MiniMax H3 (Hailuo 3) — 2K AI video up to 62% cheaper

Docs/Video/Tekst naar afbeelding

Minimax H3· Tekst naar afbeelding

Tekst naar afbeelding

POSThttps://api.foxwire.ai/v1/media/generations

Pricing Model

768P 2,55 ₽/s · 2K 5,1 ₽/s × 4–15s

Execution Flow

Asynchronous submission model. Submitting requests queues background workers.

#Een taak indienen

Mediageneratie is asynchroon: je krijgt direct een taak-id, zonder te wachten.

bash

curl https://api.foxwire.ai/v1/media/generations \
  -H "Authorization: Bearer $CORRY_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "model": "minimax-h3", "prompt": "a cute robot mascot, flat vector logo" }'
# → { "id": "cmus7q0x9000108l3f5g2abcd", "status": "queued" }

#Aanvraagparameters

Parameter	Type	Verplicht	Standaard	Beschrijving
`prompt`	string	Ja	—	Text prompt describing the video (max 7000 chars). Required in every mode.
`first_frame_url`	string	Nee	—	First-frame image URL (image-to-video). JPG/PNG/WEBP/HEIC ≤30MB, 256–5760px, W/H ratio 0.4–2.5.
`last_frame_url`	string	Nee	—	Last-frame image URL; must be paired with first_frame_url.
`reference_image_urls`	string[]	Nee	—	Reference images (max 9). Mutually exclusive with first/last frame.
`reference_video_urls`	string[]	Nee	—	Reference videos (max 3, each 2–15s, total ≤15s; MP4/MOV ≤50MB). Mutually exclusive with first/last frame. Per-second price doubles when reference video is used.
`reference_audio_urls`	string[]	Nee	—	Reference audio for voice/timbre (max 3, each 2–15s, total ≤15s; WAV/MP3 ≤15MB).
`resolution`	string	Nee	`768P`	Output resolution; higher costs more per second — see the pricing page. Opties: 768P, 2K
`duration`	number	Nee	`5`	Video duration in seconds, integer 4–15. Billed per second.
`ratio`	string	Nee	`adaptive`	Aspect ratio. Text-to-video requires a concrete ratio (we default to 16:9 if omitted); image-to-video always follows the input image (adaptive). Opties: adaptive, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16
`callback_url`	string	Nee	—	Optional public http(s) URL. We POST the task result here the moment generation finishes (success or failure).

#Poll for the result

Generation runs asynchronously — the submit call returns immediately with a task id. Poll this endpoint with that id every few seconds until status becomes succeeded (result ready) or a terminal error (failed / timeout).

GEThttps://api.foxwire.ai/v1/media/generations/{id}

bash

# take the id from the submit response, then poll every ~3s
curl https://api.foxwire.ai/v1/media/generations/cmus7q0x9000108l3f5g2abcd \
  -H "Authorization: Bearer $CORRY_KEY"
# repeat until "status" is "succeeded" (or "failed" / "timeout")

#Antwoordvoorbeeld

Bij indienen

json

{
  "id": "cmus7q0x9000108l3f5g2abcd",
  "status": "queued",
  "type": "video",
  "model": "minimax-h3"
}

Bij succes (na pollen)

json

{
  "id": "cmus7q0x9000108l3f5g2abcd",
  "status": "succeeded",
  "type": "video",
  "model": "minimax-h3",
  "result": { "urls": ["https://cdn.sozdai.ai/video/1782537315223-b49c0c1b771f.mp4"] },
  "cost": "1.6000",
  "currency": "USD"
}

Antwoordvelden

Veld	Type	Beschrijving
`status`	string	Taakstatus: queued / submitted / succeeded / failed / timeout.
`type`	string	Mediatype: image / video / music.
`model`	string	Het model dat je hebt aangevraagd.
`result`	object	Aanwezig bij succes. result.urls is een array van permanente bestands-URL's.
`error`	string	Foutmelding, aanwezig wanneer de taak is mislukt.
`cost`	string	Voor deze taak in rekening gebracht bedrag — een USD-string (bijv. "0.1200"). Zie het currency-veld.
`currency`	string	Currency of cost — always "USD".

What is MiniMax H3

MiniMax H3 (community names: Hailuo 3.0 / Hailuo 03) is a next-generation video model that renders 2K natively. It is fully multimodal: besides text it accepts first/last frame images and whole sets of references — images, video clips and audio, including voice timbre for talking characters.

Three modes cover the production cycle: text-to-video, image-to-video with first/last frame control, and reference-to-video for transferring style, motion and voice. Each request generates 4–15 seconds.

On Sozdai the 768P tier — still in closed beta on the official API — is open to everyone, and per-second prices are 54–62% below the official rate card.

Highlights

54–62% below official pricing

768P at $0.03/s vs $0.08 official; 2K at $0.06/s vs $0.13. Verifiable against the public rate card.

Exclusive 768P tier

Closed beta on the official API — available to everyone here, at half the 2K price.

Native 2K output

No upscaling: crisp textures and stable faces up to 15 seconds.

Multimodal references

Up to 9 images, 3 videos and 3 audio clips per request — style, motion and voice transfer.

First/last frame control

Pin both ends of the clip and let the model build the transition.

Trial credits on sign-up

Try the model for free; pay in RUB or crypto afterwards.

FAQ

How is it billed?+

Per second: 768P $0.03/s, 2K $0.06/s. A request containing reference video doubles the per-second rate; the calculator shows the exact cost before you run.

Why is it cheaper than MiniMax directly?+

We buy compute in volume across several inference providers, so we can price below the official rate card. Same model, same quality.

What are the limits?+

4–15 seconds per request; up to 9 reference images, 3 videos (≤15s total) and 3 audio clips.

What if generation fails?+

You are only charged for successful generations — failed or timed-out tasks are refunded automatically.

Is there an API?+

Yes — programmatic access over a standard HTTP API. See the API tab above for docs and a live sandbox.

Commercial use?+

Yes, generated videos can be used in personal and commercial projects.

MiniMax H3 (Hailuo 3.0)