# MiniMax H3 dolly-zoom prompt: hold subject size while depth changes

Prepare one continuous shot in which a stationary foreground target keeps approximately the same image size while the background depth relationship changes. The aim is to recognize a coupled move, rather than to decorate a simple zoom with a cinematic name.

**Original practice input and untested method.** No model generation, completed edit or output review has been performed.

## 1. Read the original input and define the outcome

The clean S01 input shows one ochre pillar with a coral inset, standing on a teal plinth in a fictional corridor. Three nested rectangular frames and floor lines provide visible background references. The pillar is not a real product, and the drawing does not document an unseen adjoining room. Download the PNG for the first-frame slot; keep the SVG and landmark table beside your review sheet. The separate two-panel guide is an authored comparison sketch, not a supplied ending frame or a successful H3 result.

- [S01 Stationary foreground and corridor — PNG](/learning-assets/original-b-next-camera-20261022/dolly-zoom-subject-size-background-depth/S01-first-frame.png) · [Editable SVG](/learning-assets/original-b-next-camera-20261022/dolly-zoom-subject-size-background-depth/S01-first-frame.svg)

[Authored review guide, not an output](/learning-assets/original-b-next-camera-20261022/dolly-zoom-subject-size-background-depth/review-guide.svg)

## 2. Assign motion and preserve the named relationships

The target remains stationary in the scene. The requested camera retreat and counteracting zoom are the two parts of one effect: they should approximately retain the pillar height while changing the apparent spacing of the corridor behind it. Track the pillar top and plinth contact separately from the near and far frame edges. The drawing coordinates identify those features only; they are not camera keyframes or an undocumented numerical API. A full-frame crop enlarges the foreground and background together and cannot establish the intended depth change.

[Drawing landmarks and role definitions](/learning-assets/original-b-next-camera-20261022/dolly-zoom-subject-size-background-depth/scene-roles.json)

## 3. Prepare the manual single-shot task

Open one new image-to-video draft and use S01 only. Begin with a modest retreat and an opposite framing adjustment, not an orbit, a pillar rotation or a long walkthrough. Keep the whole plinth visible so its contact can be reviewed. Before submitting, read the camera and subject sentences independently: only the viewpoint and framing are assigned motion. Retain the exact input and wording with any future take. If the requested combination is not produced, that is an unresolved result; do not describe an attractive push-in as a completed dolly zoom.

Use **MiniMax H3 Base image-to-video**. In the documented controls, enable **Add end frame** to expose **First Frame**, upload only the current shot PNG there, and leave **Last Frame** empty. This does not use Fast, which requires paired frames. Recheck the actual controls and current credit estimate before manual submission. The download and draft button do not upload or pay for a generation. SVGs, the separate review guide and drawing-coordinate JSON stay local; they are not extra uploaded frames or special camera commands. No fixed duration, numeric lens control or exact ending is implied.

[Download the complete practice kit](/learning-assets/original-b-next-camera-20261022/dolly-zoom-subject-size-background-depth/practice-kit.zip)

## 4. Use the complete original prompt

**S01 — Stationary foreground and corridor (the only default draft)**

```text
Begin from this original corridor illustration. The ochre pillar, coral inset and teal plinth remain stationary and rigid. Make a small smooth camera retreat along the current viewing direction while adjusting the framing in the opposite sense so the pillar keeps approximately its opening image height. Let the visible corridor frame spacing and depth relationship change around that retained foreground size. Preserve the three existing frames, floor connections and plinth contact. Do not simply scale the whole picture, move the pillar, orbit, cut, stretch the walls or reveal a new room. Stop both camera and framing change together and hold the same target. This is a qualitative visual brief, not numerical lens control.
```

[Download this complete prompt](/learning-assets/original-b-next-camera-20261022/dolly-zoom-subject-size-background-depth/prompt-S01.en.txt)

This is one source shot. The default button carries only this prompt; the clean PNG must still be uploaded manually.

## 5. Inspect actual phases with the blank sheets

First watch the whole shot to find unintended cuts or subject drift. Then compare opening, intermediate and ending frames using the same pillar boundary each time. In the blank sheet record normalized pillar height, the three frame references and any uncertainty; do not copy the authored guide coordinates into a column intended for observed video. Approximate subject-size stability and a visible background-depth change are both needed. A matching final pillar size alone could conceal an ordinary zoom followed by an abrupt correction.

1. One stationary pillar and plinth remain identifiable without stretching or a ground slip.
2. Pillar image height stays approximately near its opening value while background frame relationships visibly change.
3. The change is continuous, with no cut or whole-image scale-only substitution.
4. The ending settles without cropping the target or inventing a new room.

[Blank acceptance sheet](/learning-assets/original-b-next-camera-20261022/dolly-zoom-subject-size-background-depth/acceptance.blank.csv) · [Blank source-shot log](/learning-assets/original-b-next-camera-20261022/dolly-zoom-subject-size-background-depth/shot-log.blank.csv)

## 6. Change one failed relationship at a time

If everything grows together, remove words asking for a close-up and restate the counteracting camera/framing relationship. If the pillar slides across the floor, strengthen only its fixed plinth contact before changing the camera range. If corridor frames bend or multiply, reduce the move and retain the originally visible space. If the target is cropped, choose a smaller range rather than inventing a second target at the ending. Keep each attempted change separate; neither a source replacement nor a second model choice isolates the original camera instruction.

## 7. Keep the materials and state the limits

A future accepted take ends with the same pillar settled on the same plinth, roughly the opening image height, and a distinguishable new background-depth relationship. Save the original PNG, exact prompt, untrimmed result and filled observations together. Leave result fields blank until actual frames exist.

The existing tracking-walk page follows a moving person, while transparent-object parallax changes the view through a fixed object. This exercise instead couples camera distance and framing around a stationary foreground size anchor. It does not guarantee lens control, a calibrated trajectory, perfect perspective, factual hidden surfaces or successful generation.

- [Tracking walk with constant screen direction](/templates/tracking-walk-constant-direction)
- [Background parallax through a transparent object](/templates/transparent-object-background-parallax)