MiniMax Hailuo H3 AI Video Generator — Text & Reference to Video 🎬

Generate a video with sound from a prompt, a photo, or a set of references — powered by MiniMax Hailuo H3, the multimodal video model that produces native audio along with the picture. No GPU, no subscription, no monthly fees: you pay per second of finished video. A pay-as-you-go alternative to Google Veo, Runway, Kling, and Sora.

Perfect for TikTok / Reels / YouTube Shorts automation, faceless video channels, product and character ads, consistent-character series, and n8n / Make / Zapier workflows and AI agents.

Three modes, no mode switch

The Actor picks the model from what you attach — there is nothing extra to configure:

What you give it	Mode	What it does
prompt only	Text to video	Generates the whole clip, with sound, from your prompt alone.
imageUrl (+ optional endImageUrl)	First / last frame	Animates your photo as the opening frame, and interpolates to the closing frame if you give one.
referenceUrls (1–5)	Reference to video	Carries the characters, products, styles, or voices from your references into a brand-new shot.

The difference between the two image modes matters: a first frame is a frame the video literally starts on, while a reference is a subject the model recreates in a new composition. They are different models, so you pick one or the other — not both in the same run.

Why this Actor?

• 🔊 Sound included — H3 generates native audio with the video: ambience, effects, and speech, from one prompt.
• 🎭 Consistent characters & products — attach references and the same face, outfit, or product shows up shot after shot.
• 🖼️ Animate your own photos — give a first frame, and optionally a last one, for controlled start-to-end motion.
• 💸 Cheap — from $0.04 per second — roughly half MiniMax's own 768p list rate and a fraction of what fal charges for H3. Pay only for successful videos; failed runs are never charged.
• 📐 Social formats built in — 9:16 (TikTok, Reels, Shorts), 16:9 (YouTube), 1:1, 21:9 cinematic, 4:3 and 3:4.
• ⏱️ Flexible length — any duration from 4 to 8 seconds, priced per second.
• 🔌 Automation-ready — call it from the Apify API, n8n, Make, Zapier, LangChain, or any AI agent via MCP.

How it compares

	This Actor (Hailuo H3)	Veo / Runway / Kling
Pricing model	Pay per second of video (from $0.04/s)	Monthly subscription / credits
Commitment	None — pay only for what you run	Recurring plan, credits expire
Failed runs	Never charged	Often consume credits
Audio	Native, generated with the video	Varies / extra step
References	Up to 5 images, video, or audio (+$0.01 each)	Usually one image
API & automation	Native Apify API, n8n, Make, Zapier, MCP	Varies / limited
Watermark	None	Often on lower tiers

How it works

1. Write a prompt describing the shot — and optionally attach a first frame image or reference URLs.
2. The Actor generates the video and stores it in the run's dataset and key-value store.
3. Download the MP4 from the output, or pipe the URL into the next step of your workflow.

Input example

Text to video:

{
    "prompt": "A lone astronaut walks across a red desert at sunset, camera slowly dollies in, wind and gravel underfoot",
    "resolution": "480p",
    "aspectRatio": "9:16",
    "duration": 6
}

First / last frame — animate a photo:

{
    "prompt": "slow push in, gentle wind, birds chirping",
    "imageUrl": "https://example.com/first.jpg",
    "endImageUrl": "https://example.com/last.jpg",
    "resolution": "480p",
    "aspectRatio": "16:9",
    "duration": 4
}

Reference to video:

{
    "prompt": "She picks up the bottle, turns to camera and smiles, soft studio light",
    "referenceUrls": [
        "https://example.com/model.jpg",
        "https://example.com/product.png"
    ],
    "resolution": "720p",
    "aspectRatio": "9:16",
    "duration": 6,
    "cfgScale": 3.5
}

Output example

{
    "videoUrl": "https://api.apify.com/v2/key-value-stores/.../records/video.mp4?signature=...",
    "prompt": "A lone astronaut walks across a red desert at sunset",
    "mode": "text-to-video",
    "imageUrl": null,
    "endImageUrl": null,
    "referenceUrls": [],
    "resolution": "480p",
    "aspectRatio": "9:16",
    "duration": 6,
    "requestId": "req-...",
    "model": "minimax/minimax-h3-t2va"
}

Pricing

You are charged per second of successful video, based on resolution — a 6-second 480p clip costs $0.24:

Resolution	Price per second	4s video	6s video	8s video
480p	$0.04	$0.16	$0.24	$0.32
720p	$0.06	$0.24	$0.36	$0.48

Choose any duration from 4 to 8 seconds. Text-to-video and first/last-frame cost exactly this; reference-to-video adds $0.01 per reference (up to 5, so at most $0.05 on top). A 6s 480p clip with 3 references is $0.27. Failed generations are never charged.

