Wan 3.0 AI Video Generator
Type a line or drop in a photo. Get back up to 30 seconds with the sound already in it. First clip free, no card.
SIGN UP FREE · 1 CLIP · 480PSIGN UP FREE · 1 CLIP ON WAN 3.0 · 480P · 3s · WITH SOUND
- First clip free · no card
- 30 seconds in one pass
- Sound in the same file
- 480P / 720P / 1080P
- Failed runs cost nothing
ReceiptYour free clip is Wan 3.0, not 2.7model ID and task ID printed under every result
Every clip on this page shows its receipt
- model
- wan3.0-video
- resolution
- 848×480
- length
- 5.0s
- frame rate
- 30 fps
- audio
- on
- task
- 0385dc79…
See the raw response
This is the JSON Alibaba's API returned for this clip, untouched. Every field on the strip above is read out of it. If the two ever disagree, the strip is the bug.
Served through Alibaba Cloud Model Studio. TRY WAN3 is an independent interface — we are not Alibaba, and we cannot grant rights the model provider does not grant.
Seven Wan 3.0 prompts — each one beside the clip it describes
Every card shows a clip and the exact prompt that produced it — not a summary. One press loads it into the generator at the badge's ratio, length and resolution.
- 6 of 7Alibaba's own, with their prompt
- 1 of 7read back from the footage — labelled
- 0rewritten or shortened
- Our own runs replace these the day the key is live
16:9 · 720P · 15SThe full prompt, not a summary
A teenage girl in an oversized grey sweater and a pleated skirt walks across a campus forecourt in the middle of the day, other students in uniform passing without looking at her; her right arm is a glossy pink prosthetic. Then she stops and armour plates unfold from that arm and travel across her body until she is inside a full pink mech suit, the visor closing last. After that the camera…
Street noise and footsteps first, then servo whine and plate locks, then a low rumble under the brute, no music.
One prompt, one pass, fifteen seconds — three shot sizes and no cut
TextA car becomes an animal in four beats, and the fourth one is the ending
TextFifteen seconds of one continuous move, across the glacier
ReferenceTwo reference subjects held across an orbital handoff
TextAstronaut in the ruins
TextStreet chess turnaround
ImageThe dragon opens its eyes
ReferenceDesert train heist
- Camera
Name one move. Not "cinematic".
- Sound
Say what it sounds like — silences included.
- Change
Describe how the frame changes across its length, not one frozen moment.
- Dialogue
Its own sentence, or it gets narrated instead of spoken.
Four ways into the Wan 3.0 AI video generator
One model covers all four. Wan 3.0 folded text, image, reference and editing into a single endpoint, so you are not switching products between jobs — and a reference set up for one shot still applies to the next.
Wan 3.0 text to video: what one prompt has to carry
A prompt here does two jobs at once: the picture and the soundtrack come out of the same pass. Beside this is one prompt revealed a layer at a time — subject, camera, light, sound — with the clip changing as each lands.
Five dancers in black chiffon stand in a bank of ground fog,
lit from directly behind by one hard white source, no fill.
They gather, they open outward, they draw back in and hold.
Breath, bare feet on a wooden floor, the rustle of fabric. No music.
- Subject · camera · light
- Sound — same prompt
16:9 · 720P · 12Swan3.0-video · 16:9 · 720P · 12s · sound in the same pass
Those four lines are the shape of the prompt behind this clip; the whole thing runs to 876 characters and is printed in full on [/prompt-ideas](/prompt-ideas). Wan 3.0 takes 20,000 — about 3,000 words, or a twelve-page script, for a clip of at most thirty seconds. The room is for being specific, not for writing more.
Wan 3.0 image to video: first frame, last frame, or both
Give Wan 3.0 one still and it starts there. Give it two and it has to arrive — a last frame is a destination, not a suggestion, which is what makes two-ended runs the predictable way to work.

WAN 3.0 · 16:9 · 720P · 5Swan3.0-video-prime · 16:9 · 720P · 5s
Wan 3.0 generates everything between
- 0 images = text to video
- 1 image = one anchor
- 2 images = both ends pinned
- JPEG · PNG · WebP · BMP · ≤20MB
- No alpha channel — flatten transparent PNGs
JPEG, PNG, WebP or BMP, up to 20MB. The input that fails without an obvious reason is a PNG carrying transparency: Wan 3.0 takes no alpha channel, so flatten it before you upload.
