MiniMax H3 API vs Kling 3.0, Seedance 2.0, and Hailuo 2.3

MiniMax H3: native 2K video on HiAPI

For native 2K scenes with generated sound, start with MiniMax H3 at $0.1189 per second. Choose Kling 3.0 for resolution and audio controls, Seedance 2.0 for reference-driven workflows, or Hailuo 2.3 for fixed-price short clips.

If you arrived here searching for "Hailuo 3 API," the current HiAPI model ID covered here is minimax-h3. Compare the input workflows, audio controls, and HiAPI prices below. The JSON snippets are starter requests; the embedded videos are existing model previews, rather than renders of those exact starter prompts.

The four contenders

Prices are current HiAPI rates from the public /api/pricing endpoint. Seedance ranges include both standard and reference-video tiers.

MiniMax H3 (minimax-h3) — native 2K scenes

MiniMax H3 supports text, frame images, and multimodal references. Reach for it when the brief needs a complete scene — a character talking, ambient sound, music cue — generated in one call instead of stitched together from a silent clip plus a separate TTS pass.

{
  "model": "minimax-h3",
  "input": {
    "prompt": "A barista in a sunlit café explains today's roast to camera, warm ambient chatter and espresso machine hiss in the background",
    "duration": 15,
    "resolution": "2K"
  }
}

A barista in a sunlit café explains today's roast to camera, warm ambient chatter and espresso machine hiss in the background

Use this starter prompt → Set the remaining input fields to the values shown above.

At $0.1189/s, a 15-second 2K clip has a budget of $1.7835. H3 also accepts up to five reference images and up to three reference videos; use image_urls and video_urls for those inputs, separately from first/last-frame control.

VIDEO

Kling 3.0 (kling-3.0/text-to-video) — tiered control over resolution and audio

Kling 3.0 is the pick when you need to dial resolution and audio independently to hit a budget, rather than accepting a model's fixed audio-always-on or audio-never behavior.

{
  "model": "kling-3.0/text-to-video",
  "input": {
    "prompt": "Drone shot pushing through a neon night market, vendors calling out, crowd noise rising",
    "duration": 10,
    "resolution": "1080p",
    "aspect_ratio": "9:16",
    "sound": true
  }
}

Drone shot pushing through a neon night market, vendors calling out, crowd noise rising

Use this starter prompt → Set the remaining input fields to the values shown above.

Sound is off by default — omit sound and you get a fully silent track, not a quiet one. The tiered pricing rewards turning it off when you don't need it: 720p silent is $0.10/s, 720p with audio jumps to $0.143/s; 1080p runs $0.129/s silent vs $0.193/s with audio; 4K tops out at $0.479/s (audio pricing not separately tiered at 4K in the current schedule).

VIDEO

Seedance 2.0 (seedance-2.0) — reference-guided generation and 4K output

Seedance 2.0 accepts reference images through reference_image_urls (up to nine images, including first/last frames) and reference videos through reference_video_urls (up to three). H3 also supports reference videos, using its own video_urls field. Choose the model and input family that match your source material.

{
  "model": "seedance-2.0",
  "input": {
    "prompt": "Extend the pocket watch transformation sequence, gears reassembling in reverse, warm brass tones",
    "aspect_ratio": "16:9",
    "duration": 5,
    "resolution": "720p",
    "generate_audio": true
  }
}

Extend the pocket watch transformation sequence, gears reassembling in reverse, warm brass tones

Use this starter prompt → Set the remaining input fields to the values shown above.

Audio defaults to on here (opposite of Kling 3.0). Pricing depends on resolution and whether reference_video_urls is supplied: 720p costs $0.293/s normally or $0.179/s with a reference video. The generate_audio switch does not select these pricing tiers. The full tiered table: 480p $0.136/s ($0.082/s with a reference video), 720p $0.293/s ($0.179/s with a reference video), 1080p $0.729/s ($0.443/s with a reference video), 4K $1.486/s ($0.915/s with a reference video) — the reference-video tier applies at each listed resolution.

VIDEO

Hailuo 2.3 (hailuo-2.3/text-to-video) — fixed price, three fields, no math required

Hailuo 2.3 is the simplest schema of the four and the only one priced per-video instead of per-second:

{
  "model": "hailuo-2.3/text-to-video",
  "input": {
    "prompt": "Time-lapse of storm clouds rolling over a coastal cliff",
    "duration": "6",
    "prompt_optimizer": true
  }
}

Time-lapse of storm clouds rolling over a coastal cliff

Use this starter prompt → Set the remaining input fields to the values shown above.

Note duration is a string — "6" or "10", not a number; sending 6 as an integer fails with a 400. There's no resolution or audio parameter to configure at all. A 6-second clip is a flat $0.40; 10 seconds is $0.80. If you're generating short clips at high volume and want predictable per-clip cost without tracking resolution tiers, this is the model that removes that variable entirely.

VIDEO

Which one to actually call

• Need generated sound and native 2K? minimax-h3, with text, frame, and multimodal-reference paths.
• Need to choose resolution and audio independently? kling-3.0/text-to-video. For frame-image input, use kling-3.0/image-to-video.
• Working from source footage? Compare H3 and seedance-2.0. Seedance has a separate reference-video price tier and output options through 4K.
• Need a fixed budget per short clip? hailuo-2.3/text-to-video, at $0.40 for 6 seconds or $0.80 for 10 seconds.

All four run through the same async task API — submit with POST /v1/tasks, poll GET /v1/tasks/{id} until data.status is success or fail. On success, select the type: "video" entry in data.output and download its url before expireAt (Unix seconds). If you've already integrated one hiapi video model, switching between these four is a matter of swapping the model ID and adjusting the handful of fields above — use the Seedance 2.0 parameters or the Hailuo 2.3 parameters for the complete input contract. The H3 and H3 Max pricing guide explains per-second batch budgets.

Grab an API key from the dashboard and start with whichever model matches the job above — the schema differences are small enough that testing two of them side by side on the same prompt takes only a few minutes.