Start with a prompt or still, choose a 5–15 second format, and make the next version while the direction is clear. Host testing reports a 5-second 768P take in under three seconds.
A 5-second 768P H3 Max take made from a detailed scene prompt.
Make a short video, choose the controls that matter, and move into the next version without losing the direction.
The MiniMax H3 Max video generator lets you turn one clear instruction into a 5–15 second video: write the action, subject, setting, and camera move, or upload a still to lock the opening composition. Choose duration, resolution, and framing before you generate. The host reports that a 5-second 768P take renders in under three seconds in its August 2026 testing; actual completion time can vary with the queue and input. Use the first result to adjust one visible decision, then make the next take.
| MiniMax H3 Max video generator | prompt to video | still to video | short AI video workflow | 480P and 768P video |
Two reliable starting points
Use text when the scene starts in your head; use a still when the opening composition is already right.
A useful text-to-video prompt names a subject, action, setting, and camera instruction. “A ceramic teapot rotates on wet black stone, close focus, slow orbit” gives the model a visible job. Add light, texture, weather, or a transition only when it helps you judge the take. You get a compact visual draft to compare against the brief.
Image-to-video is the calmer route when a product, character, layout, or key visual must begin in a known place. Upload a still as the start frame, describe the movement you want to introduce, and use an optional end frame when the shot needs a clear destination. That makes H3 Max useful for turning a storyboard panel into motion, trying a gentle product orbit, or testing how a social still could open into a clip without abandoning its original composition.
Text-to-video offers 21:9, 16:9, 4:3, 1:1, 3:4, and 9:16 framing, with 16:9 as the default. Choose a wide frame for a landscape beat, a square frame for a feed, or a vertical frame for a phone. Image-to-video follows the start frame, so the reference still carries its established composition into the motion.
At 768P, output is 1344×768 at 24 FPS with natively synced audio. Review timing, camera energy, and whether the scene reads at the chosen length in one pass. Then change the instruction that caused the miss instead of rewriting the entire idea.
Start with an intentional input, set the output, and use each take to make the next decision.
Give the generator a single moment to solve. Name the subject, what happens, where it happens, and how the camera behaves. For example: “A red running shoe lands on a rain-dark track, low tracking camera, streetlight reflections.” If a still defines the opening composition, upload it before you write the motion direction.
Choose a whole-second duration from 5 to 15 seconds. Pick 480P when you want to explore an idea, then use 768P for a sharper reviewable take; 768P is the default setting. Text-to-video also offers 21:9, 16:9, 4:3, 1:1, 3:4, and 9:16. Image-to-video keeps the start frame’s proportions, so prepare the reference in the format you expect to publish.
Choose whether movement, composition, timing, or scene detail matters most before you watch. A product may need the label to remain legible; a character shot may need the camera distance to hold; a transition may need a stronger final image. Turn that observation into the next direct instruction.
Keep the elements that worked and alter the part you can name. Change “fast” to a specific camera move, add the direction of travel, simplify a crowded background, or provide a more decisive end frame. Re-running a clear variation makes comparison easier than replacing every line at once. When a take earns a place in your edit, keep the prompt and settings with it so the visual language can be continued deliberately.
Choose the controls that fit the destination instead of stretching one output across every job.
Choose an integer duration from 5 to 15 seconds, with 5 seconds as the default. A shorter take is useful for a product gesture, reaction, or transition; a longer take gives a scene room for a camera move and a completed action. Plan one beat per generation rather than asking a single clip to carry a whole narrative.
This generator offers 480P and 768P, with 768P selected by default. Treat 480P as a practical way to test direction, then bring the chosen shot back at 768P when detail matters for review. A 768P output is 1344×768 at 24 FPS, so the visual decision can be made against a clear, consistent format.
Text prompts can render in 21:9, 16:9, 4:3, 1:1, 3:4, or 9:16. That range lets the same idea start as a cinematic wide shot, a familiar web video, a feed-ready square, or a vertical social clip. Decide the output home early; reframing after generation changes the balance of the image.
Image-to-video begins with a required start frame and can use an optional end frame. The first image anchors layout, while the second can clarify where motion should arrive. Use this pair for before-and-after transitions, matched campaign stills, product reveals, or a storyboard sequence where beginning and end need to feel related.
A 768P video includes natively synced audio. Listen during review: sound can change whether movement feels abrupt, whether a scene has enough atmosphere, or whether the take belongs in the sequence you are assembling.
H3 Max keeps prompt, reference, duration, resolution, and frame selection close to the generation action. That makes it easier to compare a few purposeful versions of one shot. Preserve the condition that worked, revise the condition that did not, and build a small library of tested directions instead of relying on an unrecoverable one-off result.
