APIMODELSAPIMODELS
All
Models/MiniMax Hailuo H3
MiniMax

MiniMax H3 API

minimax-h3

MiniMax H3 is a general-purpose multimodal video model rather than a stack of single-purpose tools. It reads text, images, video and audio in the same pass, works out what the brief is actually asking for, and returns one coherent clip with picture, motion and sound already in agreement — no separate dubbing, lip-sync or motion-transfer step to stitch together afterwards. Practically, that means one endpoint covers the whole range: send a prompt on its own for text-to-video, or attach up to 9 reference images, 3 reference videos and 3 reference audio clips to pin down character, camera movement and voice at once. Every reference input is optional. Choose any whole-second length from 5 to 15 and an aspect ratio of adaptive, 21:9, 16:9, 4:3, 1:1, 3:4 or 9:16. Two resolution tiers, both billed per second: native 2K at $0.145/s (5s = $0.725, 15s = $2.175) and 768P at $0.097/s (5s = $0.485, 15s = $1.455) when the more affordable tier is enough. Reference inputs are priced separately: reference video is billed by its own length at $0.135/s (2K) or $0.081/s (768P) on top of the output, reference audio is free, the first 5 reference images are free and each additional image is $0.054. A single reference video may not exceed 15 seconds. Only successful generations are charged.

Native 2K or 768PMultimodal Reference5-15sUp to 9 Ref Images3 Ref Videos + 3 AudiosSynchronized Audio
2K · per second$0.145
2K · 5s$0.725
2K · 10s$1.450
2K · 15s$2.175
768P · per second$0.097
768P · 5s$0.485
768P · 10s$0.970
768P · 15s$1.455
Reference video · per second (2K)$0.135
Reference video · per second (768P)$0.081
Reference audio$0.000
Extra reference image (past 5)$0.054
API Docs

Generate Video

Leave the references below empty for pure text-to-video — every reference input is optional.

Steer character, style or composition. The first 5 are free; each image past the 5th adds $0.054.

MP4 / MOV, ≤50MB and 2-15s each. Reference video is billed by its own length at $0.135/s (2K).

MP3 / WAV, ≤15MB and 2-15s each. Used to match voice or drive lip movement. Reference audio is free.

5s15s
Native 2K · 5s$0.72

Result

Generated video will appear here

Provide URLs and click Generate

Under a Third of Seedance 2.0 1080p

At $0.145/s it costs 29.5% of our own Seedance 2.0 1080p tier ($0.492/s) — and outputs native 2K rather than 1080p. Both figures are published on this site, so you can check the comparison yourself

One Model, Four Modalities

Text, image, video and audio are understood together rather than by separate specialist tools — visuals, motion and sound arrive already in sync, in a single pass

Native 2K, or 768P to Save

Native 2K with synchronized audio and no upscaling step — or drop to 768P at $0.097/s, a third more affordable, when the clip is a draft or ships to a small player

One Multimodal Endpoint

Prompt alone behaves as text-to-video; add images, videos or audio to steer character, camera motion and voice in the same request

Any Length 5-15s

Whole-second control rather than fixed buckets, billed per second ($0.145/s at 2K, $0.097/s at 768P) so you pay for exactly the length you asked for

Deep Reference Support

Up to 9 reference images, 3 reference videos and 3 reference audio clips — reference audio is free, the first 5 images are free and extras are $0.054 each; reference video is billed by its own length ($0.135/s at 2K), capped at 15s per clip

TL;DR MiniMax Hailuo H3 is a MiniMax video generation model, callable via APIMODELS' unified API (model name `minimax-h3`). Pricing: 2K · per second: $0.145, 2K · 5s: $0.725, 2K · 10s: $1.45, 2K · 15s: $2.175, 768P · per second: $0.097, 768P · 5s: $0.485, 768P · 10s: $0.97, 768P · 15s: $1.455, Reference video · per second (2K): $0.135, Reference video · per second (768P): $0.081, Reference audio: $0, Extra reference image (past 5): $0.054. One API key for all image / video / LLM / audio models.

