Run Alibaba's flagship Wan 3.0 in your browser: clips up to 30 seconds with sound already attached, a first-and-last-frame mode that pins exactly where the shot starts and ends, and a reference mode that reads images, video, and audio in one request.
Wan 3.0 is Alibaba's Tongyi Wanxiang 3.0, the successor to Wan 2.7, served here on the alibaba/wan-3.0 endpoint through our video provider. Three modes share one panel: text-to-video from a prompt alone, image-to-video that pins a first frame and optionally a last frame, and reference-to-video that reads up to 10 images, 5 video clips, and 5 audio tracks in a single request, each cited in the prompt by number. Its defining numbers are range: any whole second from 2 to 30, three resolutions from a 480p draft tier to 1080p, and audio generated with every clip.
The trade-offs are specific. 1080p is the ceiling — a 4K deliverable belongs to Kling 3.0. The upstream API defines a smart-duration mode that picks the clip length itself; it reserves credits at the 30-second cap, so we have not wired it up — you choose an explicit length here. Reference video seconds count toward billing, which the price preview shows before you commit. The two clips further down this page took 3m 33s (5 seconds, 720p) and 7m 32s (20 seconds, 720p) to generate here — plan around minutes, not seconds, for long takes.
Alibaba's Tongyi Wanxiang 3.0 generates clips up to 30 seconds with native audio, pins the exact first and last frame of a shot, and reads images, video clips, and audio tracks together in one reference request — billed per second from a 480p budget tier.
Most AI video models stop at 10 or 15 seconds; Wan 3.0 runs to 30 in one continuous generation. That is long enough for a full product walkthrough, a spoken scene, or a sequence of shots without stitching clips together afterwards. Pick any whole second from 2 to 30 — you pay per second, so a 6-second draft and a 30-second final are priced exactly in proportion.
Image-to-video here goes beyond animating a single photo: upload a first frame and, optionally, a last frame, and Wan 3.0 interpolates the motion between your two fixed endpoints. It is the only model in the comparison further down this page that accepts a last frame — the difference between "start from this image" and "get from exactly this image to exactly that one".
Reference-to-video accepts mixed material in one request — up to 10 images, 5 short video clips, and 5 audio tracks — and the prompt addresses each by number: "the character from Image 1, moving like Video 1, scored like Audio 1". Reference images and audio are free; reference video seconds are added to the billed duration, and the cost shows before you generate.
Audio — voices, effects, ambience — is generated with the clip, and turning it off costs the same, so there is never a surcharge for sound. Three resolutions bill in clean proportion: 480p at 40 credits per second for drafts, 720p at 80, and 1080p at 160 for finals. Iterating at 480p and finishing at 1080p keeps a 30-second production affordable.
Generated here on 26 August 2026, first take, exact prompts and settings shown. The 20-second clip exists to prove the long-take claim above — one generation, no cuts, no stitching.
One continuous handheld shot following a food vendor through a night market: she ladles broth at her stall, hands a bowl to a customer, wipes the counter, then looks up smiling as lanterns sway overhead, ambient market chatter and sizzling sounds throughout, no cuts.
A cellist practicing alone in a sunlit rehearsal room, bow moving across the strings, warm resonant cello notes filling the space, dust motes in the light, slow push-in.
It is the long-form and precision model of this group: the 30-second ceiling, the only last-frame control, and the widest reference toolbox.
Product walkthroughs, spoken scenes, and multi-beat sequences run to 30 seconds in one generation — no stitching, no continuity seams between clips, and the audio track spans the whole take.
The first-and-last-frame mode takes two images and interpolates the motion between them — a transformation, a reveal, a camera move that has to land on a specific composition. No other model on this page accepts a last frame.
Feed it the character from one image, the motion style of a video clip, and a music track, and address each by number in the prompt. Ten images, five clips, and five audio tracks can shape a single generation.
The 480p tier bills at a quarter of the 1080p rate with identical prompt behaviour, so you can iterate a 30-second cut at 1,200 credits and render the final at 4,800 only once it is right.
1080p is the ceiling — a 4K deliverable belongs to Kling 3.0. And the upstream smart-duration mode is not wired up here, so you always choose an explicit length between 2 and 30 seconds.
