Skip to content

Wan 3.0 Reference

Wan 3.0 is Alibaba’s flagship video generation model family. HiPix offers both standard and Prime (high-speed) variants, supporting text-to-video, image-to-video, and reference-to-video generation.

Generate a video from a text description only.

{
"model": "aliyun/wan3-0/text-to-video",
"prompt": "A waterfall cascading into a crystal-clear pool, surrounded by tropical flowers",
"seconds": 5,
"metadata": { "resolution": "720P", "ratio": "16:9" }
}

Animate a still image. Use first_frame to specify the starting frame.

{
"model": "aliyun/wan3-0/image-to-video",
"prompt": "The character turns and smiles, camera slowly zooms in",
"seconds": 5,
"metadata": {
"resolution": "720P",
"media": [{ "type": "first_frame", "url": "https://example.com/start.jpg" }]
}
}

Generate video using reference images, videos, and audio as creative guidance.

{
"model": "aliyun/wan3-0/reference-to-video-global",
"prompt": "Person in image 1 walks through the garden in image 2",
"seconds": 5,
"metadata": {
"resolution": "1080P",
"ratio": "16:9",
"media": [
{ "type": "reference_image", "url": "https://example.com/person.jpg" },
{ "type": "reference_image", "url": "https://example.com/garden.jpg" }
]
}
}
TypeCountPurpose
first_frame1Exact starting frame of the output video
reference_image10Visual references for characters, scenes, objects
reference_video5 (total ≤15s)Motion/style reference — billed by duration
reference_audio5 (total ≤15s)Audio for lip-sync or sound design
  • Must be publicly accessible via HTTPS
  • Images: JPEG/PNG/BMP/WebP, ≤20MB each
  • Videos: MP4/MOV, ≤100MB each, 1–15 seconds per clip
  • Audio: WAV/MP3, ≤15MB each, 1–15 seconds per clip
  • Range: 4–30 seconds (integer)
  • Default: 5
  • Use -1 for auto-duration (model decides)
ValueDimensions (16:9)Price
480P~854×480$0.075/s
720P~1280×720$0.15/s
1080P~1920×1080$0.30/s
ValueDescription
1:1Square
16:9Landscape
9:16Portrait (mobile)
4:3Classic landscape
3:4Classic portrait
adaptiveMatch reference media ratio
  • Boolean, default false
  • Set to true to add a visible watermark

Negative Prompt (metadata.negative_prompt)

Section titled “Negative Prompt (metadata.negative_prompt)”

Describe what you don’t want in the output:

{
"metadata": {
"negative_prompt": "blurry, low quality, text overlay"
}
}
FeatureWan 3.0Wan 3.0 Prime
SpeedStandard~2× faster
QualityHighHigh (same architecture)
Price$0.30/s (1080P)$0.42/s (1080P)
Best forQuality-focusedSpeed-focused
  1. Be specific in prompts — “A golden retriever puppy running through tall grass on a sunny afternoon” beats “a dog running”
  2. Use multiple reference images — provide character, scene, and style references separately
  3. Reference videos for motion — if you want specific movement, provide a short reference video
  4. Set the right ratio — use adaptive when you want the output to match your reference media
  5. Avoid content policy triggers — violent, sexual, or copyrighted content will be rejected