Updated September 5, 2026.
Quick Answer
The most useful AI video prompt formula is:
Subject + Action + Camera + Lighting + Style + Audio
Use it as a shot brief, not as a list of random keywords. Tell Seedance 2.5 what is in the frame, what changes, how the camera moves, how the scene is lit, what visual language to use, and what the viewer should hear. For image-to-video, let the source image define the subject and composition; use the prompt mainly for motion, camera direction, timing, and sound.
What an AI Video Prompt Is
An AI video prompt is a compact production brief. It describes the visible subject, the action over time, the camera language, the look of the scene, and—when supported—the sound direction. It is closer to a director's note than to a keyword list.
This distinction matters because video has a time dimension. A still-image prompt can describe a final composition. A video prompt must also describe what moves, when it moves, and how the camera reveals it. If the prompt asks for five unrelated actions, the model has to compress too much into one clip. If it gives no camera instruction, the model invents one.
For the model's capabilities, consult ByteDance's official page, then choose the workflow that matches the asset you have: text-to-video when you are creating a scene from an idea, and image-to-video when you already have a product frame, character image, or visual reference. The Seedance 2.5 tutorial covers the broader workflow, while this guide focuses on prompt construction.
Write the same brief in three levels of detail
Start with an observable action, then add only the information needed to control it. Here is a worked rewrite for a tea-shop social clip:
| Draft | Prompt | What the revision resolves |
|---|---|---|
| Vague | “Beautiful cinematic tea video.” | The subject's action and camera behavior are unspecified. |
| Testable | “A hand places a white teacup on a wooden counter. The camera stays at counter height in a medium close-up.” | Establishes one event and a fixed viewpoint. |
| Directed | “A hand places a white teacup on a wooden counter, then leaves the frame. Keep the camera at counter height and slowly push closer. Soft window light from the left, quiet ceramic contact and room tone. End on the cup with space above it for a caption.” | Adds an ending, light source, sound and a usable text area. |
For image-to-video, replace the subject description with an instruction to preserve the uploaded cup and composition. When a real brand mark must be exact, plan an overlay or a controlled filmed insert; the prompt is a request for consistency, not a guarantee of exact lettering.
Add reference roles and timed beats only when needed
Use the reference labels produced by your uploader. Labels such as @Image1 in examples are placeholders, not a universal syntax that every service accepts. Assign each input a specific role: product appearance, room layout, camera motion or sound. If two references disagree, state which one controls the conflicting property.
For a longer tea-shop scene, a compact plan might be:
Opening: establish the counter and the empty place where the cup will land.
Middle: the hand places the cup, releases it and leaves the frame.
Ending: move closer to the steam and hold the final cup composition.
Preserve: cup design, counter layout and window-light direction throughout.
Convert those beats into time ranges that fit your selected duration. Do not ask for “no cuts” and then request unrelated locations without describing how the camera connects them. If the plan needs separate scenes, specify transitions or generate shots separately.
Use the prompt library when you want a ready-made subject, and the prompt troubleshooting table when you already have a result to diagnose.
The Six-Part Formula
| Part | What to specify | Example |
|---|---|---|
| Subject | Who or what is visible | A matte-black perfume bottle on wet stone |
| Action | One continuous change | Rotates slowly as a highlight travels across the glass |
| Camera | Shot size, angle, movement | Macro close-up, slow push-in |
| Lighting | Direction, quality, color | Soft cool key light from camera left |
| Style | Visual language and finish | Premium beauty commercial, shallow depth of field |
| Audio | Ambient sound or music direction | Quiet water drops and a restrained synth pad |
1. Subject: Make the Frame Concrete
Name the main subject and the details that must remain stable. “A product” is too broad. “A matte-black perfume bottle with a silver cap and a small embossed label” gives the model a usable visual anchor.
For a character, describe only identity-relevant details: age range, hairstyle, clothing, and expression. Do not write a full biography. The goal is a stable frame, not more prose.
2. Action: Give the Clip One Motion Arc
Use a beginning, a change, and an end. For example: “The bottle remains centered while it rotates 30 degrees; a soft highlight moves from the cap to the label.” This is easier to execute than “the bottle spins, explodes into flowers, flies through a city, and returns to the table.”
