For MiniMax H3 text-to-video-with-audio (T2VA), start with three fields—integrated_multimodal_description, overall_soundscape, and non_diegetic_music—then describe what happens in playback order. Put camera movement into an ordinary action sentence, add timestamps to later shots (not the first), and keep moment-specific sound, continuous ambience, and audience-only music in their proper fields. The opening instruction changes for image-anchored modes. These are MiniMax’s documented prompt-writing recommendations, not guarantees about any generated result. Read MiniMax’s prompt guide.
A practical prompt formula
Choose the workflow first, then build the prompt around it. In T2VA, the prompt begins directly with the three core fields. In image-conditioned workflows, put the appropriate instruction for the supplied frame or references before those fields. Within the timeline, describe the visible scene and actions in the order they should unfold, including any dialogue or sound tied to a particular moment.
As an Amazon Associate I earn from qualifying purchases.
A useful T2VA skeleton is:
integrated_multimodal_description: [visual style, scene, subjects, actions, shots, and moment-specific dialogue or sound in playback order]
overall_soundscape: [continuous ambience, physical action sounds, and non-verbal human sounds]
non_diegetic_music: [audience-only score, or N/A]These field names and their roles come from MiniMax’s official guide. Treat the structure as a way to make instructions legible, not as a promise that every detail will appear exactly as written.
Free tools Windows power users keep installed
One-click scans. No signup required.
What belongs in each audio-video field?
integrated_multimodal_description: the timeline
Use this field for the visual description and for events synchronized to a particular point in the video: actions, shots, speakers, dialogue, singing, diegetic music, and other in-scene sounds. “Diegetic” means the sound belongs to the world of the scene—for example, a radio playing in a room or a door slamming on screen. Keep the timeline in playback order so a reader can associate each event with the action or shot that contains it.
#1 Best Overall
overall_soundscape: the ongoing sound bed
Describe continuous or recurring ambience and physical or non-verbal sounds here: rain, traffic, footsteps, fabric rustle, breathing, impacts, or room tone. MiniMax recommends one to four English sentences in a continuous paragraph. Do not repeat dialogue, singing, or diegetic music already specified in the timeline. Use N/A only when you want a completely silent video.
non_diegetic_music: the audience-only score
Use this field for music heard by viewers but not by characters in the scene. Specify useful musical qualities such as instrumentation, speed, rhythm, and changes in intensity. Use N/A when you do not want an audience-only score; that is distinct from requesting a completely silent video.
How to describe camera movement and timing
Write movement as an action in the shot
MiniMax’s guide breaks a camera-motion instruction into motion type, amplitude (the range of compositional change), and speed. It lists examples including zoom, push or pull, pan, truck, tilt, pedestal, arc, tracking, static framing, shake, POV, and clockwise or counterclockwise roll. Describe the movement naturally as part of the shot rather than stacking labels. The guide’s example is: “The camera pushes in with small amplitude at slow speed toward the folded letter in her hands.”
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Include amplitude or speed when they matter to the scene; the guide says medium amplitude and normal speed are usually left unstated. A static shot is also a meaningful choice when the framing should not move. For a small change in distance or angle, describe camera movement rather than adding a cut. A cut is most useful when the subject, space, state, viewpoint, or time changes enough to reveal new information.
Timestamp later shots, not the opening shot
Write shots in the order they play. The guide’s format leaves the first shot without a timestamp; number later shots sequentially and start each with a strictly increasing cut time inside the video’s duration. For example: [Shot 2] At 00:03.500, the camera cuts to ... The time marks when the new shot begins, not a duration to add to the previous shot.
Keep the requested timing plausible for the clip length selected in the interface. As of the MiniMax repository accessed in 2026, the company lists H3 video durations of 4–15 seconds; this is a MiniMax-published specification, not an independent test. MiniMax’s H3 repository is the source for the product specifications.
Choose the prompt opening for the H3 mode
The guide documents different anchoring instructions because the supplied image or references determine what is fixed. The four base modes below should not be treated as interchangeable templates.
