A product video can look convincing until a hand picks up the object. Fingers merge with the surface, the grip changes halfway through, or the product floats after release. For Seedance 2.5 hand-object prompts, describe the contact sequence as clearly as the movement: how the hand starts, what remains held, where the object goes, and when the hand lets go.
Begin with one hand, one object, and one visible action. If picking up the product is not essential, start with the hand already holding it. That removes one transition from the shot you need to evaluate.
Why can the product look correct while the action looks wrong?
Product appearance and hand–object contact are different checks. The container may retain its shape while the fingers pass through it, or a hand may look natural while the product moves without support. This guide uses Seedance 2.5 on Seedance2Video to plan and review those interactions; the examples are generation instructions, not a guarantee of physical accuracy.
Review the relationship between surfaces over time, especially when support changes from a hand to a table. Increasing output resolution will not correct a release that happens in the wrong order.
Plan contact before camera movement
Break a handling action into observable states:
| State | Question to answer | Example |
|---|---|---|
| Starting contact | Is the object already held or resting on a surface? | A hand already holds a plain cup |
| Movement | What moves, and in which direction? | The cup travels a short distance downward |
| Support | What supports the object at the destination? | The cup base rests on the table |
| Release | Does the grip remain closed or open? | Fingers release only after the cup is supported |
You do not need every state in every shot. A holding shot can keep the same contact throughout. A placement shot needs a clear change from hand support to surface support.
Avoid starting with “the person picks it up, opens it, pours, closes it, and puts it away.” Those are separate actions with different contact conditions. Decide which one communicates the product's purpose.
Choose a reference that shows the grip
For image-led work in Seedance 2.5, use a clear source image that shows the relevant hand, the object, and enough of the supporting surface to understand the scene. Check the source itself for merged fingers, an impossible grip, or hidden geometry before asking for movement.
If the hand is already holding the object, preserve that relationship in the prompt. If the product is standing alone, a new reaching hand requires the generation to construct an interaction that the source does not show. It may be appropriate, but it is not equivalent to animating an existing grip.
Keep labels and functional details in view when they matter. A shot that rotates a product into an unseen back view adds an appearance question as well as a movement question. For product identity checks, use the broader AI video QA checklist.
Three prompts for different contact tasks
The examples below are starting instructions to test against your own references. They are not measured success cases or guarantees of correct anatomy.
1. Hold a cup without changing the grip
Source: One visible hand already holding a plain cup; fingers and cup edge are readable.
Use the supplied image as the visual reference. One hand is already holding the plain ceramic cup in the starting position. Keep the same grip and finger contact throughout the shot. The cup and hand move together a small distance toward the center of the frame, then hold still. Preserve the cup's shape, rim, color, and scale. Keep the entire contact area visible. Fixed close camera, steady soft lighting, unchanged background. No rotation, release, extra hand, or newly appearing object.
This shot tests whether the object and hand remain one coherent moving relationship. Watch the edges where the fingers overlap the cup; a convincing first frame is not sufficient.
2. Place an already-held product on a table
Source: One hand holding a small plain box just above a visible tabletop.
Use the supplied image as the visual reference. One hand already holds the small box above the table. The hand lowers the box vertically until its base rests flat on the tabletop. After the box is supported and still, the fingers release their grip and the hand withdraws sideways. The box remains stationary on the table. Preserve its shape, size, color, and visible edges. Show the grip, base, and table contact throughout. Fixed camera, one continuous action, steady lighting. No sliding, bouncing, rotation, second hand, or camera cut.
Review the order of events. Does the hand release before the object reaches the table? Does the box keep following the hand afterward? These are contact failures even if the final frame looks normal.
3. Lift a product without revealing another side
Source: A small plain container on a table, with one hand positioned beside it and a clear route for lifting.
Use the supplied image as the visual reference. One hand reaches a short distance to the container, closes around its side, and lifts it straight upward by a small amount. Once lifted, the hand holds the container still. Keep the front face oriented toward the camera and preserve the container's shape, proportions, and surface details. The grip remains consistent after contact. Show the whole hand and container in a fixed medium close-up. No twisting, opening, handoff, extra fingers, or additional objects.
This is a more demanding sequence than the already-held examples. If the reach or initial grip repeatedly fails, prepare a reference with the product already held and test the lifting segment separately.
Rewrite the failing transition
| Observed defect | Focused revision | Next check |
|---|---|---|
| Fingers change after contact | Start with an existing grip and remove rotation | Contact edges through the full movement |
| Product passes through fingers | Reduce travel and keep the contact area visible | Whether the same surfaces remain in contact |
| Object floats during placement | Specify surface support before release | Frame sequence around first table contact |
| Product changes behind the hand | Reduce obstruction and shorten the hidden interval | Appearance immediately before and after obstruction |
| Hand and product move independently | Request one short shared movement with a stable grip | Relative motion between hand and object |
| More than one hand appears | Remove ambiguous actors and describe one visible hand | Entire frame, including edges and reflections |
Change one instruction at a time. Keep a note of the earliest frame where the interaction becomes wrong. If you change framing, object, movement, and lighting together, you lose a useful comparison even if the next result looks better.
For problems beyond contact—such as an unwanted orbit or changing scene lighting—use the general prompt troubleshooting guide.
Record the exact contact failure
Use a frame or time reference so that the next revision addresses an observed problem:
Take / prompt version:
Source image and selected settings:
Intended action — hold / lift / place:
Starting grip visible and plausible:
First contact failure — timestamp and description:
Object supported before release:
Product details before and after obstruction:
Chosen action — keep / simplify / regenerate / use real footage:
One prompt instruction to change:
Keep an unchanged copy of the reference and the failed output if you plan to compare attempts. A published example should show what actually happened, including limitations, rather than present a prompt as proof of success.
Decide whether the action belongs in the final video
Watch at normal speed for believable movement, then step through the reach, contact, obstruction, and release. Compare product details with the source. A subtle finger defect can become the most distracting part of a close product shot.
If the interaction is optional, replace it with a stable product shot, a camera move, or a cut between two honest states. If the video must demonstrate how the real product opens, locks, fits, or operates, use verified footage for that operation. A plausible generated action is not evidence of real functionality.
Keep handoffs between people out of the first test. They combine two grips, two actors, changing support, and possible occlusion. Plan those as a separate sequence; the multi-character prompt guide covers assigning actions to different people.
FAQ
Does writing “no extra fingers” solve hand problems?
It is a constraint, not a repair guarantee. A clear source grip, readable contact, and a simpler action provide more useful direction than adding a long list of negative instructions alone.
Why does a product change after a hand covers it?
The hidden interval makes continuity harder to inspect and maintain. Reduce the obstruction or choose a view where more of the product remains visible. Verify the actual output rather than assuming the prompt preserved it.
Should I split picking up and putting down into separate clips?
Split them when the combined interaction is difficult to review or repeatedly fails. Check the end and start states before joining the clips so the edit does not introduce a new position or grip jump.
Can I use these prompts for clothing?
The contact planning is useful, but fabric adds deformation and pattern continuity. Use the clothing product consistency guide for garment-specific details.
