Alibaba's Wan 3.0 turns text, frames, or references into 2-30 second clips at up to 1080p — with dialogue, sound effects, and music generated in the same pass. Try Wan 3.0 online for free!
At a glance
Use these facts to decide whether Wan 3.0 fits your next AI video test before opening the generator.
| Developer | Alibaba's Wan team developed Wan 3.0; SeedVideo AI is an independent access platform and is not affiliated with Alibaba. |
|---|---|
| Official launch | Alibaba Cloud announced Wan 3.0 on August 24, 2026, following a public beta that opened on August 6, 2026. |
| Duration | 2 to 30 seconds per generation — double the 15-second ceiling of the preceding Wan 2.7 model. |
| Input modes | Text to video, image to video with first/last frames, and reference to video with up to 10 images and 5 clips per request; the API also accepts audio files, documents (PDF, DOC, XLS, PPT, Markdown), and public webpages. |
| Output settings | SeedVideo AI exposes a 2-30 second duration slider, 480p / 720p / 1080p resolution, six aspect ratios including adaptive, and a sound on/off toggle for Wan 3.0. |
| Audio | Voices, sound effects, and background music are generated by default, with lip-synced dialogue; audio does not cost extra. |
| Best fit | Long-take narrative shorts, dialogue scenes, product ads with sound, and reference-led sequences that need more than 15 seconds. |
Official Wan sources
The overview film is Alibaba's official Wan 3.0 showcase; the four capability demos were generated with Wan 3.0 on this platform, using the exact prompts shown.
Last updated:
Quick answers
Wan 3.0 turns prompts, frames, and references into clips of up to 30 seconds with generated audio — these four answers cover what most people ask first.
01
Wan 3.0 is Alibaba's AI video generation model, announced by Alibaba Cloud on August 24, 2026 after a public beta that began August 6. It generates 2-30 second clips with native audio and takes text, frames, images, video, and audio as inputs.
02
Any length from 2 to 30 seconds — double the 15-second ceiling of Wan 2.7 and longer than most mainstream generators produce in one pass. On SeedVideo AI you set the exact duration with a slider before generating.
03
Yes, by default. Wan 3.0 produces voices, sound effects, and background music in the same generation, including lip-synced dialogue, and sound costs the same as silent output. A toggle turns it off when you want a silent clip.
04
Text prompts, a first frame (optionally with a last frame), and reference stacks of up to 10 images and 5 short clips. Alibaba's API additionally accepts audio files, documents, and webpages as source material.
Model overview
Wan 3.0 is the video generation model Alibaba Cloud announced on August 24, 2026, doubling the 15-second ceiling of Wan 2.7 to 30 seconds and generating voices, effects, and music by default.
Generated on this site
Four clips generated with Wan 3.0 through this platform — multi-shot consistency, lip-synced dialogue, physical motion, and non-photoreal style. The prompt under each clip is the exact one used.
A barista in a sunlit specialty coffee shop pours latte art into a white ceramic cup. Start on a close-up of the milk stream forming a rosetta, then cut to a medium shot as she slides the cup across the walnut counter to a customer, then a final close-up of steam curling above the finished pour. Same barista, same cafe, consistent lighting across all three shots. Warm tones, soft window light, gentle cafe ambience.
Two chess players in a dim study, face to face across an antique board. The older man in a grey cardigan moves his knight, taps the clock and says: "Your move." The young woman opposite smiles, slides her queen forward and answers: "Checkmate." Lip-synced dialogue, ticking clock, rain against the window. Moody warm lamplight, 35mm film look.
A parkour athlete sprints across a rooftop at dusk, vaults a ventilation duct, tucks into a roll and springs back to full sprint without breaking stride. Handheld tracking camera follows from the side, motion blur on the skyline, jacket and hair reacting to the wind, hard believable landings with dust kicked up. City lights flickering on below.
Hand-painted watercolor animation: a paper sailboat drifts down a rain-swollen gutter stream past towering blades of grass, a snail watches from a leaf as the boat spins through a tiny whirlpool and rights itself, soft pigment blooms bleeding at the edges of every shape, visible paper grain, gentle piano and rain patter.
