Practical MiniMax H3 guides
MiniMax H3 Image to Video: First & Last Frame Guide
Start with the picture you want the viewer to see, then describe what happens next. This MiniMax H3 image-to-video guide focuses on preparing a useful still, connecting two endpoint frames, and reviewing the movement between them.
Original editorial guidance with untested prompt drafts. Generation on this site is coming soon.
Browse all guidesOne image or first and last frames?
MiniMax’s API guide documents first-frame, last-frame and paired-frame inputs for H3. Use an opening image when the initial composition matters; provide an ending image when the destination matters too. Confirm which upload roles your chosen interface exposes.
For a practical first attempt, choose a small change you can describe precisely: a desk lamp switching on, a curtain moving, or a hand sliding a card onto a table. A single image leaves room for that action to develop. Two frames are useful when a client has already approved both the opening and closing composition.
If you want only the person or product from a photo in an entirely different scene, consider reference-to-video instead. Decide whether you are preserving the frame itself or borrowing an element from it before preparing your assets.
Prepare an image that leaves room for motion
Inspect your source at the size the finished video will be viewed. Is the important object recognizable? Does the crop cut through a hand, handle or moving part? If a drawer must open toward the camera, leave space for its new position. A beautiful still can be an awkward starting point for an action.
Choose the delivery crop before writing the motion. For a vertical clip, check that both the opening subject and its intended destination fit inside the narrow frame. Avoid asking the camera to repair an unsuitable composition while the subject performs a complicated action.
Write a short preservation list using details you can actually inspect: the mug’s handle stays on the right, the lamp remains on the left, and the table edge stays horizontal. Keep decorative style requests secondary to those requirements.
An image-to-video motion prompt you can adapt
The following is an original plain-language prompt draft for a first-frame interface, not an official structured rewrite or a tested result. Upload a matching image through the interface; naming a picture in text does not attach it. Set duration and output options using the provider’s controls.
This brief separates the moving part from the stationary scene. That makes the result easier to assess: the light changes, but the room should not become a new location.
Original prompt draft — adapt to your inputs; not a tested generation.
Use the uploaded image as the opening frame: a brass desk lamp beside a closed notebook on a walnut table. The camera stays still. A hand enters from the right, presses the lamp switch once, then leaves the frame. Warm light spreads across the notebook while the lamp, table and background retain their positions and shapes. Finish with the lamp illuminated and the hand out of view. A quiet switch click accompanies the press, with low room ambience and no music.
Connect the first and last frame with a plausible action
Prepare two images of the same desk: one with the lamp off and one with it on. Match the crop, object positions and viewpoint. If the second image also changes the table, notebook and wall, your brief contains several transformations instead of a lighting transition.
Write down the intermediate event. In this example, the hand reaches the switch, presses it, and withdraws. If the two endpoints show different camera angles, decide whether the camera should travel between them or whether you actually need two separate shots assembled in an editor.
The official prompt guide describes endpoint alignment and a continuous path between the images. For a provider’s structured workflow, follow its documented format; the planning draft below expresses the creative intent.
Original prompt draft — adapt to your inputs; not a tested generation.
The first uploaded frame is the unlit desk; the last uploaded frame is the same desk with the lamp on. Keep a single fixed view. A hand reaches the switch, presses it, and withdraws. Let the light settle gradually until the scene reaches the final image. Keep the notebook closed and the lamp base stationary. End after the hand has left the composition.
Troubleshoot the first frame, movement and landing separately
Wrong opening: check the upload role and selected crop before rewriting the prompt. Distorted movement: simplify the action and inspect the contact point. Wrong ending: compare the endpoint images for differences that were never explained in the action.
Review the whole clip once at normal speed, then pause at contact and near the end. For the lamp example, judge the hand, switch, lamp silhouette and final illumination separately. A correct ending does not make an awkward middle acceptable.
Save the source files, exact prompt, selected model and settings with each attempt. Change one thing on the next run. If the remaining issue is a title or a precise end hold, consider finishing that detail in an editor instead of asking generation to redo the entire action.
Frequently asked questions
Can MiniMax H3 animate a single photo?
Yes. The documented first-frame route accepts an image plus a prompt. Use an interface that exposes that mode, attach the image, and describe the subsequent action.
Do first and last frames guarantee a smooth transition?
Treat smoothness as something to review. Compatible compositions and an explicit intermediate action make a clearer brief, but you still need to inspect the generated movement.
Can I generate image-to-video on this website?
Generation here is coming soon. You can read the guides and prepare prompts now, then use an available H3 service that supports the required input mode.
Sources and editorial method
Official documentation supports the input roles and prompt conventions discussed here. Preparation tips, examples and review checklists are our editorial recommendations. We have not run comparative tests or generated the example prompts.
Sources checked September 13, 2026. Follow your selected provider’s current interface and requirements. Read our editorial method.