Start with the image you actually have
A useful image-to-video prompt describes what changes after the opening picture. The image already establishes the subject, setting and composition. Rewriting every visible detail can introduce a competing description. Begin by identifying the main action, the camera behavior and the features that must remain stable. A short, concrete instruction is easier to evaluate than a collection of cinematic adjectives with no visible endpoint.
Prepare the source before writing the motion. Check that the important subject is clear, that limbs or objects are not accidentally cut off, and that there is room for the intended movement. SeaDance video output is 16:9 only. A source composition that depends on a different shape may need preparation outside the generation workflow. A prompt cannot reliably recover unseen details.
Choose frames or references deliberately
According to ByteDance Seed, reference-based creation is a central emphasis of Seedance 2.5. According to MiniMax Research, language can describe relationships between context and a target result. On SeaDance, make that relationship explicit by choosing the correct image mode. The suggestions below are editorial writing techniques, not claims about measured success rates or guaranteed subject preservation.
| Model family | Image to Video inputs | Separate Reference mode | 5s base charge |
|---|---|---|---|
| Seedance 2.0 Mini | 1–2: first, optional last | 1–9 still images | 4 credits / $0.80 |
| Seedance 2.0 / Fast / 2.5 | 1–2: first, optional last | 1–9 still images | See each model quote |
| Kling 3.0 | 1: first frame only | Not available on SeaDance | 13 credits / $2.61 |
| MiniMax H3 / H3 Max | 1–2: first, optional last | 1–9 still images | 9 credits / $1.81 |
Prices checked November 4, 2026. Every dollar estimate uses the Basic monthly plan: $29.90 for 149 credits, or about $0.20 per credit. Dollar amounts allocate a prepaid balance; they are not individual checkout prices. Whole-request credit charges are rounded upward before conversion to dollars. New accounts start with 0 credits and require a paid plan or credit pack to generate.
In Image to Video, the first uploaded image establishes the opening frame and the optional second image provides an ending target on supported models. Kling accepts only the first. Reference mode instead uses still images as visual context; it is not an ordered sequence of endpoints. Frame inputs have no reference-image surcharge. H3 Reference-mode totals can increase with the image count, so check the quote after switching modes.
A compact prompt structure
Write the prompt as action, camera, sound and preservation. Say what the subject does, how the camera moves or stays still, which sound belongs in the scene, and what cannot change. Tie the action to a visible finish, such as the subject settling or the camera reaching a closer composition. Avoid adding unrelated events simply because the selected duration leaves room for more words.
Use one main camera move. A gentle push toward the subject is different from a pan that turns the viewpoint, and either is different from asking the subject to move through a static frame. If the subject must remain the visual anchor, say so. Keep preservation instructions concrete: maintain the face, clothing, product silhouette, label placement or room layout. Such instructions communicate intent; they are not a guarantee of exact reproduction.
Six original examples to adapt
The following prompts assume you have prepared suitable images that you own or may use. They are original starting points for short continuous shots. Replace the object names with details that are visible in your source. Keep the instruction about what stays stable when adapting the motion. None of these examples requires a specialized input beyond SeaDance's still-image modes.
Product highlight. Use the uploaded bottle as the opening frame. Keep the camera still while a soft window reflection travels across the glass, then settles. Preserve the bottle shape, cap, label placement and tabletop. Sound: quiet room ambience. Keep the background unchanged and avoid introducing hands or additional objects.
Portrait with subtle movement. Begin from the uploaded portrait. The subject glances toward the window and returns to a relaxed forward gaze. Keep the camera fixed. Preserve the face, hairstyle, clothing and background arrangement. Sound: soft room tone. Keep the expression understated and avoid adding speech or another person.
Interior camera approach. Start with the uploaded room composition. Move the camera slowly forward toward the armchair as the curtain shifts gently in a breeze. Preserve the furniture positions, wall geometry and daylight direction. End in a slightly closer view with the chair still fully readable. Sound: a faint outdoor breeze.
Outdoor subject in a static frame. Use the garden photograph as the starting image. Keep the camera still while the visible leaves sway gently and the subject's coat moves with the same breeze. Preserve the person's appearance, path and surrounding plants. Let the movement settle before the ending. Sound: soft wind and distant birds.
First-to-last product transition. Use the first image as the opening and the second as the ending target in a supported Image to Video model. The same ceramic cup turns gently into the final orientation and stops. Keep the tabletop, lighting and cup proportions consistent throughout. Use a stationary camera and quiet room ambience.
Prepared doorway endpoint. Begin from the closed-door first image and finish toward the open-door last image on a supported model. The door opens smoothly on its existing hinge while the camera remains still. Preserve the wall, handle, frame and lighting. Reveal only the room shown in the ending image. Sound: a soft hinge movement.
Common failures and specific repairs
If the subject changes appearance, first remove prompt details that contradict the source. Then name the features that matter most, such as clothing or object geometry. If the camera wanders, replace several camera instructions with one clear move and endpoint. If background objects appear unexpectedly, state which environment should remain and remove scene additions that compete with it.
If the transition feels abrupt, check whether the endpoints describe the same plausible scene. A changed viewpoint, wardrobe and environment all at once asks for much more than a simple movement. Prepare more compatible frames or use a simpler action. If motion feels rushed, remove secondary events before buying more duration. A longer clip can give an action space, but it cannot make conflicting instructions agree.
Review one problem at a time
Use Image to Video for an opening-frame animation, and read the duration guide before requesting more time. Keep the source, prompt and settings together when reviewing a take. Identify whether the failure is in subject consistency, camera behavior, action completion or sound, then revise the relevant instruction instead of replacing the entire brief.
For lower-cost exploration, the Mini model can help test a simple action, while Kling and H3 provide other output tiers. A successful draft is not proof that another model will reproduce it. Use the cost guide to reserve enough balance for checking the final model with its actual inputs.
Frequently asked questions
Can I upload an ending frame?
Seedance and H3 video models accept an optional last frame after the starting frame in Image to Video. Kling accepts only a starting frame on SeaDance.
Is Reference mode the same as first and last frames?
No. Reference mode uses still images as visual context. Image to Video assigns the opening image and, where supported, an optional ending target.
Should I describe everything visible in the photograph?
Focus on motion and preservation. Repeating the entire scene can introduce competing details, so describe new information and the features that matter most.
Can a preservation instruction guarantee identical details?
No. It communicates your intent but does not guarantee exact faces, lettering or geometry. Review the generated take against the requirements of your own project.
What should I change after a failed take?
Identify the specific failure first. Simplify conflicting motion, clarify the camera move, or prepare more compatible frames before changing several settings at once.
Sources
External background checked November 4, 2026. SeaDance settings and credit quotes are described separately above.