Wan 3.0 reference to video: one face across the whole take
Load up to 10 images, 5 video clips and 5 audio tracks into one generation and point at them by number — Image 1, Video 2. Beside this is one reference carried across a whole take: same subject, first second to last.

Image 1 · Character
16:9 · 5SSame performer, same outfit · 16:9 · 5s
- Images ≤ 10 — free
- Clips ≤ 5 · 15s total — billed
- Audio ≤ 5 — free
- Reference seconds count against the 30s ceiling
Images and audio attach for free. Reference video does not: its seconds bill at your output rate and count against the same 30-second ceiling, so a 15-second reference plus a 15-second result is the longest run you can ask for.
Turn a PPT, PDF or web page into video
The part other video models do not do. Give Wan 3.0 a deck, a report or a public URL and it reads what is inside and generates footage from the content — new video, not your slides with an avatar reading the bullets.
- A long-haul chassis that reconfigures in the field
- Eight minutes from road profile to working profile
- First customer trials on the interstate corridor in Q4
SAMPLE · 16:9 · 15SA sample render standing in for the output, and the same clip the deck-to-teaser card below uses. Our own document runs land here once the upstream key is live.
One file or one link per run · up to 100MB · up to 50 pages · doc, xls, ppt, pdf, txt, key, pages, numbers, md
It works best when the document carries a story the video can retell — a launch deck, a feature spec, a set of results worth dramatising. It works badly when the deck itself is the deliverable, or when the numbers on the slide have to appear on screen exactly as typed: Wan 3.0 renders text well for a video model, but it is generating that text, not copying it. Check anything legal or financial frame by frame before you publish.
Six jobs people are already using Wan 3.0 for
Six jobs, each with a runnable prompt and the clip that prompt describes. The clips are sample renders rather than our own output until the upstream key is live, and where a job is a bad fit for Wan 3.0 we say so rather than leave it out.
16:9 · 30s · 960 creditsProduct cut for a launch post
Thirty seconds of Wan 3.0 is long enough to place the product, live with it for a beat and land on the hero shot — one take, with the sound already under it.
16:9 · 20s · 640 creditsTurning a deck into a teaser
Drop the launch PDF in and Wan 3.0 builds the visuals out of what the slides claim — footage that dramatises the story, not your slides with an avatar reading them.
16:9 · 10s · 320 creditsA talking scene that lip-syncs
Put the line in braces and Wan 3.0 speaks it rather than narrating it. Everything outside the braces — who says it, how, and what happens next — stays description.
16:9 · 15s · 480 creditsKeeping one character across shots
Load the character reference once and cite it by number. The same face, hair and outfit survive a change of shot size and a change of light inside one take.
16:9 · 8s in, 8s out · 512 creditsFixing a shot you already have
Upload the footage and describe the change. Wan 3.0 rewrites it rather than regenerating, so the blocking and the timing you liked stay where they were.
16:9 · 12s · 96 credits at 480PB-roll from a single photo
One still becomes twelve seconds of movement. Cheapest thing Wan 3.0 does, and the fastest to iterate on.
Where Wan 3.0 is the wrong tool: hard-locked on-screen text that has to match a legal document word for word, anything longer than thirty seconds in a single pass, and 4K delivery, which it does not support. For those, generate here and finish in an editor.
Wan 3.0 specs, checked against Alibaba's own docs
Every row links to the page it came from and carries the date we last checked it. When Alibaba changes something, this table changes — and if you find a row that is out of date, tell us and we will fix it the same day.
Source= links to the primary page
WAN 3.0 · 16:9 · 720P · 15SNo, Wan 3.0 does not output 4K
Some sites advertise a 4K version. Alibaba's API reference lists three resolution tiers and 1080P is the top one. If a service hands you 4K here, that is an upscaler running after the model, not Wan 3.0 producing those pixels. We would rather lose the click than repeat a number we cannot source.
Checked by Hao Lin against the pages linked above on 25 August 2026. If a row has gone stale, tell us — this date moves when the table does.
Wan 3.0 against the models people cross-shop it with
Published numbers only. Where a spec is not on a first-party page we leave the row out rather than fill it with a hedge — a comparison table is the easiest place on the internet to get caught guessing.
