# MiniMax H3 Max Turbo vs Max: Which Fits Your Video?
By SeedRouter · Published 2026-10-20 · Updated 2026-10-20
**Choose MiniMax H3 Max Turbo for text or frame-guided video; choose Max when the shot needs reference images, video or audio.** On SeedRouter, both accept 480P or 768P and five to fifteen whole seconds. Turbo charges output seconds only. Max has separate output, reference-video and extra-image charges. The [Turbo Playground](https://seedrouter.ai/models/minimax-h3-max-turbo#playground) and [Max Playground](https://seedrouter.ai/models/minimax-h3-max#playground) expose each model's own settings.
This comparison concerns the native **content-array interface**, with model ID `minimax-h3-max-turbo` on SeedRouter. Other services use the same display name for a separately adapted interface. In particular, fal documents fractional durations, a 1080P option and additional audio controls. Those fields do not belong in the SeedRouter request.\[1] Distinguishing the interfaces avoids choosing a model for a feature the selected endpoint does not offer.
## MiniMax H3 Max Turbo vs Max at a glance
| Requirement | H3 Max Turbo on SeedRouter | H3 Max on SeedRouter |
| --------------------------------- | ------------------------------------ | ---------------------------------------------------------- |
| Output resolution | 480P, 768P | 480P, 768P |
| Duration | 5–15 whole seconds | 5–15 whole seconds |
| Prompt length | Up to 7,000 characters | Up to 7,000 characters |
| Text to video | Supported | Supported |
| Frame images | First frame, optionally a last frame | First and last frame roles |
| Reference images, video and audio | Not supported | Supported in reference mode |
| Prompt expansion | disabled, balanced, quality | disabled, balanced, quality |
| Billing dimensions | Output seconds | Output seconds, reference-video seconds, images beyond two |
The native-style Turbo limits also appear in XRToken's own request documentation and AI Video API's Turbo guide; their authentication, defaults and response envelopes differ from SeedRouter's.\[2]\[3] Use our [Turbo API reference](https://seedrouter.ai/docs/minimax-h3-max-turbo) for the exact request rather than copying a different service's wrapper.
## A real MiniMax H3 Max Turbo sample
We generated the following clip on October 20, 2026, with text alone, 768P, six requested seconds, a 16:9 ratio and quality prompt expansion. An adult dancer performs a short hip-hop sequence on a rooftop at sunset. The downloaded output is 1344 × 768 pixels and contains an AAC audio track. Its MP4 container measures 6.592 seconds; reported generation usage is six output seconds.
The exact prompt was:
```text
A premium street-dance music-video shot, one continuous six-second take. A beautiful 25-year-old adult female professional dancer with a dark high ponytail performs an energetic hip-hop groove on a broad rooftop terrace at sunset, well away from its edges. She wears a white fitted T-shirt, an open cobalt-blue bomber jacket, loose white dance trousers and white sneakers. She starts dancing immediately: a quick lateral shuffle, a precise shoulder-and-arm isolation, then a smooth half-turn and two steps toward the camera, finishing with a confident smile. Keep her full body and both feet visible, with the dancer filling most of the frame height. The camera tracks backward smoothly as she approaches; the skyline stays soft behind her. Golden backlight outlines her moving hair and jacket, with soft frontal fill keeping her face clearly visible. Natural skin texture, stable identity, coherent hands and feet, believable balance and continuous foot contact with the terrace. Her clothing follows her motion and a light breeze. Original upbeat hip-hop drums and bass, synchronized to the choreography, no vocals. No cuts, no text, no logos, no other people.
```
The sample verifies one text-to-video request at those settings. It does not verify frame-guided behavior, each prompt-expansion mode, or every output resolution. Our [Max samples](https://seedrouter.ai/models/minimax-h3-max) used a different prompt, so these clips are not a controlled speed or quality comparison. Listen to the output before deciding whether its sound fits an edit; the presence of an audio track alone does not establish sound accuracy.
## Two more dance briefs, different visual demands
These additional clips use the same model, 768P, six requested seconds, 16:9 and quality prompt expansion. They are independent text-to-video generations, with different choreography, costumes and lighting. Each downloaded MP4 is 1344 × 768 with an AAC track and a 6.592-second container duration; reported billable generation duration is six seconds.