About MiniMax Hailuo H3

MiniMax Hailuo H3 is a Video Generation API provided by MiniMax. MiniMax H3 is a general-purpose multimodal video model rather than a stack of single-purpose tools. It reads text, images, video and audio in the same pass, works out what the brief is actually asking for, and returns one coherent clip with picture, motion and sound already in agreement — no separate dubbing, lip-sync or motion-transfer step to stitch together afterwards. Practically, that means one endpoint covers the whole range: send a prompt on its own for text-to-video, or attach up to 9 reference images, 3 reference videos and 3 reference audio clips to pin down character, camera movement and voice at once. Every reference input is optional. Choose any whole-second length from 5 to 15 and an aspect ratio of adaptive, 21:9, 16:9, 4:3, 1:1, 3:4 or 9:16. Two resolution tiers, both billed per second: native 2K at $0.145/s (5s = $0.725, 15s = $2.175) and 768P at $0.097/s (5s = $0.485, 15s = $1.455) when the more affordable tier is enough. Reference inputs are priced separately: reference video is billed by its own length at $0.135/s (2K) or $0.081/s (768P) on top of the output, reference audio is free, the first 5 reference images are free and each additional image is $0.054. A single reference video may not exceed 15 seconds. Only successful generations are charged. Through APIMODELS platform, you can access this model via a unified API with transparent pay-as-you-go pricing. Current pricing: 2K · per second: $0.145, 2K · 5s: $0.725, 2K · 10s: $1.45, 2K · 15s: $2.175, 768P · per second: $0.097, 768P · 5s: $0.485, 768P · 10s: $0.97, 768P · 15s: $1.455, Reference video · per second (2K): $0.135, Reference video · per second (768P): $0.081, Reference audio: $0, Extra reference image (past 5): $0.054.

Related model: minimax-h3-lite — the budget tier of the same model — 768P at $0.02/s (a fifth of this page) and 480P at $0.01/s; no 2K, no reference video, and the same 5-10 minute turnaround

Key Features

  • Under a Third of Seedance 2.0 1080p -- At $0.145/s it costs 29.5% of our own Seedance 2.0 1080p tier ($0.492/s) — and outputs native 2K rather than 1080p. Both figures are published on this site, so you can check the comparison yourself
  • One Model, Four Modalities -- Text, image, video and audio are understood together rather than by separate specialist tools — visuals, motion and sound arrive already in sync, in a single pass
  • Native 2K, or 768P to Save -- Native 2K with synchronized audio and no upscaling step — or drop to 768P at $0.097/s, a third more affordable, when the clip is a draft or ships to a small player
  • One Multimodal Endpoint -- Prompt alone behaves as text-to-video; add images, videos or audio to steer character, camera motion and voice in the same request
  • Any Length 5-15s -- Whole-second control rather than fixed buckets, billed per second ($0.145/s at 2K, $0.097/s at 768P) so you pay for exactly the length you asked for
  • Deep Reference Support -- Up to 9 reference images, 3 reference videos and 3 reference audio clips — reference audio is free, the first 5 images are free and extras are $0.054 each; reference video is billed by its own length ($0.135/s at 2K), capped at 15s per clip

MiniMax H3 vs MiniMax H3 Lite: price, ceiling and speed

Same H3 model on two channels. Lite is a fifth of the price at 768P but stops at 768P and takes no reference video; both tiers take minutes per clip, so choose on delivery spec, not on speed.

Dimensionminimax-h3 · this pageminimax-h3-lite
768P price$0.097/s$0.02/s (a fifth)
480P price—$0.01/s
Native 2K$0.145/sNot available — 768P is the ceiling
768P · 10-second clip$0.97$0.20
Clip length5-15s, any whole second1-15s, any whole second
Aspect ratiosadaptive, 21:9, 16:9, 4:3, 1:1, 3:4, 9:1616:9 / 9:16
Reference imagesUp to 9 — first 5 free, then $0.054 eachUp to 9, all free
Reference video (motion transfer)Up to 3, billed by their own lengthNot supported (400 at create time)
Reference audioUp to 3, freeUp to 3, free (needs ≥1 image)
Turnaround (measured)Median 6.4 min, 90% within ~10 min (14-day production data)5-10 min per clip for 10-15s; 1-3 min for ≤5s; 15s can exceed 10 min under load
Best for2K delivery, motion transfer, wide / square ratiosDrafts, social verticals, iteration, volume

→ See also minimax-h3-lite — the same model at a fifth of the price when 768P is enough.