Max single-pass length
- Wan 3.0
- 2–30s
- MiniMax H3
- 4–15s
Frame rate
- Wan 3.0
- 30 fps
- MiniMax H3
- 24 fps
Resolution tiers
- Wan 3.0
- 480P · 720P · 1080P
- MiniMax H3
- 768P · 2K
Prompt limit
- Wan 3.0
- 20,000 characters
- MiniMax H3
- 7,000 characters
Reference inputs
- Wan 3.0
- 10 images · 5 clips · 5 audio
- MiniMax H3
- 9 images · 3 clips · 3 audio
Document or web page input
- Wan 3.0
- Yes
- MiniMax H3
- No
Editing a clip without regenerating it
- Wan 3.0
- Yes
- MiniMax H3
- No
Open weights
- Wan 3.0
- No
- MiniMax H3
- YesTheirs
Blind-test rank, with audio
- Wan 3.0
- #1, stated range 1–2
- MiniMax H3
- #3
And against Wan 2.7, which is what most sites are still running
Alibaba's own model table puts Wan 2.7 at 2–15 seconds, 720P and 1080P, 30 fps. Wan 3.0 doubles the ceiling to thirty seconds — but the change you will feel every day is the tier it added underneath: Wan 2.7 has nothing below 720P, and Wan 3.0 has 480P. That is where you iterate before you spend, and it is why a draft here costs a quarter of a final. The full version-to-version comparison goes through what else moved.
The honest read: pick MiniMax H3 when you need to run the weights on your own hardware, and pick Wan 3.0 when the clip has to run long, carry its own sound, or start from a document you already wrote. The three we did not table — Veo 3.1, Seedance 2.5 and Kling — each have a full comparison of their own:
Where Wan 3.0 ranks against other video models
Artificial Analysis runs blind head-to-head preference tests and publishes the Elo, the confidence interval and the number of comparisons behind it. Read all three before you quote the rank.
- Range 1–2
#1
9001500- Elo
- 1,240
- 95% CI
- ±9
- Comparisons
- 5,728
- Second place is 3 points behind — inside the interval
#2
9001500- Elo
- 1,237
- Gap
- −3
#3
9001500- Elo
- 1,227
- Gap
- −13
On the Artificial Analysis text-to-video arena, in the with-audio table, Wan 3.0 sits first with an Elo of 1,240 from 5,728 blind comparisons — snapshot 25 August 2026. Read that carefully before you quote it: second place is three points behind and the interval is ±9, so the leaderboard itself gives Wan 3.0 a range of 1–2 rather than a clean win.
What the table shows more clearly is price. At 720P, Wan 3.0 works out to roughly $6.00 per minute of finished video — level with the model ranked second and below the one ranked third.
Source: Artificial Analysis Video Arena, With Audio board ↗ · snapshot 2026-08-25
One model at every tier — pick the size of your month
Every plan runs the same wan3.0-video at all three resolutions, the full thirty seconds, with sound and commercial use. The tiers differ in how many seconds you get and how many takes come back at once — never in what the model is allowed to do.
Same monthly credits either way
- Free first clipNo card
One generation, once, on the real `wan3.0-video`. A sample rather than an allowance — every free second is cash spent before any arrives.
$0onceNo card at any point
Generate free clipAn email, no card
1 clip, once
480P · up to 3s · with sound
An email, no cardWan 3.0 has no free tier of its own — Alibaba meters it from the first second
- The real wan3.0-video, not 2.7
- Sound generated with the picture
- Model ID and task ID on the receipt
- A failed run never costs you
- 720P and 1080PPaid
- Clips longer than 3 secondsPaid
- Reference, document and editing modesPaid
- More than one clip, everPaid
Anyone advertising unmetered free access is either serving a different model or paying a bill that will not last.
At a glance
- Clips a month
- 1, once
- Takes per run
- 1
- Longest clip
- 3s
- Top resolution
- 480P
- StarterSave 50%
A launch post a week. Enough to find out whether Wan 3.0 suits the way you work, without a decision about volume.
$9.90/mo$19.80$118.80 billed yearly
Cancel anytime
480 credits / month · ≈ 12 clips
at 5s 480P · 6 at 720P · 3 at 1080P
Plan credits expire at the end of the month
- **480 credits a month** — about 12 finished clips
- 2 takes of one prompt per run
- Every resolution — 480P, 720P and 1080P
- Every length — 2 to 30 seconds in one pass
- Every input — text, image, reference, document, web page
- Sound written in the same pass, no watermark
- Commercial use included — publish it, sell it, bill for it
- Wan 3.0 Prime, the fast tier, whenever you want it
- A failed run never costs credits
- 7-day refund while the credits are untouched
- Cancel in two clicks — the price you joined at is locked
Everything under the first two lines is on this plan at $9.90. The bigger plans buy seconds and takes — never a better model.
