How to Use MiniMax H3: Step-by-Step Tutorial for 2K AI Video With Sound

How to Use MiniMax H3: Step-by-Step Tutorial

Open the MiniMax H3 generator. The steps below follow the current workspace labels. For your first attempt, use a short Text to Video shot so you can learn the controls without preparing reference files.

Step 1: Open the Workspace and Sign In When Needed

The workspace opens in Image to Video. You can inspect the three input tabs and start writing before you sign in. When you click an upload area or Generate, continue with Google to access uploads and generation.

Before you have a generation, the right-hand panel says Get inspired and displays example clips. Once you have a task, it changes to Your videos. These examples are inspiration, not videos you have created.

Step 2: Choose the Right Input Mode

Choose a mode based on what you already have. First & Last Frame is a workflow inside Image to Video, not a separate tab on this page.

For prompt-only creation, select Text to Video before moving to the next step. If you already have source material, the dedicated Image to Video guide, first and last frame guide, and Reference to Video guide cover those workflows in more detail.

Step 3: Add Your Source Files

Skip uploads in Text to Video and continue to Step 4. For other workflows, add your source files above Prompt. Sign in when prompted to upload media.

In Image to Video, upload a Start frame, an End frame, or both. Use JPG, JPEG, PNG, WEBP, HEIC, or HEIF images up to 30 MB per file. Each side must be 256–5760 pixels, with a width-to-height ratio between 0.4 and 2.5. The uploaded frame appears in its slot so you can confirm the image before writing your motion instructions.

In Reference to Video, the Reference media area has a separate row for each file type:

• Image references: up to 9 images to guide appearance, product details, style, or composition.
• Video references: up to 3 videos to guide movement, camera behavior, or timing. Their combined duration contributes to the credit estimate.
• Audio references: up to 3 audio files to guide rhythm, dialogue, music, ambience, or sound cues.

Image references must be 30 MB or smaller and meet the image rules above. Video and audio references can be up to 100 MB per file. Include at least one image or video; audio references alone cannot start a generation. These limits apply to the standard H3 workspace. H3 Max has different reference limits, and Turbo does not support Reference to Video.

Uploaded images and square video previews appear from left to right. Use Add more beside the previews to add another file within the limit. Audio players appear below their upload area, one per row. Confirm the previews and wait for uploads to finish before continuing.

Use files you own or have permission to use. In Step 4, assign each reference a clear role—for example, “Use Image 1 for the character's appearance, Video 1 for walking motion, and Audio 1 for pacing.” Your prompt should describe what changes while preserving the details supplied by your media.

Step 4: Describe the Shot

Write in the Prompt box. The counter supports up to 7,000 characters, but you don't need to fill it. Describe what happens during the clip, how the camera observes it, what the viewer hears, and where the shot ends.

Use this example as a starting brief. It gives you specific things to check later: a red ball, a low camera angle, audible paw taps, and a final close-up.

A golden retriever puppy pushes a red ball across a sunlit wooden porch, then stops and looks up at the camera. Low-angle tracking shot at the puppy's eye level. Warm late-afternoon light, shallow depth of field. Soft paw taps on the wood, a light breeze, and distant birdsong. End on a close-up of the puppy's face, tail wagging.

Keep the prompt easy to evaluate. If you add a second location, several camera moves, or a long conversation, it becomes harder to tell which instruction caused a problem. For a fuller writing framework, use the MiniMax H3 Prompt Guide.

Step 5: Set the Output and Review Credits

Open Aspect ratio to choose where the video will fit. Use 16:9 for regular YouTube videos and web pages, 9:16 for TikTok, Reels, and Shorts, or 1:1 for square posts. The workspace also offers 21:9, 4:3, and 3:4.

When you upload an image, the tool selects a suitable aspect ratio automatically unless you have already chosen one yourself. Check the selection, especially when adapting a landscape image to a vertical video: important details near the edges may no longer fit the intended composition.

Start with 768P and 5 seconds for a first draft. Switch to 2K or extend the Duration slider only when you know what the longer or higher-resolution version needs to achieve. H3 supports durations from 5 to 15 seconds.

Review the live credit estimate near Generate. A five-second H3 clip estimates 15 credits at 768P or 25 credits at 2K before additional reference charges. Changing the duration, resolution, or reference inputs can change the total.

