Does MiniMax H3 support text-to-video generation?

Yes. MiniMax H3 supports multimodal video generation using text, image, video, and audio inputs. It can generate videos from natural language instructions while maintaining consistency across visual and audio elements.

What makes MiniMax H3 different from traditional AI video models?

MiniMax H3 combines text, images, video, and audio into a unified multimodal generation system. It supports native stereo audio generation, video editing, and motion transfer instead of only prompt-based video creation.

How long and what resolution videos can MiniMax H3 generate?

MiniMax H3 can generate videos up to 15 seconds with resolutions up to 2K according to MiniMax's public release information.

Can MiniMax H3 perform image-to-video generation?

Yes. MiniMax H3 supports image-based video generation workflows, allowing users to provide reference images and generate animated video sequences while preserving subject identity.

How does MiniMax H3 compare with Sora 2 and Seedance 2.0?

MiniMax H3 focuses on controllable multimodal generation, native audio, and open-weight availability. Sora emphasizes cinematic simulation quality, while Seedance focuses on creative video generation and reference-based workflows.

Is MiniMax H3 suitable for commercial video production?

Yes. MiniMax H3 is designed for advertising, e-commerce, product visualization, gaming content, and AI-assisted filmmaking workflows.

Does MiniMax H3 provide an API for developers?

MiniMax provides video generation APIs with asynchronous task creation, status querying, and file retrieval workflows. H3-specific API availability depends on provider rollout.

Can developers integrate MiniMax H3 through CometAPI?

Yes. CometAPI provides a unified API access layer that allows developers to integrate AI models through a consistent interface and manage video generation workflows more easily.

MiniMax

Available

minimax-h3

MiniMax H3 API

minimax-h3 is a new video generation model designed for high-quality creative production. It delivers improved prompt understanding, stronger visual consistency, smoother motion, and more flexible reference-image control for cinematic video generation workflows.

Starting price

$0.1 / second

Usage-based billing

API health

Available

Production endpoint

Response

Async ready

Observed API response

Released

Aug 6, 2026

MiniMax

Overview

A practical overview of MiniMax H3, including important behavior and integration considerations.

Technical Specifications of MiniMax H3

Item	MiniMax H3
Model family	MiniMax H Series
Model type	Multimodal video generation model
Provider	MiniMax
Input modalities	Text, Image, Video, Audio
Output	Generated video with audio
Maximum video resolution	Up to 2K
Maximum duration	Up to 15 seconds
Audio generation	Native stereo sound
Video capabilities	Text-to-video, image-to-video, video editing, motion transfer
Weight availability	Planned open-weight release
Main applications	Advertising, e-commerce, product design, gaming, creative production

Source: MiniMax announcement and technical descriptions.

What is MiniMax H3?

MiniMax H3 is a multimodal video generation foundation model developed by MiniMax. Unlike traditional text-to-video systems that only transform prompts into clips, H3 focuses on director-level controllability, combining multiple input formats including text instructions, reference images, existing video clips, and audio signals.

The model is designed for professional creative workflows such as commercial video production, game asset creation, advertising generation, and AI-assisted filmmaking. It supports generating videos up to 15 seconds at 2K resolution with synchronized stereo audio.

Main Features of MiniMax H3

1. Multimodal Video Understanding and Generation

MiniMax H3 accepts multiple forms of creative input:

Text descriptions
Reference images
Existing video materials
Audio references

This allows users to create more controllable video outputs compared with traditional prompt-only generation systems.

2. Native Audio-Visual Generation

Unlike many video generation models that require separate audio synthesis pipelines, H3 supports native stereo sound generation.

This enables:

Dialogue-driven scenes
Environmental sound effects
Cinematic atmosphere
Better audio-video synchronization

3. Video Editing and Motion Transfer

H3 supports editing existing content and transferring motion characteristics between videos.

Example workflows:

Transfer human movement from reference footage
Modify existing video scenes
Apply motion styles
Generate variations of existing clips

4. Open-Weight Deployment Potential

MiniMax announced plans to release H3 model weights, allowing researchers and developers to customize the model rather than relying only on hosted inference APIs.

