Seedance Prompt Guide: Dance and Photo-to-Video Prompts

Updated 2026-10-13

On this page

A good Seedance prompt for one photo describes the motion, not the picture. The photo already shows the person, the clothes and the room. Your prompt should say what moves, in what order, how the camera behaves and what you hear.

This guide gives a simple prompt order, seven patterns for dance and photo-to-video, and examples you can copy. The examples are ours. We wrote them as starting points and did not test them on every platform.

Key facts

Order that works: reference, scene, action, camera, timing, sound. This order comes from a published Seedance 2.0 prompt guide [1].
Photo input: an image can start the video as a first frame, or it can act as a reference. The two modes are separate on the API platform we checked [2].
Length: Seedance 2.0 makes 4 to 15 seconds on that platform. Seedance 2.5 makes 4 to 30 seconds [2] [3].
Sound: Seedance makes audio with the video. The API setting is on by default and you can switch it off [2] [3].
Known weak spots (ByteDance, Seedance 2.0): detail stability, text on screen, several subjects and occasional audio distortion [4].
No prompt needed on Effect Farm: the templates at effectfarm.com write the request for you. You upload one photo.
Last checked: October 13, 2026.

Seedance prompt guide: how prompts work

Seedance is a video model from ByteDance Seed. It reads your text and your files, then makes a clip with sound. See What is Seedance for the basics.

With a photo, you have two jobs. First, tell the model what the photo is for. Second, tell it what should happen. A prompt that only repeats what the photo shows wastes words. A prompt that says what moves gives the model a plan.

The prompt order

Write the parts in the order of the video. A published Seedance 2.0 guide suggests this sequence: reference assignment, scene, action and motion, camera behavior, timing, then sound or emotional intent [1].

Part	What to write	Example
Reference	What the photo is for	"Use the photo as the first frame."
Scene	Place and light, only if they change	"A small living room, warm evening light."
Action	The motion, in steps	"The person steps left, steps right, then raises both arms."
Camera	One movement, or a fixed camera	"The camera stays still. Full body in frame."
Timing	How long each beat lasts	"Seconds 0 to 3: slow start. Seconds 3 to 8: faster."
Sound	Music, ambience or silence	"Upbeat pop beat. No voices."

Keep each part to one or two short sentences. A short, clear prompt gives the model less to misread.

Prepare the photo first

The photo matters more than the prompt. The API platform we checked lists these limits for a reference image on its 2.5 request [3]. The format is JPEG, PNG or WebP. The width-to-height ratio is between 0.4 and 2.5. Each side is between 300 and 6,000 pixels. The file is under 30 MB.

Beyond the limits, a few habits help:

Use one clear photo of one person.
Match the framing to the motion. A dance with legs needs a photo that shows legs.
Use good light and a plain background.
Avoid text, logos and crowds in the picture. ByteDance lists text on screen and several subjects as weak spots [4].
Platforms set their own rules for photos of real people, so read the rules of the platform you use. On Effect Farm, the person in the photo must be you, and you must be 18 or older.

Seven image-to-video prompt patterns for one photo and motion

1. The simple dance

Name the dance move, the rhythm and the camera.

Use the photo as the first frame. The person dances to an upbeat pop beat: step left, step right, raise both arms, spin once. The camera stays still with the full body in frame. Keep the same face, clothes and room.

2. The timed dance

Split the clip into beats. Timing helps when you want the energy to rise.

Use the photo as the first frame. Seconds 0 to 3: the person sways slowly and smiles. Seconds 3 to 8: the person dances faster with big arm movements. Seconds 8 to 10: the person strikes a pose and holds it. Fixed camera, full body.

3. The one-move loop

Ask for one repeating move. One move is easier to keep stable than a long routine.

Use the photo as the first frame. The person repeats one move: knee lift on the beat, arms swinging. The same move all the way through. Fixed camera.

4. The reaction clip

Not every photo video is a dance. A small reaction can work well from one photo.

Use the photo as the first frame. The person looks puzzled, scratches their neck, then smiles and laughs. Natural movement. The camera stays still at chest height.

5. The camera move

Choose one camera move. Do not ask for a push-in, a pan and an orbit in the same clip.

Use the photo as the first frame. The person stands still and breathes. The camera moves slowly closer to the face. Soft light. No other movement.

