Quick answer
A practical MiniMax H3 prompt names six things: the subject, the action, the environment, the camera, the lighting or visual style, and any constraints. Put the action in time order. If the shot must begin or end on a specific image, use the frame workflow and describe the path between those frames. If images, video, or audio should control identity, motion, camera behavior, or sound, use the reference workflow and assign each file a clear job.
Start with one main action and one camera move. Generate, watch the entire clip, then change one variable. MiniMax does not promise that every negative phrase, seed, or camera term will work exactly as written. Treat prompt language as direction, not a guarantee.
Last verified: August 4, 2026. minimaxh3.tv is an independent service and is not owned, operated, authorized, or endorsed by MiniMax.
A practical prompt moves from subject and action to setting, camera, visual treatment, and constraints.
The MiniMax H3 prompt formula
Use this as a working formula:
Subject + Action + Environment + Camera + Lighting/Style + Constraints
| Part | What to write | Example |
|---|---|---|
| Subject | The main person, animal, object, or scene anchor | A red paper boat |
| Action | What changes, in the order it happens | Drifts forward, catches a current, turns toward the bridge |
| Environment | Place, weather, background activity, and important objects | A shallow city canal after rain |
| Camera | Shot size, direction, movement, and speed | Low tracking shot, slow push in |
| Lighting or style | Visible treatment rather than vague quality words | Overcast daylight, muted documentary color |
| Constraints | Details that should remain stable or actions to avoid | Keep the boat red; no people enter the water |
The official H3 prompt writing guide also documents a more structured form for advanced timelines. It separates the main audiovisual timeline into integrated_multimodal_description, overall_soundscape, and non_diegetic_music. Use that structure when you need timed shots, dialogue, diegetic sound, or a score. A short browser prompt does not need all three fields.
Text to video prompt template
Text mode has no source image, so the prompt must establish the whole shot.
[Visual style and shot size]. [Subject] begins in [environment].
[Action 1], then [Action 2]. The camera [one camera movement] at [speed].
[Lighting and visible atmosphere]. End with [clear final state].
Keep [identity, object, color, or composition constraint] stable.
Example:
Live-action macro film. A red paper boat floats through a shallow city canal after rain. It drifts forward, catches a current, and turns toward a stone bridge. The camera tracks beside it at slow speed, close to the water. Overcast daylight, muted color, small ripples and wet reflections. End with the boat entering the bridge shadow. Keep the boat red and intact.
Image to video motion template
In the Start and End Frames workflow, the image already defines the subject, layout, color, and opening composition. Spend the prompt on motion and preservation.
Use the uploaded start frame as the opening composition.
[Subject or object] begins by [small first motion], then [continuous action].
The camera [one movement] while [background or lighting response].
Preserve [identity, clothing, shape, layout, or product details].
If an end frame is supplied, reach its pose and composition naturally at the end.
Avoid repeating every visible detail. The official guide recommends a continuous path from the first frame toward the last frame, especially when both are supplied.
Reference workflow template
Reference mode accepts ordered image, video, and audio inputs. Assign each one a role instead of writing "use all references."
Use Image 1 for the subject's identity and clothing.
Use Video 2 only for the camera path and movement pace.
Use Audio 3 for rhythm and timing.
Place the subject in [environment]. The subject [ordered action].
The camera follows the movement from [opening state] to [ending state].
Preserve [identity and object constraints]. Do not copy unrelated details from the references.
Text mode builds the shot from words. Frame mode describes a path from a fixed image. Reference mode assigns a job to each source asset.
Ten verifiable prompt examples
The examples below have a public source. Only the alpine lake example is a verified minimaxh3.tv prompt and output pair. The other nine are adapted from current MiniMax documentation. They are starting points, not claims about guaranteed results on this site.
| # | Mode and prompt | What it demonstrates | Evidence |
|---|---|---|---|
| 1 | Text: A cinematic sunrise over a calm alpine lake, soft mist drifting above the water, slow camera push forward. |
Subject, atmosphere, visible motion, and one camera move | Verified site prompt and output below |
| 2 | Text: A space-opera captain watches the last fleet jump away through an observation window as the bridge shakes. Slow push toward her reflection. End on the empty stars. |
Event order and ending state | Adapted from the official V2 API example |
| 3 | First frame: Preserve the dancer and studio from the uploaded image. She begins a contemporary turn, crosses the floor, and settles into a balanced pose. Slow lateral tracking shot. |
Animate a fixed opening without restating it | Adapted from the official video guide |
| 4 | First and last frame: Begin from the child's pose in the first image. Show a continuous passage of time through posture, clothing, and lighting, then arrive at the adult pose in the last image. |
A motion path between two anchors | Adapted from the official video guide |
| 5 | Reference: Use Image 1 for the model's face and beret. Keep the overcast cobbled alley. She walks forward, adjusts the beret, and smiles as the camera tracks backward. |
Clear reference roles | Adapted from the official reference example |
| 6 | Text with sound: Before sunrise, a baker opens wooden shutters and places a hot loaf on the counter. Slow push in. Street ambience, tray clinks, and one doorbell. Sparse acoustic guitar. |
Visual timeline, soundscape, and score | Adapted from the official prompt writing guide |
| 7 | First frame with dialogue: Preserve the woman, train seat, folded letter, and rainy window. The camera trucks right slowly as she looks toward the city lights and says one short line. |
Identity preservation plus a controlled speaking beat | Adapted from the official prompt writing guide |
| 8 | First and last frame: A cyclist releases the bicycle handle, raises a closed umbrella, opens it, and settles into the exact end-frame pose. Use one slow pull out. |
Observable intermediate changes | Adapted from the official prompt writing guide |
| 9 | Last frame: Begin with an intact glass near the table edge. A hand knocks it down. Track the fall and impact, then let the fragments settle into the supplied final composition. |
Work backward toward a fixed ending | Adapted from the official prompt writing guide |
| 10 | Text: A runner leaves a tunnel into daylight. Hold a static wide shot until she reaches the exit, then pan right once to reveal the open track. |
A single camera change tied to new information | Adapted from official camera-motion guidance |
This existing minimaxh3.tv clip used example 1:
More verified clips are available in the MiniMax H3 showcase. Use only examples that expose both the prompt and the result when you are studying prompt behavior.