Tips for great results

• Write a shot, not a caption — subject, action, setting, then camera. Good: "a barista pulls an espresso shot, steam rising, slow push-in, warm morning light". Weak: "coffee".
• Describe the sound you want — H3 generates audio, so "rain on the window, distant thunder" or "upbeat crowd noise" is part of the prompt.
• Order your references — put the most important subject first; it anchors the shot.
• First frame vs reference — if the video must literally start on your image, use imageUrl. If you just want the same character or product in a new scene, use referenceUrls.
• Describe motion, not appearance, in first/last-frame mode — the image already defines the look, so spend the prompt on movement and camera.
• Raise guidance scale toward 5 to stick closer to your images, lower it toward 2 for freer motion.
• Set a seed to reproduce a result you liked, then vary the prompt around it.

Use it from the API

curl -X POST "https://api.apify.com/v2/acts/danitn11~minimax-h3-video-generator/runs?token=YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{ "prompt": "A neon-lit street at night, camera glides forward, rain and distant traffic", "duration": 6 }'

Works the same from the Apify n8n node, Make, Zapier, LangChain, or the Apify MCP server for AI agents.

Templates & integrations

Drop this Actor into a no-code workflow to automate video generation end to end. In each tool, search for the Apify connector, authenticate with your Apify API token, and select danitn11/minimax-h3-video-generator with the action Run Actor (or Run Actor and get dataset items to receive the video URL back).

• n8n — add the Apify node → Run Actor, map prompt (and optionally referenceUrls) from a previous step, then push the resulting videoUrl to your publishing node.
• Make.com — use the Apify › Run an Actor module, then a Get dataset items module to read back videoUrl. Chain into Buffer, Google Drive, or a social uploader.
• Zapier — trigger (new row, form submission, etc.) → Apify: Run Actor → action (post to TikTok/YouTube, save to Drive, send a Slack message).

Common pattern — automated faceless video pipeline:

New prompt (Sheet / webhook / LLM step)
  → Run this Actor (prompt + optional referenceUrls)
  → Get videoUrl
  → Post to TikTok / Reels / YouTube Shorts (or save to storage)

Use with AI agents (MCP server)

This Actor works as a tool for any MCP-compatible AI agent (Claude, Cursor, VS Code Copilot, custom agents) through Apify's hosted MCP server. Your agent can generate videos as part of a larger task — "make a product video and post it" — with billing through your Apify account.

Server URL (with this Actor preselected as a tool):

https://mcp.apify.com?tools=danitn11/minimax-h3-video-generator

Authenticate with OAuth (you'll be redirected on first connect) or an Authorization: Bearer <APIFY_TOKEN> header.

Claude Code

$claude mcp add --transport http apify "https://mcp.apify.com?tools=danitn11/minimax-h3-video-generator" -H "Authorization: Bearer <APIFY_TOKEN>"

Claude Desktop / claude.ai — add a custom connector with the server URL above (OAuth flow handles auth).

Cursor / VS Code — add to .cursor/mcp.json (or "MCP: Open User Configuration" in VS Code):

{
    "mcpServers": {
        "apify": {
            "url": "https://mcp.apify.com?tools=danitn11/minimax-h3-video-generator",
            "headers": { "Authorization": "Bearer <APIFY_TOKEN>" }
        }
    }
}

Local stdio (npx)

{
    "mcpServers": {
        "apify": {
            "command": "npx",
            "args": ["-y", "@apify/actors-mcp-server", "--tools", "danitn11/minimax-h3-video-generator"],
            "env": { "APIFY_TOKEN": "<APIFY_TOKEN>" }
        }
    }
}

Then just ask your agent: "Generate a 6-second 9:16 video of a neon street at night, rain and traffic sounds." The agent calls the Actor, waits for the result, and gets back the video URL from the output schema.

FAQ

How long does a video take to generate? Typically 2–5 minutes, depending on duration and resolution.

Does the video have sound? Yes. H3 generates native audio together with the picture — describe the sound in your prompt.

What can I use as a reference? Any publicly downloadable image, video, or audio URL — up to 5 of them.

Can I animate my own photo? Yes — set imageUrl and it becomes the first frame. Add endImageUrl and the clip transitions to that final frame.

Can I combine a first frame and references? No. They are separate models, so the run picks one mode; sending both is rejected before anything is charged.

Is there a watermark? No. The output MP4 is clean and ready to publish.

What model is used? MiniMax Hailuo H3 — minimax-h3-t2va for text-to-video, minimax-h3-fl2va for first/last frame, and minimax-h3-ref2va when references are attached.