### Silver-jacket pop choreography on a neon stage
View the exact submitted prompt
```text
An eye-catching pop dance music video, one continuous six-second shot. A beautiful 25-year-old adult East Asian female professional dancer with expressive eyes and long dark hair performs a precise, energetic pop choreography on a spacious neon-lit stage. She wears a reflective silver cropped jacket over a fitted black top, high-waisted black dance shorts and black ankle dance boots. Soft white key light keeps her face clear; magenta and electric-blue stage lights make a vivid background. Start dancing immediately: two snappy side steps, one coordinated shoulder-and-arm accent and a controlled half-turn, finishing toward the camera with a playful confident smile. Keep the full body and both feet visible, occupying most of the frame height. Smooth lateral camera tracking follows her without distracting zooms. Her hair and silver jacket move naturally with the dance. Photorealistic skin, stable face, coherent fingers and limbs, grounded footwork. Original upbeat electronic drums and bass, no vocals. No cuts, no text, no logos, no other people.
```
### Scarlet-dress jazz turns in a sunlit studio
View the exact submitted prompt
```text
A polished fashion dance campaign, one continuous six-second take. A beautiful 25-year-old adult female professional dancer with glossy shoulder-length dark hair performs an expressive jazz dance solo in a bright spacious ivory studio. She wears an elegant scarlet-red sleeveless dance dress with a flared skirt and closed-toe red dance heels. Warm side light creates a clear face and a long soft floor shadow; the red dress is the strong visual focus against the pale background. She moves from the first frame: two crisp diagonal steps with sweeping arms, one smooth turn that opens the skirt, then a confident final pose with a direct look and a smile toward the lens. Keep her entire body and feet visible, filling most of the frame height. The camera tracks in a gentle shallow arc, maintaining distance and fluid continuity. Natural skin texture, stable facial identity, coherent hands and legs, believable weight shifts and fabric movement. Original lively jazz-funk rhythm with drums and bass, no vocals. No cuts, no text, no logos, no other people.
```
The silver-jacket scene uses crisp weight shifts and a turn against saturated neon. The red-dress scene emphasizes travelling steps and a flaring skirt in an open studio. Compare the whole sequence, including foot contact and the transition into the final pose, rather than judging only the poster.
## Start with the asset that defines the shot
For a dance scene, decide what must remain fixed. If you need a newly imagined performer and setting, text can describe the costume, choreography and camera movement. A first-frame photograph becomes more useful when the opening composition needs to match an existing character image. Add a last frame when the shot must end at another specified composition, and supply the opening image with it.
That is the point at which Turbo and Max stop being interchangeable. A photograph used as a **first frame** establishes the opening of the clip. A **reference image** guides appearance without necessarily becoming the opening frame. Turbo's frame inputs do not stand in for Max's reference mode. Nor can a motion-reference video be converted into a first frame while preserving the same instruction.
If the brief says “follow this camera move” and the source is a video, evaluate Max. If it says “begin on this character photograph and animate the pose,” Turbo has the relevant frame-based controls. In either case, inspect the performer's face, hands, foot placement and clothing across the full take. A good opening still does not establish that the character remains consistent during motion.
## Compare the cost of the whole request
Turbo's charge is the generated output duration multiplied by the selected resolution's rate. A first frame, a first-and-last-frame pair, and prompt expansion do not add separate input charges. The live estimate changes when you change output duration or resolution.
Max separates output-video and reference-video seconds. It includes the first two image inputs, then charges for later images; reference audio has no separate input charge. A Max request containing several reference images and a video therefore needs more than the output-second rate to estimate its cost.\[4]
### MiniMax H3 Max Turbo
MiniMax H3 Max Turbo charges generated output seconds at the selected resolution. First and last frames and prompt expansion have no separate charge. These are live SeedRouter rates.
| Resolution | USD/output second |
| --- | --- |
| 480P | $0.0228515625 |
| 768P | $0.0365625 |
Failed tasks are not charged. Reference images, video and audio are not supported.
### MiniMax H3 Max
MiniMax H3 Max prices output and reference video separately, using output resolution. The first two input images are free. Reference audio has no input charge. These are live SeedRouter rates.
| Resolution | USD/output second | USD/reference-video second | USD/extra image |
| --- | --- | --- | --- |
| 480P | $0.045703125 | $0.05054765625 | $0.0661375 |
| 768P | $0.073125 | $0.1307109375 | $0.0661375 |
Final charges use reported output seconds, input-video seconds and images beyond two, at the rate captured for the request. Failed tasks are not charged.