Community testing has already explored local deployment workflows through ComfyUI and optimized inference pipelines.

5. High-Resolution Video Generation

MiniMax H3 supports generation up to 2K resolution, targeting professional content production workflows.

Compared with earlier consumer AI video tools, higher resolution enables:

Brand advertisements
Product marketing
Film previsualization
Professional social media content

MiniMax H3 Benchmark Performance

Currently, MiniMax has not published standardized benchmark results comparable to:

VBench
MovieGenBench
EvalCrafter
Video-MME

Therefore, there are no verified public scores available.

However, MiniMax reports several capability advantages:

Capability	MiniMax H3
2K video generation	Supported
Native audio	Supported
Text-to-video	Supported
Image-to-video	Supported
Video editing	Supported
Motion transfer	Supported

MiniMax H3 vs Sora vs Veo vs Kling

Model	Provider	Main Strength	Best For
MiniMax H3	MiniMax	Multimodal video + audio + editing	Commercial video creation
Sora	OpenAI	Cinematic realism and world simulation	Creative filmmaking
Veo	Google	High-quality video understanding	Enterprise media generation
Kling	Kuaishou	Character motion and realism	Short videos/social content

MiniMax H3 vs Sora

MiniMax H3 focuses more on:

Open ecosystem
Commercial generation
Multimodal control
Video editing workflows

Sora focuses more on:

Physical simulation
Cinematic consistency
Complex scene generation

MiniMax H3 vs Kling

Compared with Kling:

MiniMax H3 emphasizes:

Audio integration
Multimodal inputs
Professional workflow support

Kling has strong advantages in:

Human motion
Character animation
Consumer video creation

MiniMax H3 Use Cases

AI Advertisement Production

Marketing teams can generate:

Product commercials
Brand videos
Social media advertisements

Workflow:

Product image → Prompt → Marketing video

E-commerce Video Generation

Online sellers can automatically create:

Product showcases
Lifestyle scenes
Promotional clips

Game Development

MiniMax H3 can assist with:

Character animation
Cinematic trailers
Concept visualization

Film Previsualization

Directors can quickly test:

Camera angles
Scene design
Storyboards

How to Use MiniMax H3 API with CometAPI

CometAPI provides a unified API access layer for AI models, allowing developers to integrate emerging video generation models without maintaining multiple provider SDKs.

Using MiniMax H3 through CometAPI follows the same workflow as other video generation APIs:

Step 1: Get CometAPI API Key

Register an account on CometAPI.
Navigate to API Console.
Create an API key.
Use the generated key for authentication.

Example:

Authorization: Bearer YOUR_COMETAPI_API_KEY

Step 2: Send MiniMax H3 Video Generation Request

MiniMax-style video generation APIs typically use an asynchronous task architecture:

Submit generation request.
Receive task_id.
Query generation status.
Retrieve generated video file.

MiniMax official video APIs follow this workflow pattern.

Example request refer to the MiniMax H3 API section for details.

Step 3: Query Video Generation Status

After submitting the task:

{
  "task_id": "xxxxxxxx",
  "status": "processing"
}

Poll the task status until completion.

Successful response:

{
  "status": "success",
  "file_id": "video_xxxxx"
}

Step 4: Retrieve Generated Video

The completed video can then be accessed through the returned file ID.

Why Use MiniMax H3 API Through CometAPI?

1. Unified Access to Multiple AI Video Models

Developers can integrate MiniMax H3 alongside other video models:

Seedance
Kling
Sora
Veo
Runway

without maintaining separate API implementations.

2. Simplified API Management

Instead of handling:

Different authentication systems
Different request formats
Different SDKs

developers use one API interface.

3. Faster Model Switching

AI video quality changes rapidly.

CometAPI allows applications to experiment with different generation models without rebuilding the entire production pipeline.