Step 6: Generate, Review, and Download

Click Generate. If you are signed out, complete sign-in first, then review the restored prompt and settings before starting. If a required prompt or frame is missing, complete that field rather than repeatedly clicking the button.

Your task appears in Your videos. Standard H3 generations usually take 2–5 minutes. The progress bar is labeled Estimated progress; its percentage is a waiting estimate, not a measurement of frames already rendered.

You can return later through My Creations. Once the task finishes, play the entire clip with sound on. Check the opening, the main action, the camera movement, and the final frame. Listen for the sounds you requested, rather than judging the result from its thumbnail alone.

Use Download on the completed result card to save the video. Keep the prompt and source files with your downloaded draft so you can reproduce the setup for another attempt.

Step 7: Make One Deliberate Revision

Decide which single problem matters most before generating again. For the puppy example, a shot that tracks too quickly needs a camera-speed change; it doesn't necessarily need different lighting or a new subject description.

Replace “Low-angle tracking shot at the puppy's eye level” with “Slow low-angle tracking shot that keeps pace with the puppy.” Keep the other instructions and output settings the same, then compare the two versions.

If the change helps, keep it. If it introduces a new problem, return to the earlier wording and try a different adjustment. This makes each new generation a useful test instead of a complete restart.

Refine a Result Without Starting Over

For Image to Video, separate appearance from motion. A supplied product photo is your visual anchor; your prompt should explain what moves, what stays still, and how the camera behaves. Avoid inventing new product colors or materials that contradict the image.

Keep the exact shoe shape, colors, stitching, and logo placement. The sneaker rotates slowly on a white turntable while the camera pushes in. A soft shadow moves across the floor. A quiet studio hum and a single click as the turntable stops. End on a three-quarter hero view of the shoe.

When reviewing this draft, look separately at the stitching, the camera motion, and the final angle. If the shoe looks right but rotates too far, change the rotation instruction. If the label distorts, reduce motion and strengthen that preservation rule rather than adding more style adjectives.

For additional examples, visit MiniMax H3 text to video or learn how to animate an image with MiniMax H3. Use those pages for workflow-specific inspiration after the basic controls feel familiar.

Fix Common Workspace Problems

Some problems happen before a video is generated. Check the field or task state first so you can distinguish a setup issue from a creative result that needs revision.

If a task fails, keep its details before trying again. Failed or rejected generations are automatically refunded according to this workspace's pricing policy. For a missing result or a credit discrepancy, contact support with the task information so the specific generation can be checked.

Credits and Waiting Time

Check the estimate for each generation rather than assuming every attempt costs the same. Standard H3 uses 3 credits per output second at 768P or 5 at 2K. Reference video adds a charge at the selected output rate; the first five reference images are included, and additional images cost 2 credits each.

A five-second 768P draft uses 15 credits before extra reference charges. Use the MiniMax H3 pricing page for current credit packs, rates, and refund details rather than treating an example in this guide as a fixed quote for every workflow.

Standard H3 usually takes 2–5 minutes. If faster testing is your priority, explore MiniMax H3 Max, whose typical generation time is 10–30 seconds. It has its own resolution options and credit rates, so review that workspace before switching.

Check the Video Before You Publish

Download and review the version you actually plan to use. A successful task means a video was delivered; it does not guarantee that every face, logo, line of dialogue, or product detail is accurate.

• Inspect details across the clip. Check faces, hands, labels, and small objects beyond the opening frame.
• Listen to the complete audio. Confirm that dialogue is audible and not cut off, and that background sound fits the scene.
• Check framing and the ending. Make sure the subject remains visible and the last frame works for your intended placement.
• Confirm your rights. Review the Terms of Service and your rights to uploaded media before paid ads or client work.

Keep a copy of the prompt and note what changed between versions. A short record of your decisions is more useful for your next shot than a collection of unlabeled downloads.

MiniMax H3 FAQ

Continue With Your Next Shot

Now you know how to use MiniMax H3: choose an input, describe the shot, review settings and credits, then click Generate. Save a useful first draft and make the next attempt test one clear decision.

minimax-h3.app is an independent service and is not affiliated with or endorsed by MiniMax or Hailuo AI. All trademarks belong to their respective owners.