Create Half-Minute Scenes with Sound in Wan 3.0
Last verified: November 3, 2026
Wan 3.0 is the newest generation of Alibaba's Wan video model family . It turns a text prompt, a pair of key frames, or a mix of image, video and audio references into a clip up to 30 seconds long with its own soundtrack. Alibaba opened it in public beta on Alibaba Cloud Model Studio in early August 2026 . There, developers call the standard version by its model name, wan3.0-video . Alibaba calls the series an All-in-One video generation model: text-to-video, image-to-video from a first frame or a first-and-last-frame pair, and reference-based generation all run through the same model .
Length is the headline capability. The standard version renders up to 30 seconds at 30fps in one generation . That's enough time for an action to build, peak and resolve inside one clip instead of across stitched fragments. Alibaba lists major upgrades for the wan3.0 series in audio-video generation and multi-modal reference. Its sibling, Wan 3.0 Prime, is the accelerated version. It generates much faster while keeping output quality high, with the same 480P, 720P and 1080P tiers and the same 30-second ceiling . Both versions run here in four modes (Text to video, Image to video, First to last frame and Reference to video), with sound generated alongside the picture.
Explore Wan AI's Models
Clip Limits and Inputs at a Glance
The limits that shape each clip, from length and frame rate to what you can upload.
Clip length per run
3–30 seconds, in one-second steps
Frame rate, standard Wan 3.0
30fps
Output resolution
480P, 720P or 1080P
Framing options
Auto, 1:1, 3:4, 4:3, 9:16, 16:9
Reference inputs
1–10 images, plus up to 5 videos and up to 5 audio files
Audio track
Generated with the picture via Generate Audio, on by default in every mode
Before You Generate: Six Checks for Long Takes
Most weak results come from untouched defaults or uploads that work against the prompt. Catch them here.
Set Video Duration before you write the action
The duration selector starts at 3 seconds, far short of the 30 both versions can reach. Pick the length first. Then write a sequence of actions that fills it, such as an entrance, a turn and a reveal, instead of one static moment stretched across the whole clip.
Test at 480P, finish at 1080P
Resolution starts at 480P. Use it to check whether composition, pacing and sound work, then switch to 1080P for the version you'll publish. The 1080P run is a new generation, not an upscale of the draft, so watch it in full before you use it.
Pick the frame shape for where the clip will play
Aspect Ratio starts at 1:1. Before you generate, switch to 9:16 for vertical feeds, 16:9 for widescreen, or 3:4 and 4:3 for in-between layouts. Reframing a finished clip later crops away the composition the model built.
Make your first and last frames compatible
First to last frame needs both images: jpg, jpeg, png or webp, up to 10 MB each and at least 300×300 pixels. Keep the subject, lighting and lens feel similar between them, and say in the prompt how one turns into the other. Two unrelated pictures leave the model an awkward path to bridge.
Keep reference footage plus output within 30 seconds
Alibaba's API reference says that when video is part of the input, the input video's length plus the output length must not exceed 30 seconds . When you attach clips under Add Video, lower Video Duration so the two add up to 30 seconds or less. For example, pair 10 seconds of reference footage with a 20-second output.
Write the soundtrack into the prompt
Generate Audio is on by default, so every run comes with sound unless you switch it off. Name who is speaking and put their lines in quotes, then add ambience and music cues so the track follows the action. Turn the switch off if you plan to score the footage yourself.
Wan 3.0 or Kling 3.0: Choose the Generator for Your Next Clip
Both families run on this site, so the choice comes down to the controls each generator gives you. Each row shows what that generator lets you pick or upload here, version by version: Wan 3.0 and Wan 3.0 Prime on one side, Kling 3.0 and Kling 3.0 Turbo on the other.
