Prompt MiniMax for image-to-video

Last updated: October 12, 2026

MiniMax H3 animates an approved still into a clip that acts on beats (a look, a gesture, a reaction) with sound generated in the same pass. Pick it when the frame is signed off and the character has to perform: a spell charge, an idle, a reaction shot.

Let the agent write it

You don't have to write any of this yourself. The Layer agent knows how to prompt each model it runs, from the same guidance this page is built on. Attach the still in chat, name the model and say what should happen, for example: Animate this with MiniMax H3 Max: the mage looks up and raises his staff as the flame swells, slow push-in, 5 seconds. The agent writes the prompt, shows you its plan, and runs it when you approve. Use the rest of this guide when you want to write or adjust the prompt yourself, and see Write a brief for the agent for how to brief it.

When to use it

Use MiniMax when the clip is a performance: a character who turns, looks, lifts, reacts, in an order you set. It lands a sequence of beats in one take and keeps the character's face, costume and colours from the first frame. It also takes a last frame, for a move between two approved poses.

It drifts when the camera does several things at once: the background starts to change into something the frame never had. Keep to one move per clip. To generate a clip with no image at all, see Create a video.

Versions on Layer

MiniMax H3

The full model, at 768p, 1440p or 2160p. Clips run 5 to 15 seconds, with a first and a last frame, and sound in every clip. Use it for the final render.

MiniMax H3 Max

A version of H3 tuned to follow the prompt more closely, at 480p, 768p or 1080p. It adds a Camera move control for drawing the camera's path around the subject. This guide's examples were made with it, at 480p.

MiniMax H3 Max Turbo

The same inputs as H3 Max without Camera move, and quicker to return. Use it to find the motion before you render the keeper.

How to write the prompt

Style, then one camera move written as a sentence, then the action beat by beat, then what must stay and what must not appear:

Stylised painterly game art. The camera pushes in with small amplitude at slow speed toward the mage. He lifts his head and looks out toward the valley, then slowly raises the staff with both hands as the ember on its tip swells into a ball of fire, his cloak stirring in the rising heat. He holds the pose as the flame crackles. Keep his face, red hooded cloak and staff exactly as in the first frame. Do not add subtitles or any other text.

MiniMax's own guide builds an image-to-video prompt as first-frame anchor, action onset, development, result. The frame already fixes who is in the shot and how it looks, so the words go on what happens next. That prompt is the one in the clip above.

What works

Camera moves as sentences

Write the move with its type, amplitude and speed: "The camera pushes in with small amplitude at slow speed toward the mage". The types MiniMax documents are push in, pull out, zoom in, zoom out, pan, truck, tilt, pedestal, arc shot, tracking shot, static shot, shake and roll. We ran this sentence against the bracket command [Push in] on the same frame and seed. Both pushed in and landed every beat, but the bracket version swung much further, into a low-angle close shot, and stretched the staff into a long pole. The sentence kept the move measured. Even "small amplitude" travels a fair way, so ask for a static shot when the framing must hold.

A static shot for idles

For an idle or a loop on a menu, say "The camera holds a static shot; the framing does not change." In our test the figure, ledge and castle stayed in place for the whole clip while the cloak, hood and ember moved and the mage turned his head.

Action in beats

Give the beats in order, each with what triggers the next: "lifts his head and looks out toward the valley, then slowly raises the staff with both hands as the ember swells into a ball of fire". Then say how it ends ("he holds the pose"), so the clip has somewhere to land.

Open with the style

Start with the medium of the frame: "Stylised painterly game art", "2D-animated", "3D CG", "claymation". MiniMax's guide puts the style word first in every shot. Use concrete visual and sound details over words like "cinematic" or "beautiful".

Sound in the prompt

Every clip has sound. Name what should be heard ("soft wind and the quiet crackle of the ember") and say "no music" when you want none. Ambient-only sound can come out quiet: our idle's sound sat far below the other clips', so plan to level it in the video editor. For dialogue, put the line in quotes and say who speaks it; for a voiceover, write that the character "says in an off-screen voiceover" and that their lips stay closed.

MiniMax's labelled layout

MiniMax's guide documents a layout with a first-frame line and three labelled fields. We ran the spell charge this way on the same frame and seed:

For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.

integrated_multimodal_description: [Shot 1] Stylised painterly game art, the young fire mage shown in <Picture 1> stands on the mossy ledge, preserving his face, red hooded cloak, staff and the castle and valley behind him. The camera pushes in with small amplitude at slow speed toward the mage. He lifts his head and looks out toward the valley, then slowly raises the staff with both hands as the ember on its tip swells into a ball of fire, his cloak stirring in the rising heat. He holds the pose as the flame crackles. Do not add subtitles or any other text.

overall_soundscape: A soft wind moves across the ledge, the ember hisses and then roars into a crackling fireball, and cloth rustles as he raises the staff.

non_diegetic_music: N/A

Both versions kept the mage and the castle on-model and ended on a big flame. The labelled run thrust the staff forward where the prose run raised it overhead as asked; neither was better overall. With Prompt enhancement on, MiniMax rewrites prose into this layout before it generates, so you don't need it. It earns its place when you want to set the sound (overall_soundscape) and the music (non_diegetic_music, or N/A for none) separately from the picture.

