AI VIDEO / MINIMAX H3
Turn a written scene, a first or last frame, or a set of references into a 4–15 second video. Choose 768p or 2K, review the credit estimate and download the finished clip.
THE TOOL
MiniMax H3 is a video generation model that accepts a written scene and, depending on the mode, images or other reference media. This MiniMax H3 AI Video Generator brings those inputs into a browser workspace without installing an app or configuring a local model. You can create a new scene from text, guide a clip with its first or last frame, or combine visual references with supporting audio.
This is an independent MiniMax H3 tool, not MiniMax’s official website or a model-download service. Its controls differ from the Economy and Standard options in our Pika H3 Video Studio. Choose the workspace for the controls you need rather than treating the names as interchangeable. Here, output duration runs from 4 to 15 seconds, with 768p and 2K resolution choices.
Use the result as a clip to review and edit. Generating a video does not guarantee exact lettering, unchanged facial details, or a finished campaign. Watch the complete output and check the parts that matter to your project before sharing it.
PLAN THE SHOT
INPUTS AND CONTROLS
Each mode solves a different starting problem. Keep the inputs relevant and let the prompt explain how they work together.
Write a subject, a visible action and a camera direction. Choose from six frame shapes: 21:9, 16:9, 4:3, 1:1, 3:4 and 9:16. A short, specific description is usually easier to inspect than a list of competing ideas. Add a sound cue when it belongs to the moment rather than using it to introduce a second scene.
Image to Video accepts either endpoint independently. A first frame establishes the opening; a last frame supplies an ending. With both, describe a plausible transition between them. Do not expect unrelated compositions to join seamlessly. This mode does not send a separate aspect-ratio setting, so prepare your images for the intended composition.
Reference to Video accepts up to nine images, three videos and three audio files. Include at least one image or video. Explain which reference supplies the subject, movement or atmosphere; simply adding more media can create competing instructions. Keep the set focused and remove references that do not contribute to the requested shot.
Select a whole number from 4 to 15 seconds and choose 768p or 2K. Give the action time to begin and finish. Higher resolution changes the output size, but it does not fix a contradictory prompt or guarantee accurate small details. Compare results at the same duration when you are deciding which setting suits your work.
Reference mode offers adaptive framing alongside the six explicit ratios. Use an explicit shape when you already know the destination, such as a portrait placement or wide presentation. With adaptive framing, inspect the returned dimensions before planning a final crop. Your publishing platform may have its own size requirements.
Upload images up to 10 MB each and video or audio files up to 100 MB each through this workspace. Use the upload cards to attach your files; image previews help you check the selected material. A pending task remains available while you browse other pages in the same session. Use its existing status controls to continue checking rather than submitting another chargeable request.
HOW TO USE MINIMAX H3
Open Text to Video for a written idea, Image to Video for an endpoint image, or Reference to Video for a collection of media. The page opens in text mode to keep the first step simple. Switching modes does not start a task.
Write what should happen, then upload the necessary material. For reference videos, enter their combined input duration in seconds. This is separate from the requested output duration and affects billing. Wait for each upload to finish and check the selected files before submitting.
Select duration, resolution and any available frame-shape setting. Sign in, check your account balance and submit once. The percentage is estimated: it can stop at 99% while the service finishes. A successful response moves it to 100% even if it arrives earlier.
Preview the opening, movement and ending. Enlarge the video when checking faces or product details. Download the clip, or use My creations to find your work. For a revision, change one instruction and keep the other settings stable so the difference is easier to understand.
Need a clearer description? The Pika H3 prompt guide offers general scene-writing examples you can adapt. Its model-specific settings belong to the Pika workspace; use the MiniMax controls on this page for your actual request.
MATCH THE TASK
Explore a product in a setting before committing to a larger production. Start from an approved product image when its silhouette matters. Describe a simple reveal, a small camera move or a change in light. Keep written packaging and logos on your review checklist.
The useful output is a direction to evaluate, not proof that every brand detail is correct. Compare the result with your source image and keep any final factual or promotional claims outside the generation brief until they are checked.
Build a single visual moment for a larger edit: an opening shot, a transition or a character reaction. Choose the frame shape for the placement and leave space where you plan to add captions. A short clip should make sense even before music and overlays are added.
Review how the action reads on a small screen. If the subject is hard to follow, simplify the composition or movement before increasing duration. Assemble separate shots in your editor when the story needs more than one scene.
Use references when describing a visual relationship is harder than showing it. An image can establish a subject, while a video supplies a movement idea. Explain each reference’s role and remove contradictions before submitting. Only use material you have permission to upload.
Keep a record of your prompt and settings when comparing attempts. Evaluate whether the intended characteristic survived, rather than expecting a frame-for-frame copy. Test changes separately so you can decide which instruction improved the result.