4. Developer-Friendly Workflow

Typical applications:

AI video SaaS platforms
Marketing automation tools
E-commerce content generators
AI filmmaking tools
Social media video generators

FAQ

Pricing

Use the quick estimate above for a single run, then review the full CometAPI and official-price comparison.

5-second test

$0.5

Estimated CometAPI usage cost

10-second test

$1.00

Estimated CometAPI usage cost

15-second test

$1.50

Estimated CometAPI usage cost

Pricing for MiniMax H3

Explore competitive pricing for MiniMax H3, designed to fit various budgets and usage needs. Our flexible plans ensure you only pay for what you use, making it easy to scale as your requirements grow. Discover how MiniMax H3 can enhance your projects while keeping costs manageable.

Resolution	Comet Price (USD / Second)
768p	Per Second: $0.08
2K	Per Second: $0.13

API

Copy a working endpoint and code example, then open the complete API reference when you need every parameter.

One API, production-ready model access

Authenticate once, call the model endpoint and keep the same billing and observability workflow across providers.

POST/v1/videosGET/v1/videos/{task_id}/contentGET/v1/videos/{task_id}

Sample code and API for MiniMax H3

Access comprehensive sample code and API resources for MiniMax H3 to streamline your integration process. Our detailed documentation provides step-by-step guidance, helping you leverage the full potential of MiniMax H3 in your projects.

POST

/v1/videos

GET

/v1/videos/{task_id}/content

GET

/v1/videos/{task_id}

#!/usr/bin/env bash
set -euo pipefail

# 1. Create the task.
task=$(curl --fail-with-body --silent --show-error \
  https://api.cometapi.com/v1/videos \
  -H "Authorization: Bearer $COMETAPI_KEY" \
  --form-string 'model=minimax-h3' \
  --form-string 'prompt=A cinematic view of clouds moving over green mountains, locked camera, no text.' \
  --form-string 'seconds=5' \
  --form-string 'size=1344x768')
# Add --form-string 'images=https://example.com/reference.jpg' for image-to-video.

task_id=$(jq -r '.id' <<<"$task")
echo "Task: ${task_id}"

# 2. Poll until the task finishes.
while true; do
  task=$(curl --fail-with-body --silent --show-error \
    --retry 3 --retry-all-errors --retry-delay 2 \
    "https://api.cometapi.com/v1/videos/${task_id}" \
    -H "Authorization: Bearer $COMETAPI_KEY")
  status=$(jq -r '.status' <<<"$task")
  echo "Status: $status"
  [[ "$status" == "failed" ]] && { echo "$task" >&2; exit 1; }
  [[ "$status" == "completed" ]] && break
  sleep 10
done

# 3. Download the completed video.
curl --fail-with-body --silent --show-error \
  --retry 3 --retry-all-errors --retry-delay 2 \
  "https://api.cometapi.com/v1/videos/${task_id}/content" \
  -H "Authorization: Bearer $COMETAPI_KEY" \
  --output "${task_id}.mp4"
echo "Saved: ${task_id}.mp4"

cURL Code Example

#!/usr/bin/env bash
set -euo pipefail

# 1. Create the task.
task=$(curl --fail-with-body --silent --show-error \
  https://api.cometapi.com/v1/videos \
  -H "Authorization: Bearer $COMETAPI_KEY" \
  --form-string 'model=minimax-h3' \
  --form-string 'prompt=A cinematic view of clouds moving over green mountains, locked camera, no text.' \
  --form-string 'seconds=5' \
  --form-string 'size=1344x768')
# Add --form-string 'images=https://example.com/reference.jpg' for image-to-video.

task_id=$(jq -r '.id' <<<"$task")
echo "Task: ${task_id}"

# 2. Poll until the task finishes.
while true; do
  task=$(curl --fail-with-body --silent --show-error \
    --retry 3 --retry-all-errors --retry-delay 2 \
    "https://api.cometapi.com/v1/videos/${task_id}" \
    -H "Authorization: Bearer $COMETAPI_KEY")
  status=$(jq -r '.status' <<<"$task")
  echo "Status: $status"
  [[ "$status" == "failed" ]] && { echo "$task" >&2; exit 1; }
  [[ "$status" == "completed" ]] && break
  sleep 10
done

