MiniMax H3 AI Video Generator: Text to Video, Image to Video

Generate realistic AI videos with sound with MiniMax H3, from a text prompt, from a photo, or from up to 2 reference images of your product or character. You get a 5- or 10-second MP4 with audio in 16:9, 9:16 or 1:1, in HD (1344 × 768) or SD. It is a text to video and image to video API you can call from the Console, Python, JavaScript, Make, Zapier or n8n, with no GPU, no AI provider account and no subscription. It costs $0.205 per 5-second HD video ($0.133 in SD), about a third less than MiniMax elsewhere on Apify, and videos stopped before generation starts cost nothing.

deAPI AI tools

• Need cheaper clips, or a free try? → LTX-2.5 AI Video Generator: Text to Video & Image to Video
• Need a clean product cutout first? → AI Background Remover API: Remove BG to Transparent PNG
• Photos too small? → AI Image Upscaler 4x: Upscale Images to 4K & 8K in Bulk

What does MiniMax H3 AI Video Generator do?

It is an AI video generator built on the MiniMax H3 video model, known for lifelike motion and detail. For each video you ask for, the Actor:

1. Reads your prompt, and downloads and checks your images if you gave any.
2. Has MiniMax H3 generate the video, picture and sound together, at 24 frames per second.
3. Saves the MP4 to your Apify storage. It comes back as a download link with its resolution, length, seed and what was billed.

There are three modes:

• Text to video. You describe the scene and the model invents it.
• Image to video. Your picture becomes the first frame and the prompt says how it moves. You can also give an end frame, and the video moves from one picture to the other.
• Reference to video. Give up to 2 pictures, for example a product and a person, or a character and a place. The model keeps them recognizable in a new scene that you describe with <Picture 1> and <Picture 2>.

Why use it?

• Realistic motion and detail. MiniMax H3 handles people, animals and products with lifelike movement.
• Keep your product or character. Reference to video puts the same mug, sneaker or mascot into any scene, which suits ads and product videos.
• Video with sound. Ambience, effects and music are generated with the picture.
• HD or SD. HD for final videos; SD costs about 35 % less, for drafts and tests.
• Every format for social. 16:9 for YouTube, 9:16 for TikTok, Reels and Shorts, and 1:1 for feeds.
• Batch up to 10 videos per run, each with its own prompt, images and settings.
• Pay per video, no subscription. You pay when a video starts generating; inputs refused before that cost nothing.

What data can you get?

Data	Details
🎬 Video	A download link (videoUrl) to the MP4 in your run's key-value store.
📐 Format	quality, aspectRatio, width, height, durationSeconds, fps.
🔊 Sound	hasAudio: whether the file has a sound track.
🎲 Seed	The seed used, so you can reproduce the video.
🧭 Mode	text-to-video, image-to-video or reference-to-video, plus the image links you gave.
💵 Billing	billedEvent: the charge for this video.

How to generate a video with MiniMax H3

1. Open the Actor and write a prompt: the subject, what happens, the setting, the camera and the light.
2. Optionally add a start image to animate a picture, or reference images to keep a product or character.
3. Pick the quality (HD or SD), the duration (5 or 10 s) and the aspect ratio.
4. Click Start. A 5-second HD video usually takes about one minute.
5. Open the Output tab and download the MP4.

Writing a good prompt

• Describe one shot, not a story: "a red fox trotting through fresh snow in a pine forest at sunrise".
• Name the camera: "slow push in", "the camera orbits the product", "aerial drone shot".
• Name the light and the mood: "soft golden light", "neon reflections on wet asphalt".
• With reference images, say what each picture does: "<Picture 1> standing on a wooden table in a cozy cafe, rain on the window".

Sizes and lengths

Aspect ratio	HD	SD	For
16:9	1344 × 768	896 × 512	YouTube, websites, presentations
9:16	768 × 1344	512 × 896	TikTok, Instagram Reels, YouTube Shorts
1:1	992 × 992	640 × 640	Instagram and Facebook feeds

Videos are 5 or 10 seconds at 24 fps, with sound, in H.264 MP4. Auto takes the shape of your start (or first reference) image, or 16:9 for text to video.

How fast is it?

In our tests a 5-second HD video took about 70 s, a 5-second SD video 35–50 s, and a 10-second HD video about 2.5 minutes. A run keeps two videos in progress at a time. When many people generate at once, a video can wait in a queue first.

