An accelerated video model from Alibaba Tongyi Wan, with multimodal references, native audiovisual generation, editing, and extension for advertising, product showcases, and short-video iteration.
Alibaba
Wan 3.0 Video Prime is the accelerated variant of Tongyi Wan’s video model. Alibaba describes its capabilities as aligned with the standard Wan 3.0 Video version, with improved end-to-end speed for workflows involving frequent creative comparisons and revisions.
It combines text-to-video, first-frame and endpoint guidance, multimodal references, editing, and extension, alongside generated dialogue, background music, and sound effects. Each task has its own asset-combination rules; input types are not universally interchangeable.
| Feature | Value |
|---|---|
| Accelerated inference | Reduces iteration waits; actual duration depends on settings and service conditions. |
| Unified task support | Organizes generation, editing, and extension around one model. |
| Combined references | Communicates subject, movement, and sound through concrete assets. |
| Generation up to 30 seconds | Provides narrative room, subject to input-mode restrictions. |
| Audiovisual creation | Incorporates dialogue and environmental sound into generation. |
| Multiple resolutions and ratios | Supports content preparation for different presentation channels. |
Prime is the accelerated variant, with officially aligned capabilities and a focus on end-to-end speed. It is not a separate higher-resolution tier; compare duration and quality using identical assets and settings.
Its accelerated positioning does not guarantee the same speed multiplier on every request. Keep duration, resolution, and assets constant, then measure completion time and success across multiple runs.
With video input, the official API limits combined input and output duration to 30 seconds. Calculate the input duration before configuring reference-video or extension tasks and check the applicable output range.
Alibaba’s official endpoint-frame mode accepts only first-frame and last-frame asset types, excluding audio references. Use the appropriate multimodal-reference mode for combined visual and audio guidance, following the platform’s rules.
Assign a clear purpose to each asset, such as subject identity or movement. Identify details that must remain intact, then check whether subjects, actions, and sounds are matched correctly in the output.
Open its LinkModel page and configure the published model identifier, authentication method, and request format. Test a short clip before enabling references, editing, or extension, and follow the platform’s task-status and result-retrieval flow.
Use fixed prompts and assets to assess subject consistency, motion continuity, audiovisual synchronization, and editing success. Also verify output duration, resolution, generation time, and failure recovery for the parameter combinations used in your workflow.
| Output specification | Our pricing | Official |
|---|---|---|
| 1080P | $0.238 / second | $0.28 / second |
| 480P | $0.058 / second | $0.068 / second |
| 720P | $0.119 / second | $0.14 / second |