Motion and physical detail
Track a white horse running through shallow water, capturing splashes and a coherent gait in morning backlight. Keep physical motion natural.
Direct continuous movement, physical detail and lighting together.No matching models
Explore all modelsCreate videos with native audio using MiniMax H3. Start with text, first/last frames, or combined image, video and audio references. Available resolutions: 480P / 768P / 2K. 768P / 2K support 4–15 seconds; 480P starts at 5 seconds.
{
"model": "minimax-h3",
"input": {
"seconds": 5,
"resolution": "480P",
"aspect_ratio": "16:9",
"aigc_watermark": false,
"prompt": "A breathtaking photorealistic science-fiction film shot for a premium AI technology platform. An immense elegant circular orbital observatory of brushed titanium and dark glass floats above the luminous blue curvature of a planet. A precisely engineered ring, fine panel seams, small warm windows and slender structural ribs; believable physical construction. The camera makes one slow smooth lateral orbit with subtle forward movement, creating clear parallax between the crisp station in the foreground and the planet below. A distant sun slowly emerges from behind the edge of the station, tracing a thin silver rim across the metal; clouds drift far below. Rich ink-blue space, delicate violet nebula wisps, restrained ice-blue atmosphere, silver and tiny warm highlights. Immediately compelling and clearly legible from the first frame. The entire main ring stays inside the central 70 percent of the frame with generous breathing room. Extremely crisp detailed metal, physically accurate light, controlled highlights, deep rich contrast, cinematic large-format photography. One continuous six-second shot, stable geometry, measured graceful motion. No cuts, no titles, no lettering, no logo, no watermark, no HUD, no interface, no particles flying toward the camera, no strobing, no explosive action, no excessive lens flare, no heavy motion blur. Landscape 16:9 composition."
}
}The example follows your Playground parameters.
curl --request POST 'https://newrouters.com/v1/tasks' \
--header 'Content-Type: application/json' \
--header "Authorization: Bearer $MODEL_API_KEY" \
--data '{
"model": "minimax-h3",
"input": {
"seconds": 5,
"resolution": "480P",
"aspect_ratio": "16:9",
"aigc_watermark": false,
"prompt": "A breathtaking photorealistic science-fiction film shot for a premium AI technology platform. An immense elegant circular orbital observatory of brushed titanium and dark glass floats above the luminous blue curvature of a planet. A precisely engineered ring, fine panel seams, small warm windows and slender structural ribs; believable physical construction. The camera makes one slow smooth lateral orbit with subtle forward movement, creating clear parallax between the crisp station in the foreground and the planet below. A distant sun slowly emerges from behind the edge of the station, tracing a thin silver rim across the metal; clouds drift far below. Rich ink-blue space, delicate violet nebula wisps, restrained ice-blue atmosphere, silver and tiny warm highlights. Immediately compelling and clearly legible from the first frame. The entire main ring stays inside the central 70 percent of the frame with generous breathing room. Extremely crisp detailed metal, physically accurate light, controlled highlights, deep rich contrast, cinematic large-format photography. One continuous six-second shot, stable geometry, measured graceful motion. No cuts, no titles, no lettering, no logo, no watermark, no HUD, no interface, no particles flying toward the camera, no strobing, no explosive action, no excessive lens flare, no heavy motion blur. Landscape 16:9 composition."
}
}'Set MODEL_API_KEY in your environment. Keep your secret key on the server. Full documentation ↗
Estimate one request, then choose a budget for your workflow.
| Usage | Quantity | Rate | Cost |
|---|---|---|---|
| 768P | 5 sec | $0.1/ sec | $0.5 |
Public USD estimate. Your account rate and final recorded usage determine the actual charge.
| Specification | Our price | Official reference |
|---|---|---|
| 768P | $0.1 / second | — |
| 2K | $0.12 / second | — |
| 480P | $0.08 / second | — |
Capability-based prompts and platform-generated examples are labeled separately, with their sources.
Official capability source ↗ · 2026-10-14
Track a white horse running through shallow water, capturing splashes and a coherent gait in morning backlight. Keep physical motion natural.
Direct continuous movement, physical detail and lighting together.Use the supplied scene and character references for a clip of the character entering a courtyard and turning toward the camera. Preserve design consistency with a slow pullback.