Input

Field	Type	Description
prompt	string	What the video shows. Up to 2,500 characters.
imageUrl	string	Optional start image (image to video): JPG, PNG or WebP, at least 64 px a side, up to 25 MB.
lastFrameUrl	string	Optional end image, with a start image only.
referenceImageUrls	array	Optional, instead of a start image: up to 2 reference pictures, <Picture 1> and <Picture 2> in the prompt.
quality	hd / sd	Default hd.
duration	"5" / "10"	Length in seconds. Default "5".
aspectRatio	auto / 16:9 / 9:16 / 1:1	Default auto.
seed	integer	Optional, for a reproducible video.
videos	array	Optional batch: up to 10 objects with prompt and any of the fields above.

{
    "videos": [
        {"prompt": "<Picture 1> on a marble kitchen counter, morning light, the camera slowly orbits", "referenceImageUrls": ["https://example.com/product.jpg"]},
        {"prompt": "Steam rising from the coffee mug, slow push in", "imageUrl": "https://example.com/mug.jpg"},
        {"prompt": "Waves rolling onto a beach at dusk, aerial shot", "duration": "10", "aspectRatio": "9:16", "quality": "sd"}
    ]
}

Output

One row per video, in the order of the input:

{
    "prompt": "<Picture 1> standing on a wooden table in a cozy cafe, rain on the window behind, the camera slowly orbits the mug",
    "mode": "reference-to-video",
    "referenceImageUrls": ["https://upload.wikimedia.org/wikipedia/commons/thumb/1/1f/Coffee_Mug_%28Unsplash%29.jpg/1280px-Coffee_Mug_%28Unsplash%29.jpg"],
    "status": "success",
    "videoUrl": "https://api.apify.com/v2/key-value-stores/.../records/01-picture-1-standing-on-a-wooden-table-in-a-cozy-caf.mp4",
    "kvsKey": "01-picture-1-standing-on-a-wooden-table-in-a-cozy-caf.mp4",
    "model": "MiniMax H3",
    "quality": "sd",
    "aspectRatio": "16:9",
    "width": 896,
    "height": 512,
    "durationSeconds": 5.21,
    "fps": 24,
    "hasAudio": true,
    "seed": 11,
    "fileMB": 0.54,
    "generationSeconds": 47,
    "billedEvent": "video-sd-5s",
    "error": null
}

A video that could not be made has "status": "failed" and an error that says why. One that was left out, for example by the per-run limit, has "status": "skipped" and is not charged. A video whose generation had already started is charged even if it then fails; its error says so.

How much does it cost?

You pay per video delivered. There is no subscription and no minimum.

Video	HD	SD
5 seconds	$0.205	$0.133
10 seconds	$0.312	$0.204

• Text to video, image to video and reference to video cost the same.
• For comparison, other MiniMax video Actors on Apify charge about $0.30 per 5 seconds.
• You pay when a video starts generating. Anything stopped before that is free: an image that can't be used, a prompt refused on submit, a free-plan or per-run limit, or your budget. Once generation has started the video is charged, even if you abort the run.
• This Actor runs on paid Apify plans only. To try AI video for free, use the LTX-2.5 AI Video Generator.
• At most 10 videos per run.

Use cases

• Product videos and ads: put your product into any scene with reference images, in 9:16 for social ads.
• Social media content: realistic vertical clips for TikTok, Reels and Shorts.
• Character consistency: the same mascot or character across a series of clips.
• Animate photos: bring product shots, portraits and landscapes to life.
• Automated pipelines: write prompts with an LLM, then generate videos in bulk through the API.

Tutorial: generate videos with the API

1. One video in the Console

Write a prompt, click Start, and download the MP4 from the Output tab.

2. A product video from reference images

Add a product photo to Reference images and write a prompt such as "<Picture 1> on a marble kitchen counter, morning light, the camera slowly orbits".