# 3. Download the completed video.
curl --fail-with-body --silent --show-error \
  --retry 3 --retry-all-errors --retry-delay 2 \
  "https://api.cometapi.com/v1/videos/${task_id}/content" \
  -H "Authorization: Bearer $COMETAPI_KEY" \
  --output "${task_id}.mp4"
echo "Saved: ${task_id}.mp4"

Python Code Example

import os
import time
from pathlib import Path

import requests


api_key = os.environ["COMETAPI_KEY"]
base_url = "https://api.cometapi.com"
headers = {"Authorization": f"Bearer {api_key}"}


def get(path: str, timeout: int) -> requests.Response:
    for attempt in range(3):
        try:
            response = requests.get(
                f"{base_url}{path}", headers=headers, timeout=timeout
            )
            response.raise_for_status()
            return response
        except requests.RequestException:
            if attempt == 2:
                raise
            time.sleep(2)
    raise RuntimeError("unreachable")

# 1. Create the task.
form = {
    "model": (None, "minimax-h3"),
    "prompt": (
        None,
        "A cinematic view of clouds moving over green mountains, "
        "locked camera, no text.",
    ),
    "seconds": (None, "5"),
    "size": (None, "1344x768"),
}
if image_url := os.getenv("MINIMAX_H3_IMAGE_URL"):
    form["images"] = (None, image_url)

response = requests.post(
    f"{base_url}/v1/videos",
    headers=headers,
    files=form,
    timeout=60,
)
response.raise_for_status()
task_id = response.json()["id"]
print(f"Task: {task_id}")

# 2. Poll until the task finishes.
while True:
    response = get(f"/v1/videos/{task_id}", timeout=30)
    task = response.json()
    status = task["status"]
    print(f"Status: {status}")
    if status == "failed":
        raise RuntimeError(task)
    if status == "completed":
        break
    time.sleep(10)

# 3. Download the completed video.
response = get(f"/v1/videos/{task_id}/content", timeout=300)
output = Path(f"{task_id}.mp4")
output.write_bytes(response.content)
print(f"Saved: {output}")

JavaScript Code Example

import fs from "node:fs";

const apiKey = process.env.COMETAPI_KEY;
const baseUrl = "https://api.cometapi.com";
const headers = { Authorization: `Bearer ${apiKey}` };
const sleep = (ms) => new Promise((resolve) => setTimeout(resolve, ms));

async function get(path) {
  for (let attempt = 0; attempt < 3; attempt += 1) {
    try {
      const result = await fetch(`${baseUrl}${path}`, { headers });
      if (!result.ok) throw new Error(`${result.status} ${await result.text()}`);
      return result;
    } catch (error) {
      if (attempt === 2) throw error;
      await sleep(2000);
    }
  }
  throw new Error("unreachable");
}

// 1. Create the task.
const form = new FormData();
form.append("model", "minimax-h3");
form.append(
  "prompt",
  "A cinematic view of clouds moving over green mountains, locked camera, no text.",
);
form.append("seconds", "5");
form.append("size", "1344x768");
if (process.env.MINIMAX_H3_IMAGE_URL) {
  form.append("images", process.env.MINIMAX_H3_IMAGE_URL);
}

let response = await fetch(`${baseUrl}/v1/videos`, {
  method: "POST",
  headers,
  body: form,
});
if (!response.ok) throw new Error(`${response.status} ${await response.text()}`);
const taskId = (await response.json()).id;
console.log(`Task: ${taskId}`);

// 2. Poll until the task finishes.
while (true) {
  response = await get(`/v1/videos/${taskId}`);
  const task = await response.json();
  console.log(`Status: ${task.status}`);
  if (task.status === "failed") throw new Error(JSON.stringify(task));
  if (task.status === "completed") break;
  await sleep(10000);
}

// 3. Download the completed video.
response = await get(`/v1/videos/${taskId}/content`);
const output = `${taskId}.mp4`;
fs.writeFileSync(output, Buffer.from(await response.arrayBuffer()));
console.log(`Saved: ${output}`);

Capabilities

Scan the model facts that matter before you choose an architecture or estimate production workload.

