MiniMax H3 Image to Video: Frame Control

The short answer

MiniMax H3 image to video turns a still into a continuous shot. Upload a first frame, describe what should change, and choose a duration and aspect ratio. If the ending matters, add a last frame and the model will aim for that image instead of inventing its own ending.

The same model also handles text-to-video, but the workflow is different. With an image, the subject and light are already decided. Your prompt should spend its words on motion, timing and sound, not on re-describing the frame.

What the workflow gives you

The studio exposes the controls that matter for a controlled shot:

• First frame — the image the clip starts from. JPG or PNG, up to 10 MB.
• Last frame — optional. Use it when the clip must land on a specific composition.
• Duration — 5, 10 or 15 seconds.
• Resolution — 768P for drafts, 2K for final output.
• Aspect ratio — 21:9, 16:9, 4:3, 1:1, 3:4 or 9:16.
• Native audio — stereo sound is generated with the picture.

A five-second 2K clip costs 25 credits; the same clip at 768P costs 15. A fifteen-second 2K clip costs 75 credits. The pricing page shows the full credit table.

When to use image-to-video instead of text-to-video

Use image-to-video when the frame already carries the idea: a product photo, a character design, a location scouting still, a client-approved composition. The model's job is to move it, not to redesign it.

Use text-to-video when you want the model to invent the subject, the environment and the camera in one pass. That is faster for concepts and b-roll, but less precise when a brand asset or a specific person must stay recognizable.

A practical rule: if you would reject a generation because the product shape changed, start from an image.

The five-step workflow

1. Choose the frame. Pick an image with a clear subject and a little room for motion. A frame that is already cropped tightly leaves the camera nowhere to go.
2. Decide the ending. If the shot needs to land on a specific composition, prepare a last frame that matches the first frame's subject, lighting and style.
3. Write the motion prompt. Describe the change: camera move, subject action, light shift and sound. Keep it to one beat.
4. Pick duration and ratio. Five seconds for a loop or a detail; ten or fifteen for a reveal or a longer camera move. Match the ratio to the destination.
5. Review and iterate. Watch the clip with sound. If the motion is right but the ending drifts, add a last frame. If the motion is wrong, rewrite the verb before you change the model settings.

First frame only vs first and last frame

First frame only gives the model freedom to resolve the ending. It is the better choice for ambient shots, portraits and anything where a natural conclusion is acceptable.

First and last frame constrains the journey. It is the better choice for product reveals, before-and-after transitions, logo landings and shots that must cut cleanly into an edit.

With a last frame, describe how the subject gets there. "The camera pushes in while the bottle rotates a quarter turn" is a path. "A bottle" is not.

Write the prompt around change

An image-to-video prompt has four useful slots:

• Camera — push-in, pull-back, orbit, handheld follow, static.
• Subject motion — a head turn, fabric lifting, a hand entering the frame, a product rotating.
• Light change — a cloud passing, a neon sign flickering, the sun moving behind a building.
• Sound — room tone, rain, footsteps, a low score.

Example: "Slow push-in on the perfume bottle, the label catches the light as the camera moves, silk fabric shifts in the background, soft room tone and a single glass chime."

Notice that the prompt does not explain what the bottle looks like. The first frame already did that.

Three use cases

Product reveal. Start on a detail, pull back to the full product, and land on a clean packshot. Use a last frame if the final composition must match a template. Describe the camera move and the surface sound; keep the product motion small so the shape stays stable.

Portrait with a subtle turn. A still portrait becomes a short clip when the subject turns toward the camera, the curtain moves, or the light shifts. Large gestures are harder to keep coherent; a small, motivated movement usually reads better.

Transition shot. Use a first and last frame to move between two compositions. The prompt should describe the transition itself: a wipe of light, a camera move through a doorway, a match cut on a color.

Settings that change the result

Duration changes how much story fits. A five-second clip can hold one beat; a fifteen-second clip can hold a setup and a payoff. Longer is not automatically better, especially when the motion is subtle.

Resolution changes detail, not composition. Draft at 768P, then render the approved shot at 2K. This is the simplest way to keep credit spend under control without giving up final quality.

Aspect ratio changes what the frame can hold. A 21:9 shot has room for a lateral move; a 9:16 shot needs a vertical composition and a subject that stays in the center band. Choose the ratio before you write the prompt.

Common mistakes

• Describing the still. The image already contains the subject and light. Describe the change.
• Asking for a cut. H3 generates a continuous shot. Plan cuts in the edit.
• Too much motion in five seconds. A push-in, an orbit and a subject turn in one short clip usually produces drift.
• Mismatched frames. A last frame with different lighting or a different product angle forces the model to reconcile two unrelated images.
• Ignoring the first frame's composition. If the subject is at the edge of the frame, a push-in may crop it. Leave room in the source image.

FAQ

Can I use any image as a first frame?

Use an image you have the rights to use, with a clear subject and enough resolution for the output. JPG and PNG up to 10 MB are supported. Avoid images with heavy motion blur or a subject that is already cropped at the edge.

Do I need a last frame?

No. Add one only when the ending must be exact. For ambient shots and portraits, a first frame plus a clear motion prompt is usually enough.

How much does an image-to-video clip cost?

A five-second 2K clip costs 25 credits, a ten-second clip costs 50 and a fifteen-second clip costs 75. At 768P the same durations cost 15, 30 and 45 credits. Failed renders are refunded automatically.

Can I generate sound from an image?

Yes. Native stereo audio is generated with the picture. Describe the ambience or effect you want in the prompt, then review the result. You can also replace the audio in your editor if the scene needs a specific track.

Conclusion

MiniMax H3 image-to-video is a control workflow: you supply the frame, the model supplies the motion and sound. Keep the prompt focused on change, use a last frame when the ending matters, and draft at 768P before you commit to 2K.

Open the MiniMax H3 video generator to try a first frame, or read the MiniMax H3 prompt guide for a reusable shot-brief structure.

Sources

• MiniMax H3 launch post — official model announcement, 31 July 2026.
• Hailuo AI MiniMax H3 page — official product page with resolution, duration and audio details.