How to Use MiniMax H3 Max: Full Setup and Cost Guide

MiniMax H3 Max vs Turbo: Which One to Use

Both versions generate at 5–15 seconds and 480p or 768p — this workspace doesn’t offer 1080p or 2K on either one.

Choose MiniMax H3 Max Turbo when:

• The task is text-to-video or first/last-frame

• You’re iterating on early drafts and want the lower per-second cost

• You don’t need image or audio reference guidance

Choose MiniMax H3 Max when:

• The task needs image references, audio references, or both

• You’ve already confirmed the prompt on Turbo and are generating a final version that needs reference control

A cost-efficient workflow: draft and refine on Turbo first, then move only the confirmed prompt to Max if the task requires reference guidance Turbo doesn’t offer.

This split mirrors a pattern worth applying to any credit-metered generation tool: separate the cost of finding the right prompt from the cost of producing the final asset. Most of the value in iteration comes from cheap, fast attempts that let you see what the model actually does with a given instruction — the version that ends up in a deliverable is usually the last of many, not the first. Turbo is built for that first phase; Max earns its higher per-second cost only once you’re past it and need capability Turbo simply doesn’t have.

It’s also worth noting what doesn’t change between the two versions: prompt structure, aspect ratio options for text mode, and duration range are identical. Switching from Turbo to Max for a confirmed prompt doesn’t mean rewriting anything — it means changing the Model version control at the bottom of Output settings and, if the task calls for it, attaching reference files that Turbo doesn’t accept in the first place.

How to Use MiniMax H3 Max on ZMS AI: Step by Step

Step 1: Choose an Input Mode and Select MiniMax H3 Max

Open the MiniMax H3 Max workspace on ZMS AI. The first dropdown sets your input mode:

• Text to video — describe the shot from nothing but a prompt.

• First / last frame — a starting image is required; an ending frame is optional and adds 15 credits.

• Mixed references — combine up to 12 files total (images plus optional audio). Each audio file runs 2–15 seconds, totaling no more than 15 seconds combined. Video-reference submission isn’t available in this workspace.

Then select MiniMax H3 Max from the model dropdown next to it. Max and Turbo aren’t two separate entries here — both live under this one model selection, with the specific version chosen later in Output settings.

Step 2: Write the Prompt

Describe the subject, one clear action, the camera movement, and how the shot ends. If audio matters to the shot, name the specific sound and when it happens in the same prompt, rather than treating audio as a separate concern.

In Mixed references mode, state what each file is for: “match the bottle shape and cap color from the image, use the room tone from the audio clip.” Assigning a role to each file avoids conflicting guidance across a multi-file reference set.

Step 3: Open Output Settings — Aspect Ratio, Resolution, Duration, Prompt Expansion, and Model Version

Open Output settings from the third dropdown (it shows a summary like “16:9 · 5s · 768p”). This one panel holds every remaining setting:

• Aspect ratio: 21:9, 16:9, 4:3, 1:1, 3:4, or 9:16 for text mode; Mixed references mode adds Adaptive.

• Resolution: 480p or 768p.

• Duration: 5 to 15 seconds, in 1-second steps.

• Prompt expansion: Disabled generates from your exact wording. Balanced (default) and Quality both rewrite your prompt with added detail before generating.

• Model version: Max or Turbo — this is the last control in the panel, easy to overlook if you’re used to picking a version before anything else.

Set Prompt Expansion to Disabled when testing whether a specific instruction actually works — otherwise you may be seeing the model respond to its own expanded version of your prompt, not what you wrote. Use Balanced or Quality once you’re generating a version you plan to keep.

Since Model version sits at the bottom of this same panel, it’s worth deciding Max vs. Turbo before you open Output settings, rather than defaulting to whichever was last selected — see the comparison above for which one fits your task. Click Done to confirm; your full settings collapse back into the dropdown summary so you can check them before generating.

Check the credit estimate before generating, and again after any setting or file change — it updates live and doesn’t carry over from your last check.

Step 4: Generate, Track, and Review

Open Task Center to follow progress, then View output once the task completes, or find it later in My Creations. Review motion, object details, and audio timing before using the clip anywhere — a completed task doesn’t guarantee the composition matches your intent.

If a result is close but not quite right, change one instruction and regenerate, rather than rewriting the full prompt — that’s the only way to isolate what actually caused the difference. This matters more on H3 Max than it might on a slower model: because generation is fast and Turbo’s cost per attempt is low, the discipline of changing one variable at a time is cheap to maintain and actually tells you something, rather than being a nice-to-have you skip under time pressure.

MiniMax H3 Max Credit Cost Breakdown

Example: a 5-second Max text draft at 480p costs 75 credits; the same draft on Turbo costs 40. A 5-second Max image task at 768p with one starting frame costs 100 credits; adding an ending frame brings it to 115. Cost scales linearly with duration — a 15-second clip costs three times as much as a 5-second one at the same settings.