The tables above show current SeedRouter rates for both models. For a text-only five-second shot, compare five times the output rate at the same resolution. For Max reference mode, add the corresponding reference-video and extra-image charges. Also budget for the number of attempts: every new successful render is a separate paid request, even if you make only a small prompt change. Failed tasks are not charged.
Container duration and billed duration can differ slightly, as the six-second dance sample illustrates. Use the estimate before submission and account usage after completion. Trimming a finished clip in your editor does not reduce the amount already generated.
## Pick resolution and duration before polishing the prompt
Both models in this comparison offer 480P and 768P. Start by deciding where the clip will appear: a small preview, a full-width showcase or a vertical social edit. Inspect at that intended size. More pixels can make texture easier to see, but they do not guarantee an accurate silhouette, stable lettering or a successful camera move.
Both accept whole-second durations from five through fifteen seconds on this interface. If the final edit needs four seconds, generate an allowed duration and trim afterward. Keep room for the action to begin and settle. Several competing actions in a short dance clip are harder to assess than a clear sequence of steps and a single turn.
For a useful comparison, hold the prompt, ratio and duration constant, then change one setting. Note what improved and what changed unexpectedly. The three samples show actual outputs, but they cannot answer which model will be best for every scene.
## Prompt expansion and audio
The expansion choices are `disabled`, `balanced` and `quality`, with balanced as the default. Disabled is appropriate when you want to submit a detailed brief without prompt expansion. Changing this option can change interpretation; the word quality names a mode rather than promising that a result will meet your acceptance criteria.
Turbo has no target-soundtrack or reference-audio field on this interface. Describe intended sound in the prompt, then inspect the generated track, or add your own recording during editing. Max accepts reference audio in reference mode, alongside image or video context. Audio used as guidance is not the same promise as placing an exact supplied recording in the final soundtrack.
## Avoid mixing contracts from different services
A search for “MiniMax H3 Max Turbo” can return fal's adapted model, native-style integrations and community workflows. Their names overlap; their schemas do not. Fal's text and image documentation includes controls outside the native-style interface used here.\[1] A request with its fractional duration or target-audio parameter will be rejected by SeedRouter rather than silently reinterpreted.
The same caution applies to seed controls. Some native-style services document a seed, while this interface does not expose one. Do not use a seed field from another guide and assume it makes results reproducible. Check the actual endpoint's parameter list, its allowed values and its media rules before adapting a working request.
## Frequently asked questions
### Is MiniMax H3 Max Turbo faster than Max?
There is no controlled timing comparison in this article. One completed request cannot establish typical latency or a speed multiplier. Compare repeated runs under equivalent settings before making a delivery-time commitment.
### Can I switch by changing only the model ID?
A text-only request using shared resolution, duration and ratio settings can retain its general shape. A Max request with reference images, video or audio cannot be moved to Turbo unchanged. Read the [Turbo request fields](https://seedrouter.ai/docs/minimax-h3-max-turbo) and remove unsupported assumptions before submitting.
### Does Turbo support last-frame-only generation?
Not on this interface. Provide a first frame when adding a last frame. If the opening composition is unimportant, text-only generation may be simpler than preparing an arbitrary opening image.
### Which should I try for a product reveal?
For a written scene or a reveal beginning with a product photograph, start with the [Turbo Playground](https://seedrouter.ai/models/minimax-h3-max-turbo). If reference footage, multiple appearance references or audio guidance defines the brief, inspect the [Max Playground](https://seedrouter.ai/models/minimax-h3-max). Choose the input mode first, then compare a finished take against the visual details your project requires.
## Sources
\[1]
fal: H3 Max Turbo text-to-video documentation
. A separate hosted interface; checked October 20, 2026. Its fields are not the SeedRouter contract.
\[2]
XRToken: MiniMax H3 request parameters
. Its native-style Max and Max-Turbo rules distinguish frame inputs from reference mode; checked October 20, 2026.
\[3]
AI Video API: MiniMax H3 Max Turbo
. Its Turbo-specific input limits and output-only billing; checked October 20, 2026. Its wrapper fields and defaults differ from SeedRouter.
\[4]
MiniMax: Create Video Generation Task
and
MiniMax pricing
. Max request and billing dimensions; checked October 20, 2026. SeedRouter prices above come from the live pricing component.