alibaba / wan 3.0 prime

Overview

Wan 3.0 Prime is Alibaba’s accelerated Wan 3.0 video generator. It turns your written direction into a full sequence of frames, then packages them as an MP4 clip. You can anchor the motion with a starting frame, or lock both a start and end frame to guide the transition. It’s useful when you need a short, editable video draft without shooting footage.

What you can build

• Text-to-video concept clips for ads, trailers, and storyboards
• Image-to-video “photo comes alive” animations from a single hero image
• Start-to-end transitions by giving both a first frame and a last frame
• Cinematic camera moves (push-in, orbit, crane, handheld) from shot-style prompts
• Social format variations by switching aspect ratios (square, portrait, landscape)

Inputs

• A required written prompt that describes the scene, subject motion, camera motion, lighting, and style
• An optional first-frame image to start the video from a specific composition and subject look
  • Accepted formats: JPEG/JPG, PNG (no transparency), BMP, WEBP
  • Size limits: 240–8000 px on each side, max 20 MB
  • Aspect ratio limit: up to 8:1
• An optional last-frame image to steer the ending composition (same format and size limits as the first frame)
• A required target duration, selected as a fixed number of seconds
• A required output quality tier (480p, 720p, or 1080p)
• An optional aspect ratio setting
  • Adaptive follows the prompt and any provided frames
  • Fixed ratios force a specific frame shape (for example 16:9 or 9:16)
• An optional toggle to include an audio track in the output
• An optional prompt expansion toggle that rewrites and extends short prompts for more detail
• An optional numeric seed to reproduce similar variations across runs

Outputs

• One generated MP4 video file
• Video metadata alongside the file, typically including: - The final pixel dimensions (width and height)
  • Frame rate information
  • Actual generated duration
  • The seed used for the result

Limitations

• The maximum supported duration is 30 seconds.
• The maximum supported resolution tier is 1080p.
• The output frame rate is fixed by the model’s service.
• PNG inputs with transparency aren’t supported.
• Very wide or very tall source images can reduce motion quality, even when accepted.
• If you provide both first and last frames, the model may not match them perfectly.
• Low-quality inputs can cause drifting details.
  • Common risks include heavy JPEG artifacts, small faces, tiny text, and blurry edges.
• Prompt expansion can change wording and emphasis, so results may shift from your original intent.

Safety & compliance

• Use only content you have rights to upload and transform.
• Don’t generate disallowed content, including sexual content involving minors or instructions for wrongdoing.
• Avoid using real people’s likeness or voice without permission.
• Follow your local rules for labeling AI-generated media when publishing.