Minimax H3 Reference To Video
MiniMax H3 (Official) supports reference-to-video generation: provide reference images, videos, and audio clips to guide video creation. Combines multimodal references with text prompts to produce 2K video with native stereo audio up to 15 seconds.
Key Features
- Reference-guided video generation — use images, videos, and audio as style/content references.
- Up to 9 reference images, 3 reference videos, and 3 reference audio clips per generation.
- Combine multimodal references to guide both visual style and audio characteristics.
- 2K resolution output with native stereo audio, up to 15 seconds.
- Ideal for character-consistent content, style transfer, and audio-guided generation.
Parameters
| Parameter | Required | Description |
|---|---|---|
| prompt | Yes | Text description to guide video generation alongside reference materials (max 7,000 characters). Use <Picture N>, <Video N>, <Audio N> tags to reference uploaded media. |
| reference_images | No | Array of reference image URLs to guide visual style (max 9). Formats: JPG, JPEG, PNG, WEBP, HEIC, HEIF. Max 30 MB each. |
| reference_videos | No | Array of reference video URLs for motion/style guidance (max 3). H.264/H.265 in MP4/MOV, 2-15 seconds each, total duration ≤ 15s. Max 50 MB each. |
| reference_audios | No | Array of reference audio URLs for audio guidance (max 3). WAV or MP3, 2-15 seconds each. Max 15 MB each. Requires at least one reference image or video. |
| duration | No | Video length in seconds: 4-15, integer only (default: 5). |
| aspect_ratio | No | Output ratio (default: adaptive). |
| resolution | No | Output resolution: 768P or 2K (default: 2K). |
How to Use
- Upload reference materials: images for visual style, videos for motion, audio for sound.
- Write a prompt describing how to combine the references into the output video.
- Select duration and resolution.
- Click Generate — the model blends your references with the text prompt.
Code Examples
import os
import requests
response = requests.post(
"https://aircube.ai/api/v3/minimax-h3/reference-to-video",
headers={
"Authorization": "Bearer " + os.environ["AIRCUBE_API_KEY"],
"Content-Type": "application/json",
},
json={
"prompt": "Character speaks naturally, following the reference voice and visual style",
"reference_images": [
"https://example.com/character-ref.jpg"
],
"reference_audios": [
"https://example.com/voice-ref.mp3"
],
"duration": "5s",
"resolution": "2K"
},
timeout=300,
)
data = response.json()
if data["success"]:
print("ID:", data["data"]["id"], "Status:", data["data"]["status"])
else:
print("Error:", data["error"]["message"])Pricing
| Resolution | Duration | Cost |
|---|---|---|
| 768p | 4s | $0.36 |
| 768p | 5s | $0.45 |
| 768p | 6s | $0.54 |
| 768p | 8s | $0.72 |
| 768p | 10s | $0.90 |
| 768p | 12s | $1.08 |
| 768p | 15s | $1.35 |
| 2k | 4s | $0.52 |
| 2k | 5s | $0.65 |
| 2k | 6s | $0.78 |
| 2k | 8s | $1.04 |
| 2k | 10s | $1.30 |
| 2k | 12s | $1.56 |
| 2k | 15s | $1.95 |
Billing rules
- 768p / 4s: $0.36.
- 2k / 4s: $0.52.
- Longer durations scale proportionally.
- Failed generations are not charged.
Best Use Cases
- Character-consistent video generation across multiple scenes.
- Style transfer from reference images to video content.
- Voice/audio-guided generation using reference audio clips.
- Creating variations of existing content with consistent visual identity.
Pro Tips
- Combine multiple reference types for best results — e.g., reference images for visual style + reference audio for voice.
- Reference audio requires at least one reference image or video in the same request.
- Total reference files across all types cannot exceed 12.
- Reference videos: H.264/H.265 in MP4/MOV, 2-15 seconds per clip, total duration ≤ 15s, max 50 MB each.
Notes
- Reference-to-video and image-to-video (first_frame/last_frame) modes are mutually exclusive.
- Audio references must be WAV or MP3, 2-15 seconds, max 15 MB each.