At a glance
- Clips a month
- ≈ 12 at 5s 480P
- Takes per run
- 2
- Credits per dollar
- Baseline
- Refund window
- 7 days
- Most people land herePro
One finished clip every working day, with three alternates each. The draft-at-480P, finish-at-1080P loop, sized for a working week.
$29.90/mo$59.80$358.80 billed yearly
Cancel anytime
1,760 credits / month · ≈ 44 clips
at 5s 480P · 22 at 720P · 11 at 1080P
Plan credits expire at the end of the month
- **1,760 credits a month** — 3.7× Starter4×
- ≈ 44 finished clips a month, or 11 at full 1080P
- **4 takes of one prompt per run** — pick one instead of re-rolling2× takes
- **+32% credits per dollar** than StarterBetter rate
- Room to draft at 480P and finish at 1080P in the same week
- Every resolution — 480P, 720P and 1080P
- Every length — 2 to 30 seconds in one pass
- Every input — text, image, reference, document, web page
- Sound written in the same pass, no watermark
- Commercial use included — publish it, sell it, bill for it
- Wan 3.0 Prime, the fast tier, whenever you want it
- A failed run never costs credits
- 7-day refund while the credits are untouched
- Cancel in two clicks — the price you joined at is locked
The jump from Starter is quantity, takes and rate. The model, the resolutions and the thirty seconds were already yours.
At a glance
- Clips a month
- ≈ 44 at 5s 480P
- Takes per run
- 4
- Credits per dollar
- +32% vs Starter
- Refund window
- 7 days
- StudioBest rate
Client volume, and the plan where you stop rationing 1080P. Forty full-resolution finished clips a month, at the lowest credit price on this page.
$79.90/mo$159.80$958.80 billed yearly
Cancel anytime
4,960 credits / month · ≈ 124 clips
at 5s 480P · 62 at 720P · 31 at 1080P
Plan credits expire at the end of the month
- **4,960 credits a month** — 10× Starter, 2.8× Pro13×
- ≈ 124 finished clips a month, or **31 at full 1080P**
- **+65% credits per dollar** — the best rate on this pageBest rate
- Enough 1080P that you stop drafting at 480P first
- 4 takes of one prompt per run
- Sized for a client roster rather than one channel
- Top up any month without changing plan
- Every resolution — 480P, 720P and 1080P
- Every length — 2 to 30 seconds in one pass
- Every input — text, image, reference, document, web page
- Sound written in the same pass, no watermark
- Commercial use included — publish it, sell it, bill for it
- Wan 3.0 Prime, the fast tier, whenever you want it
- A failed run never costs credits
- 7-day refund while the credits are untouched
- Cancel in two clicks — the price you joined at is locked
Studio is volume and rate, nothing else. If you are not spending a month’s credits, Pro is the honest answer — we would rather write that here than take the difference.
At a glance
- Clips a month
- ≈ 124 at 5s 480P
- At full 1080P
- ≈ 40 a month
- Credits per dollar
- +65% vs Starter
- Refund window
- 7 days
What one Wan 3.0 video
actually costs
Move the slider and the number updates before you commit. Credits come off when the clip comes back, not when you press the button — a failed run costs nothing. Every plan runs the same model at all three resolutions.
5s · 720P
This clip: 80 credits
≈ $2.72 per finished 5-second 1080P clip on Pro · ≈ $0.68 at 480P
Failed runs cost 0. Drafting at 480P costs a quarter of 1080P.
A worked example, because per-second numbers are hard to feel: a 12-second clip at 720P is the length most people land on for a product cut. A 5-second 480P draft of the same shot costs a fraction of that, which is why the sensible loop is to iterate at 480P until the framing is right and spend the 720P run once. Batch mode charges per clip, so four variants cost four times one — there is no bulk discount and we are not going to pretend there is.
Wan 3.0 has no free tier of its own — Alibaba meters it from the first second, and anyone offering unmetered access to Wan 3.0 is either running a different model or absorbing a bill they cannot absorb for long. Your first clip here is free because we pay for it: one generation, once, 480P, up to three seconds, on the real `wan3.0-video`. An email, no card. It is a sample rather than an allowance. After that, plans start at $9.90 a month.
- Five seconds each; longer clips cost proportionally more
- Yearly pricing shown — monthly is double for the same credits
- Failed runs cost 0
Four published Wan 3.0 reviews, quoted — including the unflattering ones
One benchmark and three reviews, none of them ours and none of them Alibaba’s. Every quotation below is a link to the page it was published on, with the date we read it. Two of the four are corrections rather than praise, and both stay up.