Use Cases

Marketing Videos

Quickly generate brand promotion videos for ad campaigns and social media marketing.

Social Media Content

Create compelling short-form video content for platforms like TikTok, Instagram, and YouTube.

Product Demos

Generate product feature demonstrations and tutorials to improve user conversion.

Educational Content

Produce course explanations, knowledge explainers, and training videos at an affordable price.

Why APIMODELS

  • Unified API -- One API key to access all models, no need to register on multiple platforms
  • Transparent Pricing -- Pay-as-you-go with per-call pricing published on every model page, ideal for indie developers and startups
  • Instant Access -- Start right after signup; pay by card, PayPal, Alipay or USDT
  • Full Documentation -- Detailed API docs with code examples in cURL, Python, and Node.js

Frequently Asked Questions

What is the difference between APIMODELS and the official MiniMax API?

APIMODELS offers the same MiniMax Hailuo H3 model through our aggregation platform. We provide a unified API interface so you do not need separate accounts for each provider - one API key to access all models.

What is MiniMax Hailuo H3?

MiniMax Hailuo H3 is MiniMax's multimodal video generation model. A single request produces native 2K footage together with synchronized audio, so there is no separate dubbing step. What sets it apart from most video models is that one endpoint covers every mode: reference images, reference videos and reference audio are all optional inputs. Send a prompt on its own for text-to-video, or attach media to steer character, camera motion or voice — you never switch endpoints.

How much does the MiniMax H3 API cost?

On APIMODELS it is billed per second in two resolution tiers: native 2K at $0.145/s (5 seconds is $0.725, 10 seconds $1.45, 15 seconds $2.175) and 768P at $0.097/s (5 seconds $0.485, 10 seconds $0.97, 15 seconds $1.455). 768P is a third more affordable and is the sensible pick for drafts, social verticals and anything that plays in a small frame; use 2K for delivery or a big screen. The first 5 reference images are free and each image past that is $0.054 (9 max, so at most +$0.216); reference videos and audio cost nothing extra. There is no minimum top-up and no subscription, and only successful requests are charged — a failed generation costs you nothing.

How do I call the MiniMax H3 API?

Use the unified video endpoint POST /api/v1/video/generations with model set to minimax-h3 (the alias hailuo-h3 also works). prompt and duration (a whole number of seconds, 5-15) are required. Attach references as needed: images (up to 9), video_list (up to 3) and audio_list (up to 3), each accepting a public URL or base64. Pass ratio to pin the aspect ratio. The call returns a taskId — poll GET /api/v1/video/generations?task_id=xxx, or supply callback_url at creation and we will POST you the result when it finishes. One API key covers every model on the site, reachable from mainland China with no organization verification.

What are the parameters and limits for MiniMax H3?

Resolution is fixed at native 2K — this channel offers no lower tier. Duration is any whole second from 5 to 15 rather than fixed buckets, and the aspect ratio can be adaptive, 21:9, 16:9, 4:3, 1:1, 3:4 or 9:16. Reference limits: up to 9 images (≤30MB each), up to 3 videos (MP4/MOV, ≤50MB and 2-15s each) and up to 3 audio clips (MP3/WAV, ≤15MB and 2-15s each). Prompts can be up to 20,480 characters. We transfer the finished video to our own storage and the link stays live for 30 days, so download or re-host it before it expires.

What is MiniMax H3 best for, and how does it compare to other video models?

It is the right pick when you need native 2K, synchronized audio and multimodal references at the same time: continuous shots that must keep a character consistent, ad footage that has to reproduce a specific camera move, talking-head clips with a chosen voice or lip-sync, and complex briefs that mix image, video and audio references. On value it holds up well: $0.145/s is 29.5% of our own Seedance 2.0 1080p tier ($0.492/s), and it returns native 2K rather than 1080p — both prices are published on this site, so the comparison is yours to check. If you mainly need fast vertical clips on a tight budget, try its own 768P tier first ($0.097/s); below that, LTX-2.3 (from $0.02/s at 480p) or Seedance 2.0 Fast (from $0.071/s) still go further per dollar; if you need 4K or video editing, look at Omni Flash. On APIMODELS all of these share one API key and the same endpoint, so switching models means changing a single field.