Several shots in one clip

Write later shots with a time: "[Shot 2] At 00:03.500, the camera cuts to a close-up of the staff." A cut should show something new; for a small change of framing, use a camera move instead.

First and last frames

With both frames set, describe the path between them: how the pose changes, what the hands do, how the framing moves. Don't describe the two images again, and keep it to one shot.

What to avoid

Chaining camera moves

We asked for a pan, then a push-in, then a tilt up in one prompt. The camera travelled sideways and replaced the castle and valley with mountain peaks the frame never had, and the spell barely flared. In another published test, the same kind of prompt split into separate shots. One move per clip; cut between clips for more.

Expecting a negative prompt

MiniMax has no negative prompt. Write exclusions as sentences in the prompt: "Do not add subtitles or any other text." Text that should appear, such as a sign or a HUD label, goes in double quotes, word for word.

Leaving the move to the rewriter

Prompt enhancement fills in what the prompt leaves out, including how far and how fast the camera moves. State the amplitude and speed yourself so the choice stays yours. When the prompt is already complete and you want it read as written, set Prompt enhancement to Off.

Settings that matter

First frame and Last frame pin where the clip starts and ends. Duration runs from 5 to 15 seconds; find the motion at 5, then render the length you need. Aspect ratio sets the shape and resolution; match the shape of your still. Under Advanced settings, Prompt enhancement controls the rewrite (Auto by default), and a fixed Seed lets you change one word and compare. On MiniMax H3 Max, Camera move draws the camera's path instead of writing it. Above the prompt, Camera, Staging, Look and Performance take reference sets for each.

Prompts to start from

Spell charge

Stylised painterly game art. The camera pushes in with small amplitude at slow speed toward the mage. He lifts his head and looks out toward the valley, then slowly raises the staff with both hands as the ember on its tip swells into a ball of fire, his cloak stirring in the rising heat. He holds the pose as the flame crackles. Keep his face, red hooded cloak and staff exactly as in the first frame. Do not add subtitles or any other text.

Character idle

Stylised painterly game art. The camera holds a static shot; the framing does not change. The mage breathes slowly, his cloak and the edge of his hood stirring in a light breeze, and the ember on his staff flickers, shedding small sparks that drift upward. He turns his head slightly toward the valley, then back. Mist drifts through the valley below. Keep his face, red hooded cloak and staff exactly as in the first frame. Do not add subtitles or any other text. Sound: soft wind and the quiet crackle of the ember, no music.

Spell charge with sound fields

For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.

integrated_multimodal_description: [Shot 1] Stylised painterly game art, the young fire mage shown in <Picture 1> stands on the mossy ledge, preserving his face, red hooded cloak, staff and the castle and valley behind him. The camera pushes in with small amplitude at slow speed toward the mage. He lifts his head and looks out toward the valley, then slowly raises the staff with both hands as the ember on its tip swells into a ball of fire, his cloak stirring in the rising heat. He holds the pose as the flame crackles. Do not add subtitles or any other text.

overall_soundscape: A soft wind moves across the ledge, the ember hisses and then roars into a crackling fireball, and cloth rustles as he raises the staff.

non_diegetic_music: N/A

Use it in Layer

  1. Add the approved still to a session's board.

  2. Click View on the image, then Turn into video. Create video opens in the side panel with the image as its First frame.

  3. Click the model name at the top of the panel and choose MiniMax H3 Max. The first frame stays in place.

  4. In the prompt box, write the style, one camera move as a sentence, the action beat by beat, and the constraints.

  5. Leave Duration at 5 seconds while you find the motion.

  6. Click Create video. The cost shows on the button before you click.

  7. When the clip lands on the board, click View to open it in the viewer and play it. Then change one thing in the prompt at a time and run it again.

Learn more

  • MiniMax H3 video prompt writing guide (MiniMax): the prompt layout H3 is trained on, the first-frame instruction, camera moves written as type, amplitude and speed, and the sound and music fields.
  • MiniMax-H3 model card (MiniMax): what H3 does, and the prompt rewriter that fills in what a prompt leaves out.
  • Video generation (MiniMax API platform): prompt length, image and duration limits, and the short bracketed camera hints MiniMax still accepts.
  • H3 Max image to video API (fal): the H3 Max inputs, including the prompt expansion modes behind Prompt enhancement.
  • MiniMax H3 prompt guide and real tests (Fuser): tested runs showing what the rewriter adds, how chained camera moves can split into shots, and how quiet the native audio can come out.

Related