# One object, two hands: support the handoff before releasing

Practice one transfer between two moving hands: the receiver makes contact and establishes a visible hold before the giver opens and withdraws. The same rigid object should finish in the receiver's hand.

**Original practice method, not independently tested with MiniMax H3.** The images are authored teaching drawings, not video frames from a generated result. No model, physical simulation or video-generation service was run for this package.

## 01. Choose the correct input

[Open the clean input at full size](clean-start.png)

The coral-cuffed hand at frame left is **A, the giver**. The blue-cuffed hand at frame right is **B, the receiver**. There is one plain teal cube. These are stylized, original hand drawings with no real-person identity, brand or third-party reference media.

| File | Purpose | Upload? |
| --- | --- | --- |
| [clean-start.png](clean-start.png) | Opening composition with A holding the cube and B visibly separated | Upload this one image |
| [contact-timeline.png](contact-timeline.png) | Numbered teaching drawing for the intended order | No; inspect beside the result |
| [acceptance.blank.csv](acceptance.blank.csv) | Blank review sheet with pass, fail or not-observable outcomes | No |
| [Editable SVG and complete practice ZIP](handoff-practice-kit.zip) | Original source drawings, both briefs, prompts and inspection materials | Download for practice/review |

Open the [image-to-video generator](https://minimaxh3.studio/generator/image-to-video) and select **MiniMax H3**, the base model rather than MiniMax H3 Fast. Turn on **Add end frame** to reveal the **First Frame** and **Last Frame** slots. Upload only `clean-start.png` to **First Frame** and leave **Last Frame** empty; the base model accepts a first frame without an end frame. Check the currently accepted file type and settings. Opening the generator does not upload a file or start a job. Do not upload the four-panel timeline as a first frame: its repeated cube and hands describe different stages, not four simultaneous objects.

For a replacement, use a permitted image with both hands, the giver's grip, the receiver's open palm and the whole object visible. Start with a simple rigid shape; transparent objects, loose straps and printed labels add unrelated checks.

## 02. Separate the four contact states

[Open the four-state teaching diagram](contact-timeline.png)

| State | What should be visible | What must not happen yet |
| --- | --- | --- |
| 01 — Approach | A holds the cube. B approaches the underside with a clear initial gap. | A must not release during the gap. |
| 02 — Contact | B reaches the cube while A still visibly holds it. | A light touch is not the ending. |
| 03 — Shared hold | B's palm and fingers form a readable supporting hold; A retains contact during this transition. | The cube must not float, duplicate or jump between unrelated grips. |
| 04 — Release and hold | A opens and retreats; B continues holding the same cube. A and cube finish with a visible gap. | Do not perform a second transfer or return to A. |

The numbers express **order**, not promised timestamps or model controls. The drawing illustrates requested contact states; it does not demonstrate that a model produced them. A visible hold is a reviewable image relationship, not a measurement of weight, grip force or physical stability.

## 03. Write one handoff

Keep the camera, background and cube's identity fixed. Change the support relationship between the two hands. The following is an original English task brief:

```text
Use the uploaded drawing as the first frame. There are exactly two hands and one teal rigid cube. The coral-cuffed hand on frame left is the giver A; the blue-cuffed hand on frame right is the receiver B. Keep both roles, the cuff colors, the cube shape and the camera framing consistent.

Show one slow, continuous handoff. At first A holds the cube and B has a visible gap. B approaches from below and makes contact while A keeps its grip. Next, show a readable shared-contact interval in which B's palm supports the underside and B's fingers form a hold around the visible side. Only after that supporting hold is visible does A open its fingers and withdraw toward frame left. B keeps holding the same cube and settles.

End with exactly one cube in B's hand, A clearly separated, and no second handoff. Keep the background and camera still. Do not let the cube float between grips, become a second cube, change size, merge with fingers, fall onto an invented surface or pass through either hand. No text, arrows, extra hands, speech or scene cut.
```

Do not add an exact frame number, force threshold, simultaneous camera orbit, dialogue or a second object to this exercise. Choose only settings supported by the current generator. The wording requests a visible sequence; it does not guarantee frame-accurate timing or successful hand anatomy.

## 04. Inspect contact before the ending

Watch the whole motion once. Then pause around the first B-to-cube contact, the shared hold and A's release. An attractive final pose alone cannot establish the correct order.

1. **Count objects and hands.** Track the same cube outline through the transfer. Record an extra or disappearing object rather than overlooking it because the final count is correct.
2. **Locate first receiver contact.** Identify the first visible contact between B and the cube. Check that A still holds it at that point.
3. **Find the shared hold.** Look for B's palm under the cube and a coherent grip while A has not yet fully released. If fingers or the cube hide this interval, mark it not observable; do not infer a successful transfer.
4. **Find giver release.** Locate A's opening fingers and withdrawal. Compare this with the supporting hold, not just first fingertip touch.
5. **Review the ending.** B holds one unchanged cube; A is visibly separated. Check that camera movement, a cut or an invented tabletop has not substituted for the handoff.

In [acceptance.blank.csv](acceptance.blank.csv), enter an observed timecode only after reviewing an actual result. All outcome, timecode and note cells begin empty. Use `pass`, `fail` or `not-observable`; an obscured grip is not automatically a pass.

## 05. Revise the first visible failure

| First visible issue | Change one thing | Recheck |
| --- | --- | --- |
| A opens before B supports | Make the shared-hold-before-release clause more explicit; keep input and other settings unchanged | Is A still holding when B establishes the hold? |
| B touches but never holds | Clarify underside palm support and visible finger contact, without adding a second action | Can you inspect a supporting hold, rather than one fingertip brushing the cube? |
| Hands fuse at the meeting point | Use a replacement input with more space between fingers and a clearer object outline | Do A and B remain distinguishable during contact? |
| The cube jumps or duplicates | Reduce the travel distance and retain one continuous transfer | Can the same outline be followed through all four states? |
| A remains stuck to the cube at the end | State a clear final gap between A and the cube | Does B keep the object after A withdraws? |

If a result still hides the shared contact, record that limitation. This exercise does not prove a generally solved grasping capability or justify a claim about physical realism.

## 06. Save an honest review record

Keep the exact uploaded input, prompt, settings accepted by the generator and blank/filled review sheet together. If you generated a result, add its real task reference and observed findings; the supplied package contains no task ID, generated video or successful model-run receipt.

For comparison, [placing an object on a table](https://minimaxh3.studio/templates/hand-place-object-on-table) transfers support to a fixed surface. This exercise uses two moving hands and an explicit shared-contact interval. [Two-person relative height](https://minimaxh3.studio/templates/two-person-relative-height) trains scale relationships, while the [prompt guide](https://minimaxh3.studio/guide/minimax-h3) covers general input roles. Use the first for fixed-surface contact, the second for scene proportions, and the guide when preparing input roles.