Recommended Free Tools
| Mode | What is anchored | How to describe the timeline |
|---|---|---|
| T2VA | No supplied frame is fixed. | Begin with the three core fields and describe the scene and its development from the opening onward. |
| I2VA | The supplied image is the actual first frame. | Start with the first-frame instruction. Establish the image’s style, subjects, composition, and scene anchors, then describe how the scene develops forward. |
| FL2VA | The supplied first and last images anchor the opening and ending. | Start with the first/last-frame instruction and describe a plausible path between the two. The guide generally favors a single shot unless multiple shots are specified. |
| L2VA | The supplied image is the final frame. | Start with the last-frame instruction. Describe a plausible preceding state and the movement that converges on the supplied ending image. |
For a reference-based workflow, MiniMax’s repository also describes H3-Base-Ref2VA: it accepts text with image, video, and/or audio references. Its listed limits are up to 9 images, up to 3 video clips, and up to 3 audio clips; each video or audio clip may be 2–15 seconds, with 15 seconds total duration for each media type, and a maximum of 12 files across input types. Interfaces and limits can change, so check the official repository and the current workflow before relying on those figures.
Best Value
Dialogue and voiceover instructions
For a speaking or singing subject, MiniMax recommends giving each speaker a stable ID such as (S1) and reusing it across shots. Put the speaker’s identifying phrase, ID, action, and delivery outside the <d> block; put the language tag and the exact user-provided words inside it. Preserve supplied dialogue and punctuation verbatim rather than paraphrasing them.
For voiceover, use the phrase says in an off-screen voiceover. The guide says to state immediately after the dialogue block that the corresponding on-screen character’s lips remain closed. This helps distinguish narration from speech that should be visibly spoken by someone in the scene.
Illustrative prompt
This example demonstrates field separation, a camera action, and a timed cut; it is a prompt-writing illustration, not a tested generation result.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitchesintegrated_multimodal_description: A quiet kitchen at dawn, cool window light across a wooden table. [Shot 1] A woman folds a letter beside a cup; the camera pushes in with small amplitude at slow speed toward her hands. (S1) She looks at the letter and says in a calm voice: <d>en-US: I thought I had more time.</d> [Shot 2] At 00:04.000, the camera cuts to the rain-streaked window as she sets the letter down out of frame.
overall_soundscape: Soft rain against the window continues throughout. The cup makes a light ceramic sound on the table, followed by a brief rustle of paper.
non_diegetic_music: Sparse, slow piano notes with long gaps; the music grows slightly fuller near the end.Here, speech is attached to the timeline event, rain and handling sounds form the ongoing soundscape, and the piano is a score for the viewer rather than a sound source inside the kitchen.
Before you submit the prompt
- Confirm the opening instruction matches T2VA, I2VA, FL2VA, L2VA, or the reference workflow you selected.
- Describe visible actions and their order rather than relying on mood words alone.
- Make camera direction clear; add amplitude and speed only where they affect the intended movement.
- Leave the first shot untimestamped; number later shots in sequence and use increasing cut times within the clip.
- Use a cut for a meaningful change of information, not merely a small reframing.
- Keep spoken or sung content in the timeline, retain supplied wording and punctuation, and use consistent speaker IDs.
- Separate moment-specific in-scene sound from continuous ambience and audience-only music; use
N/Afor the relevant silence or absent score as directed by the guide. - Check image/reference count and duration limits against the current official workflow before using them.
MiniMax’s published output specifications
MiniMax’s H3 repository, accessed in 2026, lists 24 FPS, 32 kHz stereo audio, a default 768-pixel shorter side, support for a range of aspect ratios, and 2K regeneration through H3-Regenerate-2K. It also lists stable support for dialogue in 11 languages: Arabic, Chinese, English, French, German, Italian, Japanese, Korean, Portuguese, Russian, and Spanish; the repository says support for additional languages varies. These are specifications and capability statements published by MiniMax, not independently verified performance findings. The repository describes H3 as a system comprising H3-Context-IR for preprocessing and multimodal instruction refinement, H3-Base for 768p audio/video generation, and H3-Regenerate-2K for 2K regeneration; it says Context-IR is hosted and not included in the open-source release, and identifies an API for the official workflow. See the MiniMax H3 repository for the company’s current product description and access information.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