Supported capabilities

Features returned by the existing model API.

text-to-video

image-to-video

Input and output

Inputs

text

Outputs

video

Production profile

Provider: MiniMax
Context: Model specific
Commercial use: Supported

Use cases

Use MiniMax H3 for production workflows that match its video capabilities, then compare alternatives before committing to a long-term integration.

Text-to-video concepts and social content

Product demos and cinematic storyboards

Image-to-video motion and creative iteration

Other models

Compare other models available through CometAPI for different quality, latency, capability and pricing trade-offs.

Open model comparison →

Aliyun

Happy Horse 1.1

HappyHorse 1.1 is a multimodal video-generation model designed for professional content creation, advertising, short films, social media production, and storytelling. It extends the capabilities of HappyHorse 1.0—which gained significant attention after ranking highly in independent video-generation evaluations—with stronger scene coherence and improved visual fidelity.

View model

Bytedance

Seedance 2-0

Seedance 2.0 is ByteDance’s next-generation multimodal video foundation model focused on cinematic, multi-shot narrative video generation. Unlike single-shot text-to-video demos, Seedance 2.0 emphasizes reference-based control (images, short clips, audio), coherent character/style consistency across shots, and native audio/video synchronization — aiming to make AI video useful for professional creative and previsualization workflows.

View model

Aliyun

Happy Horse 1.0

Happy Horse 1.0 — A high-quality audio-video generation model that supports text-to-video and image-to-video creation. It can generate synchronized visuals, audio, and lip movements, making it suitable for short films, advertising creatives, and product showcases.

View model

Google

Gemini omni fast

Gemini Omni Fast is a lightweight multimodal video generation model designed for fast and flexible content creation. It enables efficient video generation with support for multiple input types, making it suitable for interactive and iterative workflows.

View model

Vidu

Vidu Q3

Vidu Q3 is a video generation model designed for high-quality content creation with strong visual realism and prompt adherence. It is suitable for creative video production and storytelling applications.

View model

Vidu

Vidu Q3 Turbo

Vidu Q3 Turbo is an optimized version of Vidu Q3 designed for faster video generation. It maintains good visual quality while significantly improving speed, making it suitable for rapid iteration and large-scale production.

View model

Uptime

Review live heartbeat data, endpoint availability and observed response times before moving into production.

Versions

Review available model identifiers before pinning a production integration.

minimax-h3

Keep exploring

Compare models and continue your research

MiniMax

Available

minimax-h3

MiniMax H3 API

Starting price

$0.1 / second

Usage-based billing

API health

Available

Production endpoint

Response

Async ready

Observed API response

Released

Aug 6, 2026

MiniMax

Overview

A practical overview of MiniMax H3, including important behavior and integration considerations.

Technical Specifications of MiniMax H3

Item	MiniMax H3
Model family	MiniMax H Series
Model type	Multimodal video generation model
Provider	MiniMax
Input modalities	Text, Image, Video, Audio
Output	Generated video with audio
Maximum video resolution	Up to 2K
Maximum duration	Up to 15 seconds
Audio generation	Native stereo sound
Video capabilities	Text-to-video, image-to-video, video editing, motion transfer
Weight availability	Planned open-weight release
Main applications	Advertising, e-commerce, product design, gaming, creative production

Source: MiniMax announcement and technical descriptions.

What is MiniMax H3?

Main Features of MiniMax H3

1. Multimodal Video Understanding and Generation

MiniMax H3 accepts multiple forms of creative input:

Text descriptions
Reference images
Existing video materials
Audio references

This allows users to create more controllable video outputs compared with traditional prompt-only generation systems.

2. Native Audio-Visual Generation

Unlike many video generation models that require separate audio synthesis pipelines, H3 supports native stereo sound generation.

