MiniMax H3 Lite API

Pricing (per second)

Billed for the seconds you ask for, any whole number from 1 to 15. Reference images (up to 9) and reference audio (up to 3 clips) cost nothing extra. Charged only on success, refunded on failure.

Versus minimax-h3: same model, Lite is a fifth of the price at 768P ($0.02 vs $0.097/s — a 10-second clip is $0.20 vs $0.97) but tops out at 768P (no 1080P / 2K), takes no reference video and offers 16:9 / 9:16 only; speed is about the same on both tiers, 5-10 minutes per clip. For 2K delivery or motion transfer from a reference video, use minimax-h3.

Authentication

All requests carry the API key in the header:

Authorization: Bearer YOUR_API_KEY

Generate a video

# Text-to-video (prompt only). Add images / audio_list / first+last frame for the other modes.
curl -X POST https://api.apimodels.app/v1/video/generations \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "minimax-h3-lite",
    "prompt": "A girl with headphones nods to the beat in a bright studio, slow push-in, soft light",
    "duration": 5,
    "resolution": "768p",
    "ratio": "16:9"
  }'

# Reference images + reference audio (audio drives voice and lip movement; needs >= 1 image)
curl -X POST https://api.apimodels.app/v1/video/generations \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "minimax-h3-lite",
    "prompt": "The girl in Picture 1 sings along to Audio 1, gentle head movement, studio light",
    "images": ["https://example.com/ref1.png", "https://example.com/ref2.png"],
    "audio_list": ["https://example.com/voice.wav"],
    "duration": 10,
    "resolution": "768p"
  }'

# Poll until completed (result files are kept 7 days)
curl "https://api.apimodels.app/v1/video/generations?task_id=TASK_ID" \
  -H "Authorization: Bearer YOUR_API_KEY"

Four modes, one endpoint

There is no mode switch — the request is routed by the fields you send:

• Text-to-video — prompt only.
• Reference images — prompt + images (1-9). Locks character, product or style; refer to them as Picture 1 / Picture 2 in the prompt.
• Images + audio — prompt + images (>= 1) + audio_list (1-3 clips). Audio drives voice and lip movement; refer to it as Audio 1 in the prompt.
• First + last frame — prompt + first_frame_url + last_frame_url (both required), without images / audio_list. A first frame on its own is treated as one reference image.

Parameters

Notes

• Async, and not a fast model: plan on 5-10 minutes per clip (measured: about 6 minutes for 768P·10s, 8-11 minutes for 15s, longer when the queue is busy; 1-3 minutes for clips of 5 seconds or less). Create returns a taskId; poll GET /v1/video/generations?task_id= every 15-30 seconds until completed, or pass callback_url. Result files are deleted after 7 days — download or re-host promptly.
• No reference video: a request carrying video_list / video_urls gets a clear 400 at create time — it is never dropped silently and billed anyway. Use minimax-h3 for motion transfer.
• Parameter errors (duration outside 1-15, resolution or ratio outside the enums, more than 9 images / 3 audio clips, a last frame without a first, first/last frame mixed with references) return a synchronous 400 at create time — no task, no charge.
• Field names are identical to minimax-h3, so switching between the two means changing model only.

Try it in the Playground