Is this model reliable? What is the success rate?

Its own 30-day sample is small: counted the way described below, 65 of its 67 tasks succeeded and 2 failed, too few for a meaningful percentage, so we do not quote one. Median time to a finished clip is 470 seconds. The definition is stated plainly: the infrastructure success rate counts only failures that are ours to fix (timeouts, upstream congestion, no clip returned) and excludes content-moderation rejections and malformed input, because those clear once you adjust the prompt or the source material, and neither is ever billed. Worth saying out loud: almost no video-generation platform publishes a success rate at all. Their API pages offer "stable" and "highly available" — adjectives you cannot check. We publish the real 30-day production numbers together with how they are computed, and you can ask any provider the same question.

Am I charged for failed jobs?

No. Timeouts, upstream congestion and content-moderation rejections are all free — you pay only for clips actually delivered, and reserved credits are released automatically on failure. So beyond the success rate itself, your exposure is zero.

Is this suitable for a small team or business?

No subscription, no monthly fee, no minimum spend — billing is per second of output. The smallest top-up is $10 and one API key covers every model on the platform (video, image, audio and LLMs share a single USD balance). New accounts on consumer email domains get $0.10 in free credit. Paying by Stripe lets you enter a business tax ID at checkout and a formal invoice PDF is generated automatically; PayPal invoices are manual. The account is prepaid, so calls return 402 when the balance runs out rather than continuing to bill — the ceiling is hard.

How long does one clip take?

Median 470s measured in production over the last 30 days. Video generation is heavy work and the industry sits in the minutes range; submit the job, then poll with the taskId or set a callback_url — no long-lived connection required.

Can I get low prices and stability at once, and how do I pick a platform?

Our rate is $0.097/s at 768P and $0.145/s at 2K, and failures are never billed. Judge platforms on four things rather than headline price: (1) the real success rate and how it is computed — ask whether content-moderation rejections sit in the denominator, and note that most platforms publish no number at all; (2) whether failures are billed; (3) whether input material (reference images and videos) is charged — some channels bill output seconds plus input video seconds; (4) how long result files are kept. All four are answered on this page.

How is this different from going to the official channel directly?

Beyond price, the differences are mostly about access and settlement: · Onboarding — the official channel requires business verification and an account approval flow; here you sign up and go, with a $10 minimum top-up, no subscription and no minimum spend. · Region — the official channel is domestic; this one is reachable worldwide, so overseas teams do not need workarounds. · Settlement — everything is priced in USD with Stripe, PayPal and Alipay; paying by Stripe lets you enter a business tax ID at checkout and a formal invoice PDF is generated automatically. · One key for the whole platform — video, image, audio and LLMs share a single balance, so you are not opening an account and reconciling invoices with each vendor. · Failures are never billed — timeouts, upstream congestion and content-moderation rejections all cost nothing; you pay only for clips delivered. If you already hold an official account, use only this one model, and do not mind settling in CNY through a corporate account, going direct is reasonable. This solves a different problem.

What can MiniMax Hailuo H3 do?

MiniMax Hailuo H3 supports: Native 2K or 768P, Multimodal Reference, 5-15s, Up to 9 Ref Images, 3 Ref Videos + 3 Audios, Synchronized Audio. See the APIMODELS docs for full parameters and call examples.

Can I use the MiniMax Hailuo H3 API from any country, including mainland China?

Yes. APIMODELS does not restrict access by country: the site and the API answer requests from anywhere, and apimodels.app and api.apimodels.app are reachable directly from mainland China without a proxy. You call MiniMax Hailuo H3 with one APIMODELS key, so you do not need a separate account with the model's developer.

What payment methods are supported? Can I pay with PayPal or crypto?

Yes to both. You can top up with a card (Visa, Mastercard and others, through Stripe), PayPal, Alipay, or USDT on TRON (TRC20) or BNB Smart Chain (BEP20). Card, PayPal and Alipay top-ups are credited immediately; USDT is credited once the transfer confirms on-chain. Auto top-up is optional, and a $50 top-up adds a $1 bonus ($3 on a $100 top-up). The balance is in USD and billed per use, with no subscription.

Related models

Need prompts?

We curate copy-ready prompt libraries — every entry shows its full text and a sample result, ready to adapt.