MiniMax H3 API for Video Generation
Access MiniMax H3 through the independent GoEnhance Open API. Send text, frame images, or reference media to an asynchronous REST workflow; receive an img_uuid; then poll for the result or register a callback. H3 generates native stereo audio—dialogue, sound effects, and room tone—in the same pass as the picture.
Create an API Key
Choose the H3 Model That Matches Your Inputs
hailuo_h3: Text, First Frame, or First-and-Last Frame
Use hailuo_h3 for text-to-video, image-to-video with image_url, or first-and-last-frame control with optional image_end_url. The prompt is required and can contain up to 7,000 characters. See the live playground and full parameter reference.
hailuo_h3_refs: Reference-Driven Generation
Use hailuo_h3_refs when one request needs up to 9 reference images, 3 reference videos, and 3 reference audio clips. This route uses MiniMax’s official channel and offers 768P and 2K output. See the reference-to-video endpoint.
Independent Unified API Access
GoEnhance Open API is an independent third-party platform that provides unified access to video and image models. It is not MiniMax or Hailuo, and it is not affiliated with or endorsed by MiniMax. MiniMax and Hailuo are trademarks of their owner and are used only to identify the model.
Validate these fields before submission so invalid requests are rejected before tokens are deducted.
MiniMax H3 API Parameters and Limits
Shared endpoint and asynchronous response
Submit either model to <code>POST https://api.goenhance.ai/api/v1/videos/generations</code>. The response returns an <code>img_uuid</code>; generation continues asynchronously.
hailuo_h3 duration and prompt
<code>duration</code> must be an integer from 5 through 15 seconds and defaults to 5. Non-integer or out-of-range values are rejected before token deduction. <code>prompt</code> is required and accepts up to 7,000 characters.
hailuo_h3 resolution and aspect ratio
Choose 480p, 720p, or 2K. For text-to-video only, <code>ratio</code> accepts 21:9, 16:9, 4:3, 1:1, 3:4, or 9:16. When <code>image_url</code> supplies a first frame, output follows that image and the ratio parameter is not used.
hailuo_h3 frame inputs
Pass <code>image_url</code> for a first frame and optional <code>image_end_url</code> for a last frame. This model ID covers text-to-video, image-to-video, and first-and-last-frame generation.
hailuo_h3_refs reference inputs
A single request accepts up to 9 images in <code>ref_imgs</code>, 3 videos in <code>ref_videos</code>, and 3 audio clips in <code>ref_audios</code>. Duration is an integer from 5 through 15 seconds and defaults to 5; resolution is 768P or 2K.
Native stereo audio
H3 generates dialogue, sound effects, and room tone as native stereo audio in the same pass as the picture rather than dubbing audio afterward.
Cost per call is the selected per-second rate multiplied by the requested output duration. Video models bill per second of output.
MiniMax H3 Pricing Per Second
hailuo_h3 — 480p: $0.030 per second
A GoEnhance pipeline tier for lower-cost prompt iteration. It is not an officially supported MiniMax output tier.
hailuo_h3 — 720p: $0.060 per second
A GoEnhance pipeline tier for lower-cost iteration before selecting final takes. It is not an officially supported MiniMax output tier.
hailuo_h3 — 2K: $0.090 per second
Use the 2K tier for selected takes that need MiniMax’s official output resolution.
hailuo_h3_refs — 768P: $0.099 per second
Reference-driven generation runs through MiniMax’s official channel and therefore uses different tiers and pricing.
hailuo_h3_refs — 2K: $0.143 per second
The 2K official-channel tier for reference-driven requests.
Tokens, refunds, and concurrency
Usage is billed in tokens: $10 buys 500 tokens, or $0.02 per token. Points are deducted once at submission and refunded in full if generation fails. Concurrency depends on the account Build Tier and starts at 3 concurrent unfinished tasks. Review <a href="/billing">pricing tiers and credits</a>.
Integrate the MiniMax H3 Video API
1
1. Create and protect an API key
Create a key at /keys. Send it as
Authorization: Bearer <API key>; do not expose it in browser code or logs. 2
2. Select a model and validate the request
Choose
hailuo_h3 for text or frame control, or hailuo_h3_refs for reference media. Validate duration, resolution, ratio applicability, prompt length, and input counts before submission. 3
3. Submit and persist img_uuid
POST the request to
https://api.goenhance.ai/api/v1/videos/generations. Store the returned img_uuid with your own job record because the task is asynchronous. 4
4. Poll or receive callbacks
Poll
GET https://api.goenhance.ai/api/v1/jobs/detail?img_uuid=..., or register a callback URL. A callback with status=0 fires when the task starts running; a terminal callback carries the result. 5
5. Handle validation, failure, and concurrency
Do not retry invalid duration or incompatible input combinations unchanged. Failed generations receive a full refund. Queue submissions when unfinished tasks reach the concurrency allowed by your Build Tier.
Build repeatable media pipelines around asynchronous jobs, native audio, and explicit frame or reference inputs.
What Developers Ship with H3
Programmatic ad variants
Generate batches of short campaign concepts from structured prompts, then keep only approved outputs.
Product clip pipelines
Use a product image as the first frame and generate short clips whose aspect follows the supplied image.
Localized video spots
Generate picture, dialogue, sound effects, and room tone in one pass, then route each result through your localization and review workflow.
UGC production systems
Connect submissions, job tracking, callbacks, moderation, and asset storage in an automated short-form content pipeline.
Limits, model routing, billing, audio, and platform relationship.