This enables:

Dialogue-driven scenes
Environmental sound effects
Cinematic atmosphere
Better audio-video synchronization

3. Video Editing and Motion Transfer

H3 supports editing existing content and transferring motion characteristics between videos.

Example workflows:

Transfer human movement from reference footage
Modify existing video scenes
Apply motion styles
Generate variations of existing clips

4. Open-Weight Deployment Potential

MiniMax announced plans to release H3 model weights, allowing researchers and developers to customize the model rather than relying only on hosted inference APIs.

Community testing has already explored local deployment workflows through ComfyUI and optimized inference pipelines.

5. High-Resolution Video Generation

MiniMax H3 supports generation up to 2K resolution, targeting professional content production workflows.

Compared with earlier consumer AI video tools, higher resolution enables:

Brand advertisements
Product marketing
Film previsualization
Professional social media content

MiniMax H3 Benchmark Performance

Currently, MiniMax has not published standardized benchmark results comparable to:

VBench
MovieGenBench
EvalCrafter
Video-MME

Therefore, there are no verified public scores available.

However, MiniMax reports several capability advantages:

Capability	MiniMax H3
2K video generation	Supported
Native audio	Supported
Text-to-video	Supported
Image-to-video	Supported
Video editing	Supported
Motion transfer	Supported

MiniMax H3 vs Sora vs Veo vs Kling

Model	Provider	Main Strength	Best For
MiniMax H3	MiniMax	Multimodal video + audio + editing	Commercial video creation
Sora	OpenAI	Cinematic realism and world simulation	Creative filmmaking
Veo	Google	High-quality video understanding	Enterprise media generation
Kling	Kuaishou	Character motion and realism	Short videos/social content

MiniMax H3 vs Sora

MiniMax H3 focuses more on:

Open ecosystem
Commercial generation
Multimodal control
Video editing workflows

Sora focuses more on:

Physical simulation
Cinematic consistency
Complex scene generation

MiniMax H3 vs Kling

Compared with Kling:

MiniMax H3 emphasizes:

Audio integration
Multimodal inputs
Professional workflow support

Kling has strong advantages in:

Human motion
Character animation
Consumer video creation

MiniMax H3 Use Cases

AI Advertisement Production

Marketing teams can generate:

Product commercials
Brand videos
Social media advertisements

Workflow:

Product image → Prompt → Marketing video

E-commerce Video Generation

Online sellers can automatically create:

Product showcases
Lifestyle scenes
Promotional clips

Game Development

MiniMax H3 can assist with:

Character animation
Cinematic trailers
Concept visualization

Film Previsualization

Directors can quickly test:

Camera angles
Scene design
Storyboards

How to Use MiniMax H3 API with CometAPI

CometAPI provides a unified API access layer for AI models, allowing developers to integrate emerging video generation models without maintaining multiple provider SDKs.

Using MiniMax H3 through CometAPI follows the same workflow as other video generation APIs:

Step 1: Get CometAPI API Key

Register an account on CometAPI.
Navigate to API Console.
Create an API key.
Use the generated key for authentication.

Example:

Authorization: Bearer YOUR_COMETAPI_API_KEY

Step 2: Send MiniMax H3 Video Generation Request

MiniMax-style video generation APIs typically use an asynchronous task architecture:

Submit generation request.
Receive task_id.
Query generation status.
Retrieve generated video file.

MiniMax official video APIs follow this workflow pattern.

Example request refer to the MiniMax H3 API section for details.

Step 3: Query Video Generation Status

After submitting the task:

{
  "task_id": "xxxxxxxx",
  "status": "processing"
}

Poll the task status until completion.

Successful response:

{
  "status": "success",
  "file_id": "video_xxxxx"
}

Step 4: Retrieve Generated Video

The completed video can then be accessed through the returned file ID.

Why Use MiniMax H3 API Through CometAPI?

1. Unified Access to Multiple AI Video Models

Developers can integrate MiniMax H3 alongside other video models:

Seedance
Kling
Sora
Veo
Runway

without maintaining separate API implementations.

2. Simplified API Management

Instead of handling:

Different authentication systems
Different request formats
Different SDKs

developers use one API interface.