MiniMax H3 Max Prompt Examples from the ZMS AI Gallery

The MiniMax H3 Max workspace includes worked examples you can copy directly.

The perfume close-up example uses first/last-frame mode: a starting image of a glass bottle, paired with a prompt that names exactly what must stay fixed (bottle silhouette, cap shape, color, background) and what’s allowed to move (camera position, light reflections).

The coastal gallery example runs the identical prompt on both Max and Turbo, useful for judging whether Turbo’s lower cost shows a visible quality difference for your specific type of shot.

Both examples share a pattern worth copying into your own prompts: they separate what the shot preserves from what it changes, in explicit language, rather than describing the scene as one continuous block of detail. “Keep the bottle silhouette, cap shape, and color exactly as shown” is a different instruction than a general description of the bottle — it tells the model which details are non-negotiable versus which ones it has room to render its own way.

Common MiniMax H3 Max Mistakes to Avoid

Defaulting to Max when Turbo would do the job. If the task is text-to-video or first/last-frame with no reference files, Turbo produces the same input flexibility at roughly 40% lower cost. Reach for Max only once a task actually needs image or audio reference guidance.

Leaving Prompt Expansion on Balanced while debugging a prompt. If a generation doesn’t match what you wrote, and expansion is set to Balanced or Quality, you’re not actually testing your own wording — you’re testing the model’s elaborated version of it. Switch to Disabled first to see what your exact prompt produces, then decide whether expansion helps.

Uploading reference files without assigning them a role. A pile of images and audio with no stated purpose in the prompt leaves the model guessing which file governs appearance, which governs sound, and which is just noise. State it directly for each file.

Not rechecking the credit estimate after a change. Every added reference file, extra second of duration, or resolution change shifts the total — the number you saw before your last edit doesn’t necessarily reflect what you’re about to generate. Recheck before clicking Generate, not after.

Comparing Max and Turbo results at different settings. If you’re genuinely trying to judge whether Turbo’s lower cost shows in the output, hold duration, resolution, and prompt identical between the two runs — the coastal gallery example in the ZMS AI gallery does exactly this, which is why it’s a more useful comparison than two clips generated under different conditions.

Choosing the longest duration for a first draft. Cost scales linearly with duration, so a 15-second first attempt costs three times what a 5-second test does. Confirm the concept short, then extend only the version you intend to keep.

Assuming a completed task means a correct result. A task finishing successfully confirms generation worked, not that the composition matches your brief. Review motion, object details, and audio timing before using a clip anywhere, especially before it goes into a deliverable someone else will see.

MiniMax H3 Max FAQ

How do I use MiniMax H3 Max on ZMS AI? Open the workspace, choose Max or Turbo, select an input mode (text, first/last frame, or reference for Max), write your prompt, set duration, resolution, and aspect ratio, then click Generate.

Is MiniMax H3 Max or Turbo better? Neither is better overall — Turbo costs about 40% less per second for text and first/last-frame tasks, while Max is the only version that supports image and audio references. Use Turbo by default, and switch to Max when a task needs reference guidance.

How much does MiniMax H3 Max cost on ZMS AI? Output costs 15 credits/second at 480p or 20/second at 768p on Max, and 8/second at 480p or 12/second at 768p on Turbo. References add cost separately: 15 credits per image, 1 credit per second of reference audio.

What is the maximum video length for MiniMax H3 Max? 15 seconds, in whole-second increments starting at 5. This applies to both Max and Turbo.

What resolution does MiniMax H3 Max support? 480p or 768p. This workspace doesn’t offer 1080p or 2K on either version — for 2K, use base MiniMax H3 instead.

Can I upload a video as a reference in MiniMax H3 Max? No. Mixed references mode on this workspace accepts images and optional audio only; video-reference submission isn’t available.

What does Prompt Expansion do? Balanced (default) and Quality rewrite your prompt with added detail before generating. Disabled generates from your exact wording — use it when testing a specific instruction.

Does MiniMax H3 Max generate audio automatically? Audio generates as part of the same output when your prompt describes sound, and Mixed references mode also accepts an audio file to guide tone or ambience directly. There’s no separate audio-off toggle documented in this workspace — describe sound explicitly in the prompt if the shot should be silent aside from ambient room tone, or omit sound description if you don’t need audio to be a deliberate part of the result.

Can I switch between Max and Turbo partway through a project? Yes. Both versions share the same prompt structure, aspect ratio options, and duration range, so moving a confirmed prompt from Turbo to Max — or the reverse — doesn’t require rebuilding anything, only changing the Model version control inside Output settings and, for Max, attaching any reference files the task needs.

Try MiniMax H3 Max on ZMS AI

Open the MiniMax H3 Max workspace, start on Turbo for your first draft, and switch to Max only once a task needs reference guidance. For a full model comparison, see MiniMax H3 vs Wan 3.0.