Camera and motion vocabulary
The official guide defines camera motion by type, amplitude, and speed. Add amplitude or speed only when it changes the shot meaningfully.
| Term | What the camera does | Plain prompt phrase |
|---|---|---|
| Push in / pull out | Moves physically toward or away from the subject | The camera pushes in slowly toward the letter |
| Zoom in / zoom out | Changes focal length from a fixed position | The lens zooms out to reveal the room |
| Pan left / right | Rotates horizontally from one position | The camera pans right to the doorway |
| Truck left / right | Moves sideways | The camera trucks left beside the cyclist |
| Tilt up / down | Rotates vertically | The camera tilts up from the shoes to the face |
| Arc shot | Moves around the subject | The camera arcs halfway around the sculpture |
| Tracking shot | Follows a moving subject | A low tracking shot follows the paper boat |
| Static shot | Does not move | Hold a static wide shot as the runner exits |
| POV | Uses the subject's viewpoint | POV from the passenger seat |
Do not stack camera words at the end like tags. Tie the move to an action or reveal. One deliberate move is easier to evaluate than four competing moves in a short clip.
Common prompt failures and single-variable fixes
| Symptom | Likely cause | Change one thing | Evidence or test status |
|---|---|---|---|
| The shot looks static | The prompt describes appearance but no change over time | Add one visible action with a start and end | Supported by official timeline guidance and the verified lake example |
| The subject changes identity | Too many new appearance details conflict with a frame or reference | Remove new appearance details and state what must remain stable | Official keyframe and reference guidance |
| The camera feels random | Several moves compete in one short shot | Keep one camera move tied to one reveal | Official camera-motion guidance |
| The ending is abrupt | The prompt has actions but no final state | Add one clear ending pose, location, or composition | Practical revision; not a measured success-rate claim |
| The frame transition jumps | The prompt repeats two static images without a path | Describe the intermediate changes between them | Official first-and-last-frame guidance |
| References blend together | Each file has no assigned role | Name each numbered asset and its job | Official reference workflow |
| Dialogue changes | The line is paraphrased inside a long scene description | Put the exact line in a short, isolated speaking instruction | Official dialogue guidance |
| A negative phrase has no effect | The model may not treat natural-language negatives as hard controls | Rewrite the desired positive state and remove one conflict | Prompt revision to test; no guaranteed negative-prompt control |
Prompt checklist
- Name one main subject and one main action.
- Put actions in chronological order.
- State the environment only where it affects the shot.
- Use one primary camera move for the first attempt.
- Describe visible lighting or style instead of saying "high quality."
- Preserve identity, color, layout, or product details explicitly when needed.
- Assign a job to every reference file.
- Add a clear ending state.
- Check the full prompt for conflicting directions.
- After a result, change one variable before generating again.
FAQ
How long should a MiniMax H3 prompt be?
Long enough to define the shot, but short enough that the action order stays clear. A simple shot may need three or four sentences. Timed dialogue, several cuts, or multimodal references need more structure. The official API accepts up to 7,000 characters, while the current minimaxh3.tv prompt box has a smaller limit. Follow the limit shown in the interface you use.
Does MiniMax H3 support negative prompts?
The current public site uses natural-language constraints inside the main prompt. Do not assume a separate negative-prompt field or guaranteed negative behavior. Describe the desired positive state and remove conflicting instructions.
Can I specify a seed?
The current minimaxh3.tv generator does not expose a seed control. Do not promise reproducibility from a seed that the interface does not provide.
Should I use camera terms in brackets?
MiniMax documentation includes camera instructions such as pan, zoom, and static. Plain natural-language camera actions also appear in the official H3 prompt guide. Use the form accepted by your current interface, and do not assume a term forces an exact physical result.
What should an image-to-video prompt describe?
Describe motion, camera travel, the ending state, and the details that must remain unchanged. The source image already supplies the opening composition.
How do I keep references from mixing?
Number the files and give each one a job. For example, use one image for identity, one video for motion, and one audio file for timing. Remove any reference that does not contribute to the intended shot.
Should I copy a long prompt exactly?
Copy the structure, then replace the subject, action, environment, and constraints. A prompt tied to another image or reference set may fail when copied without those inputs.
Sources and methodology
This guide was checked on August 4, 2026 against the current MiniMax Video Generation guide, the official Create Video Generation Task reference, the official MiniMax H3 video prompt writing guide, the public MiniMax H3 text-to-video generator, the showcase, and one existing public minimaxh3.tv prompt/output pair. No new paid video generation was used. No external video or community claim is included.
Ready to try a prompt? Open MiniMax H3 Text to Video. For the full browser workflow, read How to Use MiniMax H3.