When to use it
Reach for Wan 3.0 when the clip needs to run past 10-15 seconds, when sound carries the story, or when a stack of reference images and clips should steer the output.
01
Most models stop at 10-15 seconds. Wan 3.0 goes to 30 in a single generation, so a scene can build, turn, and resolve without stitching cuts together in an editor.
02
Dialogue, foley, and music come out of the same pass, lip-synced and at no extra credit cost. Write the line you want spoken into the prompt, in quotes.
03
Stack up to 10 images and 5 short clips to lock characters, products, motion, or style. In the prompt, say what each Image and Video number should control.
Three-step workflow
A stronger result starts with a clear subject, setting, motion, and soundtrack intent. Choose the model, generate a first version, then refine the prompt from what you see and hear.
Name the subject, location, camera movement, style, and mood. For dialogue, put the exact line in quotes; for sound, describe the ambience you expect.
Pick Wan 3.0 in the model menu, set duration (2-30 s), resolution, and aspect ratio, and attach frames or references when consistency matters.
Review motion, consistency, audio timing, and prompt adherence. Keep what worked, rewrite what was vague, and generate the next variation.
Next steps
Start from an image when you already have a product shot or character still, from a clip when motion should come from footage, or compare models first if you are still deciding.
01
Open the text-to-video workspace with Wan 3.0 preselected and write your first prompt.
Open text to video02
Upload a product photo or character still as the first frame, optionally pin the last frame too.
Open image to video03
Use an existing clip's motion, rhythm, or style as a reference for a new Wan 3.0 version.
Open video to video04
Compare the free trial, credit packs, and subscription options before generating at scale.
View pricing05
The other 30-second model on this site, with a larger reference stack and multi-shot editing modes.
Read the 2.5 guide06
Wan 2.5 itself is API-only and not integrated here — Wan 3.0 is, and it supersedes it on every axis.
See the Wan 2.5 pageCreative workflows
The 30-second ceiling and native audio make Wan 3.0 a fit for work that shorter, silent models cannot cover in one generation.
01
A 30-second ceiling is enough for a scene with a setup, a turn, and a payoff — generated as one continuous piece instead of stitched fragments.
02
Write lines in quotes and Wan 3.0 performs them with lip-sync, room tone, and score — usable for story beats, explainers, and character tests.
03
Generate the visual and the audio bed in one pass: product motion, ambience, and a closing beat ready for social placement.
04
Lock a character or product across shots by stacking reference images, then direct the action per shot in the prompt.
05
Describe the score and mood and the model composes to match the motion — useful for mood films, intros, and transitions.
06
Watercolor, ink, cel, and other non-photoreal looks hold their style across the full clip length, with matching sound.
Model comparison
All three run on this site. Compare them on the dimensions that decide a production: clip length, audio, reference control, and iteration cost.
| Feature | Wan 3.0 | Seedance 2.5 | Veo 3.1 |
|---|---|---|---|
| Max duration per pass | 30 seconds, any integer length from 2 s | 30 seconds via multi-shot mode | 8 seconds per generation |
| Native audio | Voices, effects, and music by default; lip-synced dialogue | Generated audio with an on/off toggle | Native audio on supported modes |
| Reference control | Up to 10 images + 5 clips per request | Up to 30 images + 10 clips per request | First/last frame and image reference |
| Resolutions here | 480p / 720p / 1080p | 480p / 720p / 1080p | 720p / 1080p |
| Credit cost at 720p | 2.4 credits per second | 7.6 credits per second | Priced per clip, see pricing page |
| Best production fit | Long takes, dialogue scenes, scored clips | Reference-heavy edits, multi-shot storyboards | High-realism hero shots and premium finals |
FAQ
Model capabilities, credit costs, input limits, and output settings all affect which option you choose.
Wan 3.0 · Up to 30 seconds, audio on
Start from a prompt, a first frame, or a reference stack, and turn it into a clip of up to 30 seconds with dialogue, effects, and music in one pass.
Start Now