Wan 3.0 tops the with-audio text-to-video board at Elo 1,240 — but on 5,728 comparisons, the fewest in the top five, and the board itself states its range as 1–2. Second place is three points behind, which is inside the error bar.
- News analysis
The "Wan 3.0 does 4K" claims circulating online are misinformation: Alibaba has no 4K tier and no mention of 4K anywhere in the announcement.
Read the review
Ilia TrikozCo-founder, GPTunneL · GPTunneL8 Aug 2026 - Hands-on review
Not 30 seconds stitched from six 5-second clips, but one continuous take with a beginning, middle and end. Audio quality is good, not studio-grade.
Read the review
Our Code WorldA week of hands-on testing · ourcodeworld.com19 Aug 2026 - Model review
It should not be treated like other downloadable open-weight models. The practical way to use Wan 3.0 is through online products or APIs, not by downloading weights and running it locally.
Read the review
AirMoreModel review · airmore.ai18 Aug 2026
Every line above is a quotation with a link, not a summary. Alibaba's own announcement is a source too, but it is not independent, so it sits in the specs table further up instead of on this wall.
How to use the Wan 3.0 AI video generator in three steps

A neon high street at night seen from behind one walking figure, footsteps, rain on awnings and distant traffic
PictureSoundDescribe or upload
Type the shot, or drop in an image, a reference clip or a document. Naming the camera move and the sound gets you closer than stacking adjectives.
Same prompt · same pass
16:9
4:3
1:1
3:4
9:16
234567891011121314151617181920212223242526272829302–30s · whole seconds · 30 fps
Pick length and resolution
Anything from 2 to 30 seconds. Draft at 480P and finish at 1080P — same model, a quarter of the cost per second.
30 fps · 480P / 720P / 1080P
Speech · effects · musicclip.mp4 · 1080PDownload the MP4
Audio is already mixed in. There is no render queue to join afterwards and no editor to open.
Sound already in the file
About 90 seconds for a short clip, longer for 30 seconds at 1080P. You can close the tab; the result waits for you.
Read before you spend: four Wan 3.0 explainers
Four questions come up before anyone pays for anything Wan 3.0 related — where to go, what it costs, what happens when it fails, and whether it is even the right model. We wrote the answers out properly rather than burying them in a FAQ line.
GuidesHow to use Wan 3.0: four routes in, and what each one asks of you
There is no single Wan 3.0 app. Four ways in, differing by what you need before you start: an Alibaba account, a phone number, a card, or nothing at all.
Editorial desk
Cost & pricingWhat a Wan 3.0 clip actually costs, once you count the attempts
The published rate is $0.05 to $0.20 a second. Four multipliers sit between that and your bill, and one charges you for footage you uploaded, not generated.
Editorial desk
GuidesWhy a Wan 3.0 generation fails, and which failures cost you money
Six ways a job dies. Two are your fault, three are not, and one quietly bills you for a full thirty-second render. The error strings, and what to check first.
Editorial desk
AlternativesWan 3.0 vs MiniMax H3: one question settles it before the spec table
Both make video with sound; only one you can download and run. If that matters the comparison ends on line one. If not, the numbers are 30 seconds and 30 fps.
Editorial desk
Wan 3.0 questions people actually ask
Thirteen questions, answered on this page.
Is this the real Wan 3.0?
Do I have to sign up?
Is Wan 3.0 free?
Is Wan 3.0 open source? Can I download the weights?
Does Wan 3.0 do 4K?
How long can one clip be?
Does it generate sound?
Can I use the videos commercially?
What happens when a generation fails?
How long does one generation take?
Can I put a real person in a Wan 3.0 video?
How is this different from a PPT-to-video tool?
Every answer is in the page itself — open a question to read it
Make your first Wan 3.0 video
The Wan 3.0 AI video generator at the top of this page is the one paying accounts use — same model, higher ceiling. Miss the shot and it cost you nothing.
- 480P · up to 3 seconds
- Sound in the same file
- An email, no card
- A failed run is never billed
SIGN UP FREE · 1 CLIP · 480PSIGN UP FREE · 1 CLIP ON WAN 3.0 · 480P · 3s · WITH SOUND
Not ready to hand over an email? Three things here need no account at all — the prompt generator, video to prompt and the prompt library. They are free because text is nearly free to generate, and we would rather say that than pretend the video is.
Written and maintained by the TRY WAN3 editorial teamPublished Last updated