3. Faster Model Switching

AI video quality changes rapidly.

CometAPI allows applications to experiment with different generation models without rebuilding the entire production pipeline.

4. Developer-Friendly Workflow

Typical applications:

AI video SaaS platforms
Marketing automation tools
E-commerce content generators
AI filmmaking tools
Social media video generators

FAQ

Pricing

Use the quick estimate above for a single run, then review the full CometAPI and official-price comparison.

5-second test

$0.5

Estimated CometAPI usage cost

10-second test

$1.00

Estimated CometAPI usage cost

15-second test

$1.50

Estimated CometAPI usage cost

Pricing for MiniMax H3

Resolution	Comet Price (USD / Second)
768p	Per Second: $0.08
2K	Per Second: $0.13

API

Copy a working endpoint and code example, then open the complete API reference when you need every parameter.

One API, production-ready model access

Authenticate once, call the model endpoint and keep the same billing and observability workflow across providers.

POST/v1/videosGET/v1/videos/{task_id}/contentGET/v1/videos/{task_id}

Sample code and API for MiniMax H3

POST

/v1/videos

GET

/v1/videos/{task_id}/content

GET

/v1/videos/{task_id}

#!/usr/bin/env bash
set -euo pipefail

# 1. Create the task.
task=$(curl --fail-with-body --silent --show-error \
  https://api.cometapi.com/v1/videos \
  -H "Authorization: Bearer $COMETAPI_KEY" \
  --form-string 'model=minimax-h3' \
  --form-string 'prompt=A cinematic view of clouds moving over green mountains, locked camera, no text.' \
  --form-string 'seconds=5' \
  --form-string 'size=1344x768')
# Add --form-string 'images=https://example.com/reference.jpg' for image-to-video.

task_id=$(jq -r '.id' <<<"$task")
echo "Task: ${task_id}"

# 2. Poll until the task finishes.
while true; do
  task=$(curl --fail-with-body --silent --show-error \
    --retry 3 --retry-all-errors --retry-delay 2 \
    "https://api.cometapi.com/v1/videos/${task_id}" \
    -H "Authorization: Bearer $COMETAPI_KEY")
  status=$(jq -r '.status' <<<"$task")
  echo "Status: $status"
  [[ "$status" == "failed" ]] && { echo "$task" >&2; exit 1; }
  [[ "$status" == "completed" ]] && break
  sleep 10
done

# 3. Download the completed video.
curl --fail-with-body --silent --show-error \
  --retry 3 --retry-all-errors --retry-delay 2 \
  "https://api.cometapi.com/v1/videos/${task_id}/content" \
  -H "Authorization: Bearer $COMETAPI_KEY" \
  --output "${task_id}.mp4"
echo "Saved: ${task_id}.mp4"

cURL Code Example

#!/usr/bin/env bash
set -euo pipefail

# 1. Create the task.
task=$(curl --fail-with-body --silent --show-error \
  https://api.cometapi.com/v1/videos \
  -H "Authorization: Bearer $COMETAPI_KEY" \
  --form-string 'model=minimax-h3' \
  --form-string 'prompt=A cinematic view of clouds moving over green mountains, locked camera, no text.' \
  --form-string 'seconds=5' \
  --form-string 'size=1344x768')
# Add --form-string 'images=https://example.com/reference.jpg' for image-to-video.

task_id=$(jq -r '.id' <<<"$task")
echo "Task: ${task_id}"

# 2. Poll until the task finishes.
while true; do
  task=$(curl --fail-with-body --silent --show-error \
    --retry 3 --retry-all-errors --retry-delay 2 \
    "https://api.cometapi.com/v1/videos/${task_id}" \
    -H "Authorization: Bearer $COMETAPI_KEY")
  status=$(jq -r '.status' <<<"$task")
  echo "Status: $status"
  [[ "$status" == "failed" ]] && { echo "$task" >&2; exit 1; }
  [[ "$status" == "completed" ]] && break
  sleep 10
done

