What’s new
- Native 30-second duration: a single generation can run up to 30 seconds, which Wan says supports more complete storytelling and creative freedom, including a "smart duration" mode that lets the model judge the right length itself.
- Omni-creation with up to 20 reference assets: images, video and audio combined can total up to 20 references, and Wan can also parse complex documents and webpages as reference material.
- Pixel-level consistency: Wan says it accurately reproduces reference detail, aimed at production workflows where delivery certainty matters — not just an approximate likeness.
- Immersive audiovisual quality and precise video editing: Wan highlights across-the-board gains in realism, texture and sound design in this version, plus instruction- or reference-based editing of existing video.
What it’s good for
- Longer shorts that need to tell a complete story in one native 30-second take, without stitching together several shorter clips.
- Brand and product work where reference detail needs to come through accurately.
- Making a targeted edit to an existing clip with a text instruction or reference, instead of regenerating the whole thing.
Prompting tips
- Prompts have generous headroom, up to 20,000 characters — enough to spell out scene, characters, and camera work in detail.
- First/last-frame mode and omni-reference mode are mutually exclusive: choosing first/last-frame rules out adding reference images, video, or audio, and vice versa.
- Reference video and audio duration is counted per clip: 1–15 seconds each, and combined with the output length, the total can't exceed 30 seconds.
- Leave the aspect ratio on adaptive unless you're sure: it's inferred from your reference material by default, and forcing a ratio can clash with what you supplied.
- Decide on audio up front: it's on by default, so switch it off explicitly if you don't want generated sound.
Omni-reference, adaptive ratio, 1080P, 20 seconds.
The girl in reference image 1 keeps the same face and white dress throughout.
Reference video 1 sets the camera rhythm: a slow lateral approach, then one steady orbit.
A garden at dawn, dappled light through the leaves, she walks through the flowers and stops under a cherry tree, looking up with a smile.
Sound: birdsong, a light breeze through the leaves, distant wind chimes.
Known limits
- First/last-frame mode and omni-reference mode can't be used together.
- NSFW filtering is off by default, and the docs note filtering effectiveness isn't guaranteed, so content compliance is on you.
- Reference video and images have resolution, aspect-ratio, and file-size limits — material outside those bounds won't be accepted.
Using it in Nomi
On the generation canvas, add a video card, pick “Wan 3.0” in the model picker and choose a mode (Text to video, First frame to video, First and last frame, Multi-reference). Add your references, write the prompt and generate. In the storyboard you can pick it per shot. The first time, connect any one of the providers above under Settings → Models.
