Back to all articles

Seedance guides

Seedance AI Prompts: Templates, Camera Directions, and Fixes

Adapt text-to-video and image-to-video prompts for products and stories, with camera examples and ChatGPT or Claude drafting instructions.

By sdanceai product teamNov 11, 20264 min read

A useful Seedance prompt describes what happens during a shot. Text-to-video needs a scene to create; image-to-video already has a visual starting point and benefits from clear motion instructions. Use the templates below as starting points, not guaranteed output recipes.

Use a simple structure

Subject + action + setting + camera + lighting + sound. Start with one visible action. Add details that matter to the result instead of stacking quality adjectives. Camera motion and subject motion are different instructions.

For example, “cinematic product, beautiful, high quality” does not tell a bottle to move. “The bottle stays still while the camera slowly pushes closer” describes an observable shot.

Product photo to video

Open Image to Video, upload a sharp photo you can use, and choose a compatible Seedance model.

Use the uploaded skincare bottle as the subject. It stays upright on the same surface, with its shape and packaging unchanged. The camera slowly pushes in. Soft light from the left creates a gentle moving reflection on the glass. No hands enter the frame. Quiet indoor ambience, no speech.

If packaging drifts, reduce the camera movement and simplify the reflection request. Check logos and lettering manually; a prompt cannot guarantee exact reproduction. For a silent model, remove the sound sentence.

Cinematic text-to-video scene

Open Text to Video:

A cyclist wearing a yellow raincoat rides slowly along a wet city street at dawn. The camera tracks alongside at wheel height in one continuous shot. Warm shop lights reflect on the pavement. The cyclist passes a closed flower stall. Gentle rain and bicycle sounds, no dialogue.

If the subject stays frozen, make the action more visible: “The cyclist pedals steadily from left to right; both wheels turn.” Keep the camera instruction separate.

Five camera instructions to adapt

Shot Starting instruction Check in the result
Fixed camera The camera stays locked while the person opens the door. The background should not drift with the subject.
Push-in The camera slowly moves toward the stationary cup. The cup should not transform as it gets larger.
Pan The camera pans right to follow the walking person. The action and camera direction should agree.
Tracking The camera moves alongside the cyclist at the same speed. The subject should stay framed during movement.
Orbit The camera makes a small, slow arc around the product. Avoid asking for a full orbit if unseen product details matter.

Do not combine an orbit, zoom, rapid cut and handheld shake in the first attempt. Establish one movement before adding complexity.

Keep a character consistent across separate shots

Reuse the same clear reference in compatible workflows. Repeat identifying details such as a navy jacket and short curly hair. Write one action per shot and keep the lighting direction compatible. References guide the output; they do not guarantee identical faces across every frame.

A starting brief for two shots:

  1. The character in the reference stands by a cafe window and looks toward the door. Static camera, soft morning light.
  2. The same character walks through the cafe doorway. Slow side tracking shot, the same jacket and hairstyle, matching morning light.

Generate and check each shot separately, then assemble the clips in an editor. Do not describe two different appearances for the same character.

Ask ChatGPT or Claude to draft a prompt

Copy this instruction into your writing assistant, then replace the brackets:

Write one Seedance video prompt for [task]. My source is [text only / description of my reference image]. Describe one subject action, the environment, one camera movement, lighting and optional sound. Keep these details unchanged: [details]. Use only the controls I provide: [available model settings]. Avoid unsupported features and do not promise perfect consistency. Give me one prompt and one simpler fallback.

Review the answer, then paste the prompt into the video generator. This workflow does not require an integration between this site and the writing assistant. Share only source material you are allowed to share.

Improve one thing at a time

  • Weak motion: replace vague mood descriptions with a physical action and direction.
  • Face or product changes: use a clearer reference, reduce motion and remove conflicting appearance changes.
  • Confusing camera: keep one camera direction and one continuous shot.
  • Missing audio: check that the selected model supports audio and that its audio control is enabled.
  • Too much story for one clip: split the story into shots that fit the model’s current duration options.

Every extra generation is a separate task. Review the credit estimate and plans before experimenting. Read the beginner guide for the full workflow and the trial guide for account eligibility.

Questions about your task? Email support@sdanceai.com with its ID. Never share API keys.