| Feature/Spec | Wan 3.0 | Kling 3.0 |
|---|---|---|
| Modes you can run | Wan 3.0 and Wan 3.0 Prime: Text to video, Image to video, First to last frame, Reference to video | Kling 3.0: Text to video, Image to video, First to last frame, Motion control. Kling 3.0 Turbo: Text to video, Image to video |
| Video Duration range | 3–30 seconds in every mode, on Wan 3.0 and Wan 3.0 Prime | 3–15 seconds on Kling 3.0 (Text to video, Image to video, First to last frame) and Kling 3.0 Turbo (Text to video, Image to video) |
| Resolution choices | 480P, 720P or 1080P in every mode, on Wan 3.0 and Wan 3.0 Prime | Kling 3.0: 720P, 1080P or 4K (Text to video, Image to video, First to last frame); 720p or 1080p in Motion control. Kling 3.0 Turbo: 720P or 1080P |
| Aspect Ratio in Text to video | Auto, 1:1, 3:4, 4:3, 9:16 or 16:9, on Wan 3.0 and Wan 3.0 Prime | Kling 3.0: 1:1, 9:16 or 16:9 |
| Prompt length limit | 5,000 characters in every mode, on Wan 3.0 and Wan 3.0 Prime | Kling 3.0: 2,048 characters (Text to video, First to last frame) or 2,500 (Image to video, Motion control). Kling 3.0 Turbo: 2,048 characters |
| Generate Audio switch | In every mode on Wan 3.0 and Wan 3.0 Prime, on by default | Kling 3.0: in Text to video, Image to video and First to last frame, on by default |
| First to last frame uploads | Wan 3.0 and Wan 3.0 Prime: First Frame and Last Frame, both required; jpg, jpeg, png or webp, up to 10 MB, at least 300×300 pixels | Kling 3.0: First Frame and Last Frame, both required; jpg, jpeg, png or webp, up to 10 MB, at least 300×300 pixels |
Where Each Generator Fits Your Workflow
Long single takes or 4K deliverables
The biggest difference is clip length versus pixel count. Either Wan version lets one generation run twice as long as the longest Kling 3.0 clip here. That's what a slow walk-and-talk, a chase across a set, or a time-lapse from dawn to dusk needs to play out without a cut. Kling 3.0 is the route when the delivery spec demands a 4K master and the shot fits inside its shorter window.
If a project needs both, plan around the constraint that can't bend. Either cut a long scene into shorter Kling 3.0 segments for a 4K edit, or keep the scene whole as one Wan generation and deliver at its top tier.
Room to direct a text-only shot
A half-minute scene needs more instructions than a short one: beats, camera moves, spoken lines and sound cues. The Wan prompt box holds at least twice as much text as any Kling 3.0 box, so you can write all of it out. With Kling 3.0, cut the direction down to essentials, which suits its shorter clips anyway.
For Text to video framing, the Wan generator adds Auto plus the 3:4 and 4:3 shapes to the square, vertical and widescreen options both sides share. That's useful for portrait feeds and classic TV proportions.
Reference-driven scenes or motion transfer
Your starting material decides this one. If you have a set of character, product or location images, maybe with a short clip or audio sample, Reference to video on Wan 3.0 and Wan 3.0 Prime builds the new scene your prompt describes from them. If you have a performance clip whose movement a character should follow, Motion control on Kling 3.0 takes that clip plus a character image. If you already know the opening and closing pictures, First to last frame takes the same pair of uploads on both sides, so choose by length and resolution instead.
Match the Generator to Your Clip Length, Resolution and Inputs
Use this quick guidance to pick the best option for your workflow.
How to choose: Pick Wan 3.0 or Wan 3.0 Prime when a clip has to run past 15 seconds (up to 30), when your prompt needs more than 2,500 characters (up to 5,000), when a Text to video shot needs Auto, 3:4 or 4:3 framing, or when you want to build from image, video and audio references in Reference to video. Pick Kling 3.0 when the deliverable must be 4K or when you want Motion control. First to last frame takes the same two uploads on both, and both turn Generate Audio on by default in the modes that show it.
Generate Your First Long-Take Clip
Six steps, from choosing a version to a finished clip.
Pick a version and mode in the model list
Choose Wan 3.0 or Wan 3.0 Prime, then Text to video, Image to video, First to last frame or Reference to video. Both versions show the same uploads and settings in each mode, so you can switch between them without relearning anything.
Add files with Image, Upload first frame or Add Image
Skip this step in Text to video. Image to video takes one picture under Image. First to last frame needs Upload first frame and Upload last frame. Reference to video needs at least one file in Add Image, and Add Video and Add Audio are optional.
Write the Prompt and try Enhance Prompt
Describe the subject, action, camera and sound in the Prompt box, which holds up to 5,000 characters. Click Enhance Prompt under the box to see fuller versions of a rough idea. Your text only changes if you pick one.
Choose Video Duration, Resolution and Aspect Ratio
Set how long the clip runs, how sharp it is and what shape it takes. Adjust all three before you generate: the starting values suit a quick square test, not a finished piece.
Set Generate Audio and Seed
Leave Generate Audio on to get a soundtrack with the picture, or switch it off for silent footage. Seed is optional; type a number into it if you want to.
Click Generate
Start the run. You get back one video made with the settings you chose.