3. From Python

from apify_client import ApifyClient

client = ApifyClient("<YOUR_APIFY_TOKEN>")
run = client.actor("deapi/minimax-video-generator").call(run_input={
    "prompt": "<Picture 1> on a marble kitchen counter, morning light, the camera slowly orbits",
    "referenceImageUrls": ["https://example.com/product.jpg"],
    "quality": "hd",
    "aspectRatio": "9:16",
})
for item in client.dataset(run["defaultDatasetId"]).iterate_items():
    print(item["status"], item.get("videoUrl"), item.get("error"))

4. From JavaScript or cURL

import { ApifyClient } from 'apify-client';

const client = new ApifyClient({ token: '<YOUR_APIFY_TOKEN>' });
const run = await client.actor('deapi/minimax-video-generator').call({
    prompt: 'A golden retriever running through a sunflower field, slow motion, warm light',
    duration: '10',
});
const { items } = await client.dataset(run.defaultDatasetId).listItems();
console.log(items[0].videoUrl);

curl -X POST "https://api.apify.com/v2/acts/deapi~minimax-video-generator/run-sync-get-dataset-items?token=<YOUR_APIFY_TOKEN>" \
  -H "Content-Type: application/json" \
  -d '{"prompt": "Waves rolling onto a beach at dusk, aerial shot", "quality": "sd"}'

MiniMax video API and integrations

Run it from Make, Zapier, n8n or a webhook, or chain it after another Actor: write prompts with an LLM, or animate the product images another Actor collected. The MP4 links in the dataset can be downloaded by any tool. The Apify API and the Python and JavaScript clients work with it as they do with any Actor.

FAQ

How do I make a video with MiniMax?

Write a prompt, optionally add a start image or reference images, pick HD or SD, 5 or 10 seconds and an aspect ratio, and click Start. The MP4 is in the Output tab after a minute or two.

What is reference to video?

You give the model up to 2 pictures and it keeps what they show, such as a product, a character or a place, in a new scene. Address them in the prompt as <Picture 1> and <Picture 2>, in the order you listed them.

How do I turn a picture into a video?

Paste a direct link to the picture in Start image URL and describe how it moves. The picture becomes the first frame of the video.

Does the video have sound?

Yes. MiniMax H3 generates the sound with the picture: ambience, effects and music that fit the scene.

How much does a MiniMax video cost?

$0.205 for 5 seconds and $0.312 for 10 seconds in HD; $0.133 and $0.204 in SD. Videos stopped before generation starts cost nothing.

Is there a free trial?

This Actor runs on paid Apify plans only, because every video is expensive to generate. The LTX-2.5 AI Video Generator has a small free allowance on the free plan.

Do I need an API key or a GPU?

No. You need only an Apify account on a paid plan. The Actor runs the model for you.

MiniMax H3 or LTX-2.5?

MiniMax H3 gives more realistic motion and detail, and keeps products and characters consistent from reference images. LTX-2.5 is about half the price and has a small free allowance, good for drafts, B-roll and social clips.

Can I make vertical videos for TikTok and Reels?

Yes. Set the aspect ratio to 9:16 for a 768 × 1344 HD or 512 × 896 SD video.

Can I use the videos commercially?

The videos are yours to use. Check that your prompt and images don't copy someone else's brand, character or likeness.

What content is not allowed?

Don't ask for sexual content, graphic violence or videos of real public figures. Such prompts may be refused; a prompt refused before generation starts is not charged.

How long are the videos kept?

In your run's key-value store, for your Apify plan's data retention period. Download the files you want to keep.

Can I make several videos at once?

Yes. Put up to 10 objects in Batch of videos, each with its own prompt, images and settings. A batch of 10 takes roughly 12 minutes (5 s HD) to 25 minutes (10 s HD). For more, start several runs.

Can I make a 30-second or 1-minute AI video?

Not in one clip: each video is 5 or 10 seconds. Generate several clips and join them in any video editor; reference images help keep the product or character the same across clips.

What resolution are the videos? Is it 1080p or 4K?

HD is 1344 × 768 (16:9), 768 × 1344 (9:16) or 992 × 992 (1:1); SD is 896 × 512, 512 × 896 or 640 × 640. All at 24 fps. There is no watermark.

Can I add my own voiceover or music?

No. The sound is generated with the picture; you can steer it in the prompt. Add your own audio in a video editor.

Is it an alternative to Kling, Runway, Sora or Veo?

For short clips, yes: you pay per video with no subscription, and you can call it as an API. Those tools offer longer videos and editing features this Actor does not have.

Can I use it from ChatGPT, Claude or another AI agent?

Yes. Every Apify Actor is available through the Apify MCP server, so an AI agent can generate videos with it.

Why did a video fail?

The error field says why: an image that could not be downloaded or read, a reference picture that was too long and thin (over 4:1), a prompt that was refused, or a service that was busy. A video that failed before its generation started is not charged; one that failed after is, and the error says which.

Support / feedback

Found a problem or need a feature? Open an issue on the Issues tab. We usually answer within a day.