Choose a starting method
The right starting method depends on how much of the opening shot is already decided.
| First input | A text prompt | A required start frame |
|---|---|---|
| Best when | The scene, action, and camera move begin as a written direction | The product, character, or composition already exists in a still |
| Framing | 21:9, 16:9, 4:3, 1:1, 3:4, or 9:16 | Follows the start frame |
| End-state control | Clarify it in the prompt | Add an optional end frame |
| Duration | 5–15 seconds | 5–15 seconds |
| Resolution | 480P or 768P | 480P or 768P |
From a creative decision to a usable draft
H3 Max is most useful when the next step depends on seeing a motion idea instead of debating it in the abstract.
A social team can turn a product line and camera cue into several short opening beats before choosing a production direction. Start with the same product description, vary the setting or movement, and compare which idea reads in vertical, square, or wide framing. The point is not to finish a campaign in one prompt; it is to make the first three seconds tangible enough for a marketer, designer, and editor to discuss the same thing.
Use a clean product still as the start frame when package shape, color, and label placement are already approved. Ask for a restrained orbit, a short push-in, or a change in light, then use an end frame when the reveal has to land in a known arrangement. This gives a product page or launch deck a moving direction to review before a full shoot or compositing task begins.
A director, producer, or designer can move a storyboard panel into a 5–15 second test that reveals pacing and camera intent. One image can establish the opening frame; a written instruction can describe the motion; an end frame can state the destination. The generated take is a practical way to find whether a scene needs a closer camera, a clearer action, or a simpler transition before the plan spreads across a larger sequence.
Editorial teams often need a visual hook while an article, announcement, or event is still timely. Build a short landscape or vertical clip around one concrete image: a city map coming alive, a title card becoming a physical sign, or a still photograph taking on a gentle camera move. The generator gives the post a motion-first draft to assess alongside the actual copy and publishing channel.
A pitch is easier to evaluate when everyone can see the tone, scale, and movement of a proposed scene. Generate short interpretations from the same core brief—one quiet, one kinetic, one close, one wide—then use the selected take to sharpen the treatment. This helps when a reference image gives the team a shared visual language but motion, camera, or transition still needs a decision.
A product or brand team can use a small video study to discuss the energy of an interface moment, campaign graphic, or animated identity direction. Keep the request focused on one behavior: panels sliding with a rhythm, a mark reacting to a gesture, or a device shot moving through a defined environment. The output can turn a verbal animation preference into a visible reference for the people who will build it.
You can create a short video from a text prompt or a still image. Text is useful for a new scene; image-to-video begins from a required start frame and lets you introduce movement to a product shot, character image, storyboard panel, or other composed visual. Each take can run from 5 to 15 seconds.
In August 2026 host testing, a 5-second 768P H3 Max take rendered in under three seconds. Actual completion time can vary with the queue and input, so treat that result as a performance reference rather than a delivery-time promise.
The generator supports whole-second durations from 5 to 15 seconds, with 5 seconds as the default. Use a shorter take for one visual beat, transition, or product movement. Use more time when the shot needs an established setting, a camera move, and a completed action; keeping one clear beat per generation makes the result easier to direct and compare.
Yes. Image-to-video uses a start frame as its required input, so the opening composition comes from your still. Add a motion description to say what should happen next. You can also provide an optional end frame when you need a clearer destination, such as a product reveal, a before-and-after transition, or a sequence that must finish in a particular layout.
H3 Max offers 480P and 768P output, with 768P selected by default. The 768P output is 1344×768 at 24 FPS with natively synced audio. Use the setting that fits the decision you are making: 480P for exploring a direction and 768P for a sharper reviewable take where image detail and rhythm need closer attention.
Text-to-video supports 21:9, 16:9, 4:3, 1:1, 3:4, and 9:16, with 16:9 as the default. Choose the ratio based on where the video will be reviewed or published rather than treating it as a last-minute crop. Image-to-video follows the shape of the start frame you upload, so prepare that image in the proportion you want to carry forward.
Start with one shot. State who or what is visible, what happens, where it happens, and how the camera behaves. Add only details that help a reviewer recognize success: a direction of travel, a material, a lighting condition, or a visible transition. If composition is already important, use a start frame and write only the movement you need.
A 768P output is rendered at 24 FPS with natively synced audio. Keep playback in the review loop because sound changes how a motion beat, atmosphere, and cut point are understood. The rendered take can give a creative team useful context to decide what to refine next, while final mixing can remain part of the wider production process.
Use an end frame when the shot needs to arrive somewhere precise. It is helpful for a transition between approved visuals, a product reveal that must finish in a known arrangement, or a storyboard moment whose final composition is already designed. If you only need a still to define the opening look, a start frame plus a clear motion instruction is the simpler choice.
Yes. Keep the subject and core action stable, then change one meaningful condition at a time: the camera direction, light, duration, frame, or reference image. This creates a useful comparison set instead of several unrelated surprises. Save settings with the take that works so the next shot can build on the visual choices already made.
Describe the movement or begin with a still, choose the format, and make the next take while the direction is fresh.
