph, roughly 60 to 120 words. Beyond that, later clauses are attended to less, so length adds noise rather than control.
Does the generated prompt guarantee the result?
No. A prompt constrains the generation; it does not determine it. Expect variation between runs and inspect each output.
Which model does the prompt target?
The builder writes for Seedance 2.5, which is preselected in the workspace, and flags its limits — 480p or 720p, 4 to 30 seconds, native audio. The same prompt works on other models with different limits.
How many credits will the clip cost?
The builder shows the Seedance 2.5 estimate for the chosen duration: 15 credits per second at 480p and 30 at 720p, for durations between 4 and 30 seconds.
Can I use the prompt for image-to-video?
Yes. In image-to-video the image supplies the subject, so keep the subject line short and spend the words on motion, camera and look.
Why does the builder warn about 4K?
Because Seedance 2.5 outputs 480p or 720p only. A prompt asking for 4K on that model does not change the output, so the builder tells you which models actually render 4K.
Can I save prompts for later?
Copy them into your own notes, or keep a shot list outside the tool. The workspace keeps the generated clips, not a prompt library.
Does the audio direction really produce sound?
On Seedance 2.5 yes — the model generates a synchronised audio track with the clip. The audio is generated, not a licensed music library, so clear rights before commercial use.
Key takeaways
- 1.Structure the prompt: subject, action, camera, lighting, look, format.
- 2.One camera move and one lighting setup per clip.
- 3.Respect the model's real limits — 480p or 720p and 4 to 30 seconds on Seedance 2.5.
- 4.Copy the prompt or send it straight to the workspace; building it is free.
Related tools
Apply the prompt to a product photo and generate.
Add reference images so the face stays stable.
Extend the same prompt into a longer sequence.
Every duration, resolution and credit rate in one table.