Wan 3.0 is the newest video generation model from Alibaba, and it is live in ComfyUI. Compared to its predecessor, it delivers native 30-second output, prompt input expanded to 20,000 characters, reference capacity increased to 10 images, 5 videos, and 5 audio clips, and web page input parsing. A newly added 480p tier covers lower-cost drafts, while ‘adaptive’ aspect ratio and ‘auto’ duration let the model pick the optimal aspect ratio and length on its own.
Generations run at 480p, 720p, or 1080p across 16:9, 9:16, 4:3, 3:4, and 1:1 aspect ratios. Output duration is anywhere from 2 to 30 seconds, and audio generation can be toggled off.
Model Highlights
Native 30-second duration
Wan 3.0 generates up to 30 seconds in one pass — no stitching. That’s enough room for complete storytelling: continuous camera moves, single-take shots, and pacing that builds across a full scene rather than resetting every few seconds.
Omni Creation
Generation can draw from up to 20 reference assets at once — 10 images, 5 videos, 5 audio clips, and one file or URL. References are addressed directly in the prompt with @ syntax (@Image1, @Video2, @Audio1), so you can control exactly where each one applies: a character card for identity, a video for camera movement, an audio clip for voice timbre. Document and webpage parsing means a proposal doc or a wiki page can serve as the source material for a finished clip.
Precision video editing
Wan 3.0 edits existing video by instruction. Remove or replace elements, relight a scene, change motion, transform a subject, or reshape the narrative — “have the man pick up the guitar and start playing” — while everything else in the frame stays put. It also extends existing footage, with total input plus output up to 30 seconds.
Cost
Model pricing is very competitive and scales with resolution. The 480p tier is new in this release, which allows you to iterate quickly while keeping costs down.
Getting Started
Update ComfyUI to v0.33.4, or access Comfy Cloud.
Find the Wan 3.0 nodes via the Node Library, or load a template from the Templates panel.
Add your prompt and any reference assets, connect outputs, and run.
Try it on Comfy Cloud, and let us know what you build.
As always, enjoy creating!
Assigning different roles to the character card, motion video and audio reference is clearer than treating them as one undifferentiated bundle. For a proposal-to-video example, I would keep a source checklist beside the shot timeline so that a persuasive visual does not quietly change the proposal's facts. I represent the Seedance 2.5 website (https://seedance2-5.co/); clear reference roles and a reviewed shot brief are useful in another video workflow too, while the specific input limits depend on the selected tool.