6. The scene change

Say what stays and what changes. Name the things to keep.

Use the photo as the first frame. Keep the person, the face and the clothes. Change the background from a kitchen to a rooftop at sunset. The person walks forward and waves.

7. The sound line

Describe sound the way you describe the picture. Say what you hear and what you do not want.

Use the photo as the first frame. The person dances. Sound: a steady drum beat and soft room tone. No speech. No music lyrics.

Mistakes to avoid

Re-describing the photo. The model can already see it. Spend your words on motion.
Too many moves. Five moves in five seconds look rushed. Pick one to three.
Two camera moves. One camera move per shot is easier for the model to follow. This is our rule of thumb.
Text on screen. ByteDance lists text rendering as a weak spot for Seedance 2.0 [4]. Leave captions for an editor.
Unlabeled references. If you give more than one file, say what each file is for. A published guide says to label every reference with its purpose [1].
Mixing modes. On the API platform we checked, a first-frame request and a reference request cannot be mixed [2] [3].
Long prompts for short clips. The 2.5 request on that platform allows up to 20,480 characters [3]. Allowed is not the same as useful. Short prompts are easier to debug.

How to fix a bad result

These are rules of thumb, not guarantees. Change one thing at a time.

If the face changes, use a sharper photo that faces the camera, and add "keep the same face".
If the motion is wild, cut the number of moves and add "smooth, natural movement".
If the camera drifts, write "the camera stays still".
If the sound is wrong, name the sound you want or switch audio off.
If hands or feet look wrong, choose a photo that shows them clearly.

A shortcut: use a template

A template writes the request for you. You upload one photo and pick a result. You do not write a prompt or choose settings.

Effect Farm is an independent site. Our templates run on Seedance. We do not name the version here. Try one of these:

Make Picture Dance: pick a preset dance for your photo.
Genjutsu AI: your photo copies one of our dances.
Jawn Dance AI: puzzled look, then a living-room dance.
Toc Toc AI Dance: a full-body dance in a small room.

One video costs 30 credits ($4.99 in the smallest pack). See pricing. For the full photo-to-dance tutorial, read how to make an AI dance video from one photo.

FAQ

What is the best Seedance prompt structure?

Write in the order of the video: reference, scene, action, camera, timing, sound. A published Seedance 2.0 guide uses this order. Keep each part short.

How long should a Seedance prompt be?

Short is fine. Three to six sentences cover the order above. The 2.5 request on the API platform we checked allows up to 20,480 characters, but a longer prompt does not make a better video.

Do I need a prompt to turn a photo into a dance video?

On Seedance directly, yes. On Effect Farm, no. The template contains the request. You upload one photo and pay for the video.

How do I make Seedance keep the face from my photo?

Use a sharp photo that faces the camera. Say "keep the same face and clothes". Ask for fewer moves. We cannot promise a perfect match, so check the result before you share it.

Can Seedance make a dance video from one image?

Yes. Seedance supports image-to-video, and the image can be the first frame. Describe the dance move and the camera in your prompt.

Can I write Seedance prompts in other languages?

Not confirmed here. The prompt guides we read use English. ByteDance says Seedance 2.0 improved Chinese dialects and singing, but that is about audio, not prompt language [4].

Does Seedance add sound to a photo video?

Yes. It makes audio with the video. You can name the sound in the prompt or switch audio off in the API settings [2] [3].

What is the difference between a first-frame prompt and a reference prompt?

A first-frame prompt starts the video from your image. A reference prompt uses your files as guides for the person, the look or the motion. On the API platform we checked, you pick one mode per request.

Make the video without a prompt

Upload one photo, pick a dance and get a video in 3–6 minutes. Open Make Picture Dance or Genjutsu AI, or see the pricing page. To compare Seedance with other models, read Seedance vs Kling vs Veo.

Sources

[1] RunDiffusion, Seedance 2.0 prompt guide: https://www.rundiffusion.com/seedance-2-0-prompt-guide [2] API documentation for Seedance 2.0 (third-party platform): https://docs.kie.ai/market/bytedance/seedance-2.md [3] API documentation for Seedance 2.5 (third-party platform): https://docs.kie.ai/market/bytedance/seedance-2-5.md [4] ByteDance Seed, official launch of Seedance 2.0: https://seed.bytedance.com/en/blog/official-launch-of-seedance-2-0