# 3. Download the completed video.
curl --fail-with-body --silent --show-error \
  --retry 3 --retry-all-errors --retry-delay 2 \
  "https://api.cometapi.com/v1/videos/${task_id}/content" \
  -H "Authorization: Bearer $COMETAPI_KEY" \
  --output "${task_id}.mp4"
echo "Saved: ${task_id}.mp4"

Python Code Example

import os
import time
from pathlib import Path

import requests


api_key = os.environ["COMETAPI_KEY"]
base_url = "https://api.cometapi.com"
headers = {"Authorization": f"Bearer {api_key}"}


def get(path: str, timeout: int) -> requests.Response:
    for attempt in range(3):
        try:
            response = requests.get(
                f"{base_url}{path}", headers=headers, timeout=timeout
            )
            response.raise_for_status()
            return response
        except requests.RequestException:
            if attempt == 2:
                raise
            time.sleep(2)
    raise RuntimeError("unreachable")

# 1. Create the task.
form = {
    "model": (None, "minimax-h3"),
    "prompt": (
        None,
        "A cinematic view of clouds moving over green mountains, "
        "locked camera, no text.",
    ),
    "seconds": (None, "5"),
    "size": (None, "1344x768"),
}
if image_url := os.getenv("MINIMAX_H3_IMAGE_URL"):
    form["images"] = (None, image_url)

response = requests.post(
    f"{base_url}/v1/videos",
    headers=headers,
    files=form,
    timeout=60,
)
response.raise_for_status()
task_id = response.json()["id"]
print(f"Task: {task_id}")

# 2. Poll until the task finishes.
while True:
    response = get(f"/v1/videos/{task_id}", timeout=30)
    task = response.json()
    status = task["status"]
    print(f"Status: {status}")
    if status == "failed":
        raise RuntimeError(task)
    if status == "completed":
        break
    time.sleep(10)

# 3. Download the completed video.
response = get(f"/v1/videos/{task_id}/content", timeout=300)
output = Path(f"{task_id}.mp4")
output.write_bytes(response.content)
print(f"Saved: {output}")

JavaScript Code Example

import fs from "node:fs";

const apiKey = process.env.COMETAPI_KEY;
const baseUrl = "https://api.cometapi.com";
const headers = { Authorization: `Bearer ${apiKey}` };
const sleep = (ms) => new Promise((resolve) => setTimeout(resolve, ms));

async function get(path) {
  for (let attempt = 0; attempt < 3; attempt += 1) {
    try {
      const result = await fetch(`${baseUrl}${path}`, { headers });
      if (!result.ok) throw new Error(`${result.status} ${await result.text()}`);
      return result;
    } catch (error) {
      if (attempt === 2) throw error;
      await sleep(2000);
    }
  }
  throw new Error("unreachable");
}

// 1. Create the task.
const form = new FormData();
form.append("model", "minimax-h3");
form.append(
  "prompt",
  "A cinematic view of clouds moving over green mountains, locked camera, no text.",
);
form.append("seconds", "5");
form.append("size", "1344x768");
if (process.env.MINIMAX_H3_IMAGE_URL) {
  form.append("images", process.env.MINIMAX_H3_IMAGE_URL);
}

let response = await fetch(`${baseUrl}/v1/videos`, {
  method: "POST",
  headers,
  body: form,
});
if (!response.ok) throw new Error(`${response.status} ${await response.text()}`);
const taskId = (await response.json()).id;
console.log(`Task: ${taskId}`);

// 2. Poll until the task finishes.
while (true) {
  response = await get(`/v1/videos/${taskId}`);
  const task = await response.json();
  console.log(`Status: ${task.status}`);
  if (task.status === "failed") throw new Error(JSON.stringify(task));
  if (task.status === "completed") break;
  await sleep(10000);
}

// 3. Download the completed video.
response = await get(`/v1/videos/${taskId}/content`);
const output = `${taskId}.mp4`;
fs.writeFileSync(output, Buffer.from(await response.arrayBuffer()));
console.log(`Saved: ${output}`);