Use the model’s supported reference mode and specify the traits that should remain consistent.A breathtaking photorealistic science-fiction film shot for a premium AI technology platform. An immense elegant circular orbital observatory of brushed titanium and dark glass floats above the luminous blue curvature of a planet. A precisely engineered ring, fine panel seams, small warm windows and slender structural ribs; believable physical construction. The camera makes one slow smooth lateral orbit with subtle forward movement, creating clear parallax between the crisp station in the foreground and the planet below. A distant sun slowly emerges from behind the edge of the station, tracing a thin silver rim across the metal; clouds drift far below. Rich ink-blue space, delicate violet nebula wisps, restrained ice-blue atmosphere, silver and tiny warm highlights. Immediately compelling and clearly legible from the first frame. The entire main ring stays inside the central 70 percent of the frame with generous breathing room. Extremely crisp detailed metal, physically accurate light, controlled highlights, deep rich contrast, cinematic large-format photography. One continuous six-second shot, stable geometry, measured graceful motion. No cuts, no titles, no lettering, no logo, no watermark, no HUD, no interface, no particles flying toward the camera, no strobing, no explosive action, no excessive lens flare, no heavy motion blur. Landscape 16:9 composition.
MiniMax H3 combines text, image, video and audio context for generation, with native stereo audio, reference-guided creation and motion transfer.
Official model information ↗ · 2026-10-14
Describe the scene, subject movement and camera direction. Select the duration, resolution and aspect ratio.
Upload one image to define the opening shot. The prompt directs the movement that follows; the output ratio follows this image.
Provide two images in order: start, then end. The model generates the motion connecting these frames.
Use up to 9 images to guide subjects, composition and style. References guide the content without fixing the first frame.
Use up to 3 clips to guide motion and camera behavior; combine them with image references.
Use up to 3 audio clips to guide sound. At 480P, also supply an image or video reference.
| Model ID | minimax-h3 |
|---|---|
| Provider | MiniMax |
| API | /v1/tasks |
| Resolution | 480P / 768P / 2K |
| Aspect ratio | 21:9 / 16:9 / 4:3 / 1:1 / 3:4 / 9:16 |
| Duration | 4–15 seconds |
| Reference images | Up to 9 |
| Frame images | Up to 2 |
| Reference videos | Up to 3 |
| Reference audio | Up to 3 |
For short storyboards or product motion studies, define duration, resolution and composition first. Add references or frame inputs only when the current request schema supports them; review motion continuity, subject consistency and delivery format.
These are acceptance checks, not guarantees of output quality or speed. A resolution label is not a fixed-pixel promise. Use the current contract and account pricing for supported inputs, combinations and cost; validate an approved small sample before batching.
Node.js video task tutorial · Batch budgeting and cost reconciliation
| Specification | MiniMax H3 | MiniMax H3 Max Turbo |
|---|---|---|
| Provider | MiniMax | MiniMax |
| Input → output | TextImageVideoAudioVideo | TextImageVideo |
| Start / end frames | Supported · 1 or 2 images | Supported · 1 or 2 images |
| Reference videos | 3 | — |
| Reference audio | 3 | — |
| Selection basis | Current model | Both declare text, image input · Same billing units |
| Resolution | 480P / 768P / 2K | 480P / 768P |
| Reference images | 9 | — |
| Starting price | $0.08 / second | $0.05 / second |
Choose by the input types, output settings and billing unit your workflow needs. Try the same prompt across models to compare results for your own use case.
Submit parameters through the Media Tasks API, then retrieve the result with the task ID. Image and video models keep their own input requirements.
Sign in and prepare a key with access to this model.
Choose parameters in Playground and copy the API example.
Review status, outputs and charges in History.
POST /v1/tasks · Authorization: Bearer
curl 'https://newrouters.com/v1/tasks' \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $MODEL_API_KEY" \
--data '{
"model": "minimax-h3",
"input": {
"seconds": 5,
"resolution": "480P",
"aspect_ratio": "16:9",
"aigc_watermark": false,
"prompt": "A quiet mountain lake at sunrise, soft morning light."
}
}'const response = await fetch("https://newrouters.com/v1/tasks", {
method: "POST",
headers: {
"Content-Type": "application/json",
"Authorization": "Bearer $MODEL_API_KEY"
},
body: JSON.stringify({
"model": "minimax-h3",
"input": {
"seconds": 5,
"resolution": "480P",
"aspect_ratio": "16:9",
"aigc_watermark": false,
"prompt": "A quiet mountain lake at sunrise, soft morning light."
