wan-3-0-video

Wan 3.0 Model Release Advances Realistic Human Rendering and Consistency Across Characters, Props, Space, and Style

Wan 3.0 has launched with a focus on realistic human rendering and reference-video consistency alongside up to 30-second multimodal video generation. The new model targets production requirements across faces, micro-expressions, body language, product structure, spatial perspective, and visual style while bringing native audio, video editing, and extension into one All-in-One model.

Model overview

Wan 3.0 is the new-generation reference-video model in the Wan video family. wan3.0-video provides standard generation, while wan3.0-video-prime significantly improves end-to-end speed with the same core capability set.

The model accepts text, images, video, audio, documents, and public web pages. It covers text-to-video, first-frame image-to-video, first-and-last-frame control, multimodal reference generation, video editing, and video extension.

Real-world fidelity becomes a primary upgrade

Wan 3.0 emphasizes more realistic and diverse human faces, reducing the tendency toward interchangeable AI-generated people. Facial features, skin detail, micro-expressions, and body language are developed together, with distinct appearance and emotional characteristics targeted even in multi-person scenes.

Reference-video generation expands consistency across four areas:

• Characters: facial traits, hairstyle, hair color, body shape, clothing, and accessories;
• Props: multi-angle appearance, hardware structure, logos, and material detail;
• Space: character blocking, camera perspective, and scene relationships;
• Style: cinematic tone, visual texture, and separation between styles in multi-look projects.

The improvements position the model for advertising, product presentation, character content, film previsualization, and continuous narrative rather than only unrelated short shots.

30 seconds, multi-shot structure, and native sound

Wan 3.0 generates 2–30 second video in a single run and includes smart duration and video extension. Output is 30fps MP4 at 480P, 720P, or 1080P with adaptive, 16:9, 4:3, 1:1, 3:4, and 9:16 aspect ratios.

Native audio covers dialogue, background music, ambient sound, and action effects. Visual motion and sound are created in the same generation process, establishing one timeline for multi-shot narrative, spoken performance, and event-driven audio.

Multimodal references and content understanding

A task can contain up to 20 multimodal reference assets, including as many as 10 images, five reference videos, and five reference-audio clips. Wan 3.0 can also read Word, Excel, PowerPoint, PDF, TXT, Markdown, Keynote, Pages, and Numbers files or use one public web page as a content source.

Brand material, product documentation, research reports, course files, or web information can become source content and be combined with visual and audio references for video generation.

Integrated editing and extension

Wan 3.0 can add, remove, replace, or modify elements in existing video, convert visual style, change lighting, and edit dialogue. Video extension continues content forward, backward, or in both directions. A post-generation change no longer necessarily requires a restart from an empty task.

Availability and limitations

Wan 3.0 Standard, Video Prime, and the very-low-cost Draft Mode are available to access and explore through wan30.io.

Audio texture and on-screen text accuracy remain areas for continued improvement. Content involving complex physical interaction, dense readable text, real people, or protected source material should retain human review and rights verification.

Model specifications

About Wan 3.0

Wan 3.0 is a new-generation All-in-One video model for complete audiovisual content, with an emphasis on longer narrative, multimodal understanding, real-world fidelity, and reference consistency.