For a 30-second scene, add a small arc rather than several unrelated scenes: “The creator opens the box, lifts the product, and turns it toward the window.” Keep the main action singular and let the camera create visual variety.
3. Camera: Use Film Language
Camera language is one of the fastest ways to change the result. A prompt can combine shot size, angle, and movement:
- Shot size: wide shot, medium shot, close-up, macro close-up
- Angle: eye-level, low angle, overhead, three-quarter angle
- Movement: slow push-in, dolly-out, pan left, tracking shot, orbit, crane up, locked-off
Use one primary movement per shot. If you need a transition, state the order: “Start in a wide shot, then slowly push in to a medium close-up.” For more camera vocabulary and examples, read the camera movement prompt guide.
4. Lighting: Describe the Source, Not Just the Mood
“Cinematic lighting” is a mood label, not a complete lighting brief. Say where the light comes from and what it does: “warm window light from the right, soft shadow on the left side of the face.” For product work, specify whether the surface should look glossy, matte, wet, or metallic.
5. Style: Choose One Visual Direction
Pick one dominant style: clean product commercial, handheld documentary, 35mm film, editorial fashion, stop-motion miniature, or clean 3D explainer. Combining “photorealistic, watercolor, anime, documentary, and luxury commercial” creates conflicting instructions.
6. Audio: Treat Sound as Part of the Shot
If your workflow supports generated audio, add a short sound direction after the visual direction. Use concrete sounds—“espresso machine hiss and quiet café ambience”—instead of “epic sound.” Keep the audio request consistent with the scene and do not ask for a full soundtrack, dialogue, sound effects, and voiceover in one short prompt unless the shot truly needs them.
Text-to-Video Example
Weak prompt
A cinematic product video of a perfume bottle, beautiful lighting, luxury, high quality, dramatic, amazing camera movement.
Strong prompt
A matte-black perfume bottle with a silver cap stands on dark wet stone. The bottle rotates slowly 30 degrees while a narrow highlight travels across the glass and reveals the label. Macro close-up, eye-level camera, slow push-in from the cap to the label. Cool blue key light from camera left, soft violet rim light behind the bottle, controlled reflections. Premium beauty commercial, shallow depth of field, clean high-end finish. Quiet water drops and a restrained low electronic pad.
The second prompt is easier to evaluate because each sentence has a job. The subject is concrete, the motion is singular, the camera is explicit, and the lighting and audio are compatible with the scene.
Image-to-Video Example
Image-to-video prompts should not re-describe everything already visible in the source image. Use the image as the visual anchor and direct the change:
Preserve the product shape, label, color, and background composition from the reference image. Slowly rotate the bottle 20 degrees toward camera right while a soft highlight sweeps across the label. Use a subtle camera push-in and keep the product centered. Maintain the original lighting and clean commercial style. Add quiet studio ambience with a single glass-water drop.
The phrases “preserve,” “keep,” and “maintain” establish invariants. They tell the model what must not drift while the motion instruction tells it what may change.
Prompt Checklist
- [ ] Is there one main subject?
- [ ] Is the action a single continuous arc?
- [ ] Did you specify shot size and camera movement?
- [ ] Did you describe the light source and shadow direction?
- [ ] Is there one coherent visual style?
- [ ] Is the audio direction short and scene-appropriate?
- [ ] For image-to-video, did you state what must stay unchanged?
- [ ] Did you avoid contradictory styles and too many scene changes?
- [ ] Can another person understand the shot without asking a follow-up question?
FAQ
How long should an AI video prompt be?
Long enough to define the subject, action, camera, lighting, style, and audio—usually one to three focused sentences. More words are useful only when they remove ambiguity.
Should I use negative prompts?
Use positive instructions first. Instead of listing ten things that must not happen, state the desired invariant: “keep the character's face and outfit consistent throughout.” This leaves less room for conflicting instructions.
Should the prompt describe every frame?
No. Describe the action arc and the key visual states. Over-specifying every frame can make the motion rigid or contradictory.
Is the same formula useful for image-to-video?
Yes, but the emphasis changes. The source image handles subject and composition; the prompt should focus on motion, camera, timing, and audio.
Put the guide into practice
Start with the Seedance 2.5 workflow, then use the text-to-video workspace to test one variable at a time. Once the structure is stable, use the 20 prompt examples or diagnose a completed result with the troubleshooting guide.