}
})
});
if (!response.ok) throw new Error(await response.text());
console.log(await response.text());import json
import requests
response = requests.post(
"https://newrouters.com/v1/tasks",
headers=json.loads("{\"Content-Type\":\"application/json\",\"Authorization\":\"Bearer $MODEL_API_KEY\"}"),
json=json.loads("{\"model\":\"minimax-h3\",\"input\":{\"seconds\":5,\"resolution\":\"480P\",\"aspect_ratio\":\"16:9\",\"aigc_watermark\":false,\"prompt\":\"A quiet mountain lake at sunrise, soft morning light.\"}}"),
timeout=300,
)
response.raise_for_status()
print(response.text)Poll the returned task ID for processing status and output assets.
curl -H "Authorization: Bearer $MODEL_API_KEY" "https://newrouters.com/v1/tasks/$TASK_ID"Documentation: Authentication · Task status and results · Error handling
| Parameter | Type | Values & constraints |
|---|---|---|
audios | array | |
images | array | |
prompt * | string | |
videos | array | |
seconds | integer | Default: 5 |
resolution | JSON | "480P" · "768P" · "2K"Default: "480P" |
aspect_ratio | JSON | "21:9" · "16:9" · "4:3" · "1:1" · "3:4" · "9:16"Default: "16:9" |
frames_images | array | |
aigc_watermark | boolean | Default: false |
{
"type": "object",
"allOf": [
{
"not": {
"required": [
"frames_images",
"images"
]
}
},
{
"not": {
"required": [
"frames_images",
"videos"
]
}
},
{
"not": {
"required": [
"frames_images",
"audios"
]
}
}
],
"required": [
"prompt"
],
"properties": {
"audios": {
"type": "array",
"items": {
"type": "string",
"minLength": 1
},
"maxItems": 3,
"minItems": 1
},
"images": {
"type": "array",
"items": {
"type": "string",
"minLength": 1
},
"maxItems": 9,
"minItems": 1
},
"prompt": {
"type": "string",
"maxLength": 7000,
"minLength": 1
},
"videos": {
"type": "array",
"items": {
"type": "string",
"minLength": 1
},
"maxItems": 3,
"minItems": 1
},
"seconds": {
"type": "integer",
"default": 5,
"maximum": 15,
"minimum": 4
},
"resolution": {
"enum": [
"480P",
"768P",
"2K"
],
"default": "480P"
},
"aspect_ratio": {
"enum": [
"21:9",
"16:9",
"4:3",
"1:1",
"3:4",
"9:16"
],
"default": "16:9"
},
"frames_images": {
"type": "array",
"items": {
"type": "string",
"minLength": 1
},
"maxItems": 2,
"minItems": 1
},
"aigc_watermark": {
"type": "boolean",
"default": false
}
},
"additionalProperties": false
}Answers about pricing, inputs, integration and results.
The current public price starts at $0.08 / second. The pricing section lists each specification. Playground checks your account price before submitting; the completed request records the final charge.
Sign in and open Quick Start above the workspace. Select or create a key with access to this model, then copy the request example.
Supported input fields: audios, images, prompt, videos, seconds, resolution, aspect_ratio, frames_images, aigc_watermark. The parameter table includes allowed values and defaults.
Yes. Use the reference mode to upload images or supply their URLs. The current limit is 9 images. Reference images guide a new generation; they do not imply mask editing support.
Save the returned task ID and query the task until it succeeds or fails. The History tab lets you reopen your tasks and reuse their parameters.
The model keeps its page and identity, but new generation requests are blocked. Existing task records remain accessible. An upstream outage is reported separately from an administrator closing the model.
Create from a prompt and reference images.
Choose a model ↗Explore text, references and frame inputs.
Choose a model ↗Integrate conversations, instructions and streaming.
Choose a model ↗Browse endpoints, request examples and integration steps.
Read documentation ↗Hello, API.Keep the task ID, handle failures and local deadlines, then save outputs.
View docs section ↗/v1/tasksDistinguish per-request and per-second rates; bound concurrency and retries.
View docs section ↗{ your next idea }From choosing inputs to reviewing outputs and charges.
View docs section ↗