First-and-last-frame generation gives an AI video model two visual anchors: how the shot begins and where it must finish. It is especially useful for product reveals, environmental changes, logo-free transitions, character movement, and shots that need a predictable edit point.
The two images are not a complete storyboard. The model still has to invent the motion between them. Good results therefore depend on designing compatible anchors and explaining the transition rather than merely describing the final image.
When to use two anchor frames
Use a first-and-last-frame workflow when:
- the final composition matters as much as the opening;
- an object must move to a known position;
- the camera must end on a specific framing;
- a material, season, time, or lighting state changes gradually;
- the generated clip must connect two edited shots;
- text-only direction repeatedly produces an uncontrolled ending.
Use a single first frame when the destination can remain open-ended. Adding a weak or incompatible last frame can reduce motion quality rather than improve it.
Design compatible anchors
The model has limited time to reconcile the two frames. Reduce unnecessary differences.
Keep identity stable
The same character or product should have matching proportions, clothing, major colors, and defining details. If the last frame quietly changes the jacket, bottle cap, hairstyle, or logo, the transition may morph.
Respect spatial continuity
Map major objects in both frames. A table on the right should not jump to the left unless the prompt calls for a camera move that explains the change.
Use a plausible camera path
Ask whether a real camera could move from frame A to frame B in the available time. A front close-up cannot become an aerial city view in four seconds without a rapid, complex reveal.
Match aspect ratio and safe area
Prepare both anchors at the same final aspect ratio. Keep important subjects away from edges that may be cropped during processing.
Control visual change
Decide which categories may change:
| Layer | Keep stable | Allow to change |
|---|---|---|
| Subject | identity, proportions | pose, expression |
| Product | geometry, label | position, rotation |
| Camera | axis, horizon | distance, modest angle |
| Environment | layout | weather, light, small motion |
| Style | medium, palette family | contrast, time-of-day accents |
Write the transition, not the endpoints
The images already show the endpoints. Use the prompt to describe the path:
The shot begins exactly from the first frame. The camera makes one slow push
forward while the subject walks two steps toward the window. Curtains move gently
in the breeze and afternoon light becomes warmer. Preserve the subject's face,
clothes, room layout, and camera axis. Finish exactly in the last-frame composition,
with the subject still and the camera settled. One continuous shot, no cuts,
no extra people, no object morphing.The strongest transition prompts answer four questions:
- What moves?
- What stays fixed?
- How does the camera move?
- How should the shot settle into the final frame?
Use a motion budget
Every shot has a finite motion budget. Spend it on the change that tells the story.
For a short clip, choose one major transition:
- subject crosses the room;
- product rotates and camera pushes in;
- day becomes night while the camera remains locked;
- closed package opens to reveal contents;
- camera rises from detail to overhead view.
If the subject changes clothes while the building transforms and the camera completes a full orbit, the model must solve three separate continuity problems at once.
Workflow: from frames to approved clip
1. Define the narrative beat
Write one sentence: “A clean bottle on a studio plinth becomes surrounded by fine mist as the camera moves into the label.” If the sentence needs multiple “and then” clauses, split it into several clips.
2. Prepare the first frame
Make it a clean starting state. Remove artifacts, ambiguous hands, accidental text, or unnecessary moving objects.
3. Create the last frame from the first
Whenever possible, edit or generate the last frame using the first as a reference. This preserves subject geometry and scene layout better than creating both independently.
4. Compare the anchors side by side
Overlay them or flip rapidly between them. Mark changes in subject scale, horizon, object position, lighting, and background perspective.
5. Draft the transition prompt
Describe motion in chronological order. Add preservation constraints after the motion, then define the final settle.
6. Generate low-cost tests
Test the shortest useful duration or preview setting first. Review motion logic, not surface polish.
7. Revise the weakest layer
If identity morphs, simplify subject motion. If the camera jumps, reduce the difference between frames. If the clip stalls, add one clear active verb and timing cue.
Troubleshooting
The video freezes near the first frame
The motion may be underspecified or the anchors may be too similar. Add an explicit action with distance and timing: “During seconds one through four, rotate the bottle 30 degrees clockwise.”
The subject morphs halfway through
Check whether the anchors disagree on identity, clothing, product details, or viewing angle. Regenerate the last frame from the approved first frame and name the constants in the prompt.
The camera teleports
Use one physically plausible camera path and reduce the endpoint difference. Explicitly exclude cuts, whip pans, and sudden zooms.
The final frame is reached too early
Add temporal pacing: “Complete only half the movement by the midpoint; settle into the final position during the last second.”
Unwanted objects appear
Remove unexplained props from both anchors. State the exact subject count and exclude additions.
Official capability note
MiniMax's official documentation describes a supported mode that accepts a first-frame image, a last-frame image, and a prompt explaining how the scene develops. Parameters and supported models can change, so confirm the current limits in the MiniMax video generation documentation before production.
Final review
- First and last frames use the same aspect ratio.
- Identity, product geometry, and scene layout agree.
- One main action fits the available duration.
- Camera movement is physically plausible.
- Prompt names both moving and fixed elements.
- The last second settles into an editor-friendly frame.
- No new text, logos, people, or accessories appear.
Create compatible anchors in the AI Image Generator, then animate them with a supported workflow in the AI Video Generator.