All tools share your account credits. Visit credit packs and pricing to top up; the calculator there describes Pika H3 Economy and Standard, not MiniMax H3. Use this generator’s own estimate and check your account credit history for settled deductions.
Describe the subject, action, framing and light before adding stylistic details.
In image-to-video mode, let your uploaded frame establish the appearance and use the prompt to explain movement. With first and last frames, describe how the subject moves between the two positions.
A glass perfume bottle stands on a pale stone shelf. A hand turns it slowly toward the window while the camera holds a close-up. Soft side lighting, a quiet room, one continuous shot.
Original writing example, not a tested output claim.
In reference mode, explain which file supplies the subject, setting or motion. The prompt-writing guide has more scene structures you can adapt.
Choose your resolution, then budget for the output and any extra reference media.
The shortest option is four seconds at 768p. The shared 15-credit welcome allowance alone does not cover that minimum.
Check the on-page estimate after changing your references. Compare available credit packs if you need a larger balance, and read the payment and refund guidance if a purchase needs support.
| Resolution | Per second | 5-second clip |
|---|---|---|
| 768p | 6 credits | 30 credits |
| 2K | 8 credits | 40 credits |
Reference video duration is charged at the same per-second rate as the selected output resolution. Each reference image beyond the first five adds 3 credits.
Example: a 5-second 768p output, 2 seconds of reference video and 6 reference images. Fractional totals round up to whole credits.
COMPARE YOUR OPTIONS
MiniMax H3 offers 768p and 2K output with clips starting at four seconds. Compare it with the Economy and Standard options in Pika H3 before choosing your settings. See current credit packs and pricing when planning your budget.
| Feature | Pika H3 · Economy | Pika H3 · Standard | MiniMax H3 |
|---|---|---|---|
| Available resolution | 480p, 768p | 480p, 768p | 768p, 2K |
| Output duration | 5–15 seconds | 5–15 seconds | 4–15 seconds |
| Text to Video | Supported | Supported | Supported |
| Image to Video | First frame required; last frame optional | First frame required; last frame optional | First frame, last frame, or both |
| Reference to Video | Switch to Standard | Images, video and audio references | Images, video and audio references |
| Reference requirements | Not available in Economy | At least one image or video; up to 12 files total | At least one image or video; up to 9 images, 3 videos and 3 audio files |
| 5-second text video at 768p | 15 credits | 25 credits | 30 credits |
| Seed and prompt expansion | Available | Available | Not offered in this workspace |
| Preview and download | Included | Included | Included |
The credit example compares one five-second text-only attempt at 768p. Reference inputs can add charges; check the estimate for your selected mode before generating. These are workspace settings, not a quality or speed ranking.
For lower-cost short tests or seed and prompt-expansion controls, explore the Pika H3 Video Studio.
A FEW PRACTICAL ANSWERS
No. This page uses the dedicated MiniMax H3 service. The Pika workspace offers Economy and Standard presets with different settings and prices. Your account credits and video library are shared, but each tool keeps its own generation controls.
You can browse this tool and prepare a prompt without spending credits. Generating a video requires enough account credits for the selected request. New accounts receive 15 one-time welcome credits, which do not expire and are shared across the site. MiniMax H3 costs 6 credits per second at 768p or 8 at 2K. The shortest text-to-video request is 4 seconds at 768p, estimated at 24 credits, so the welcome balance alone does not cover it. Reference video seconds use the same rate, and each reference image beyond the first five adds 3 credits. Review the estimate before submitting and add credits if needed.
No. Reference mode needs at least one image or video. Audio can accompany that visual material. Describe how the sound relates to the scene, and review the returned clip rather than assuming every audio cue will be reproduced exactly.
The progress bar is an estimate, not a measurement from the model. It waits for an actual completed response before showing 100%. If status checks pause, use Resume status checks. A slow response is not a reason to create another task immediately.
A failed generation returns its credits to your account. Check the error message and credit history. A network timeout is different: the service may have received your request even if the browser did not receive its task ID. Check My creations before trying again.
No. This is an online MiniMax H3 generator: enter your prompt, upload any references and review the settings in your browser. GitHub, Hugging Face and ComfyUI searches relate to a separate local-model workflow; this page does not distribute model files or an APK. Local extensions such as LoRA and advanced seed controls are not offered here.
You can browse within the site while a task runs. The same browser session retains the task ID and continues using the appropriate status endpoint. Keep a separate copy of important prompt text; browser-session storage is not a permanent project backup.
Use material you own or have permission to process. Avoid uploading private information or content that violates another person’s rights. A generated result still needs your review before publication. Read the terms of service for the site’s usage rules.