Same workspace, same credit pool — switching between any of these is one click.
| Wan 3.0 | Wan 2.7 | MiniMax H3 | Kling 3.0 | |
|---|---|---|---|---|
| Resolutions | 480p · 720p · 1080p | 720p · 1080p | 768p · 2K | 720p · 1080p · 4K |
| Clip length | 2–30s, any whole second | 2–15s, any whole second | 4–15s, any whole second | 3–15s, any whole second |
| How it is billed | Per second of output; reference video seconds added | Per second of output | Per second of output; reference video seconds added | Per second of output |
| Sound | Native, on by default — on and off cost the same | Native, always on, no toggle | Native, always on, no toggle | Optional toggle, off by default |
| Credits per second, budget tier | 40 at 480p | Not offered — 720p is the floor | 55 at 768p | 45 at 720p, silent |
| Credits per second, 1080p | 160 | 70 | Not offered — its top tier is 2K at 90 | 60 silent · 85 with sound |
| Credits for a 6-second 1080p clip | 960 | 420 | Not offered at 1080p | 360 silent · 510 with sound |
| Credits for one clip, floor to ceiling | 80–4,800 | 80–1,050 | 220–2,820 | 135–3,000 |
| Image-to-video | First frame + optional last frame | First frame only | First frame only | First frame only |
| Reference material | Up to 10 images + 5 videos + 5 audio tracks, cited as Image 1 / Video 1 / Audio 1 | Up to 5 images, cited as Image 1, Image 2 … | Up to 9 images + 3 videos + 3 audio tracks | None — the single image is the first frame |
| Aspect ratios | 5 — 16:9, 9:16, 1:1, 4:3, 3:4 | 5 — 16:9, 9:16, 1:1, 4:3, 3:4 | 6 — 16:9 through 21:9 | 3 — 16:9, 9:16, 1:1 |
| Speed we measured ourselves | 3m 33s for 5s and 7m 32s for 20s, both 720p (2 clips, August 2026) | Not measured | Not measured | Not measured |
Every resolution, clip length and credit figure in this table is read from our own model registry and the rate tables in our code on 26 August 2026 — the same tables that price your generations, not vendor marketing pages. All four models bill per second, so per-clip figures are the rate multiplied by the length you choose; Wan 3.0 and MiniMax H3 additionally count reference video seconds toward billing. The only speed numbers are the two Wan 3.0 clips we generated and timed ourselves; every other cell says Not measured rather than borrowing a number.
Compare Wan 3.0 with other video models — same workspace, one click to switch.
| Model | wan-3.0 (Tongyi Wanxiang 3.0), by Alibaba — served on the alibaba/wan-3.0 endpoint |
|---|---|
| Modes | Text-to-video, image-to-video with first-and-last-frame control, and reference-to-video with mixed image, video, and audio material |
| Resolution | 480p, 720p and 1080p — billed in 1x / 2x / 4x proportion |
| Duration | 2–30 seconds, any whole second (5 by default) — the longest range on this platform |
| Aspect ratios | 16:9, 9:16, 1:1, 4:3 and 3:4 for text- and reference-to-video; image-to-video follows the first frame |
| Reference assets | Up to 10 images, 5 video clips (15 seconds combined), and 5 audio tracks, cited in the prompt as Image 1, Video 1, Audio 1 |
| First / last frame | Image-to-video accepts a strict first frame and an optional last frame; frames and reference material are mutually exclusive upstream |
| Sound | Native — voices, effects and ambience generated with the clip; sound on and off cost the same |
| Credits | Per second of output: 40 at 480p, 80 at 720p, 160 at 1080p — the same rate across all three modes |
| Reference video billing | Reference video seconds are added to the billed duration; reference images and audio are free |
| Credit range per clip | 80 for a 2-second 480p clip, up to 4,800 for a 30-second 1080p clip (plus any reference video seconds) |
| Cancellation | Failed generations refund automatically |
| Measured speed | 3m 33s for a 5-second 720p clip and 7m 32s for a 20-second 720p clip (2 clips, 26 August 2026) |
Open the model picker and choose Wan 3.0. The panel shows its credit cost, resolution options, and duration limits before you spend anything.
Write the subject, the action, and how the camera behaves. Attach a first frame if you want the clip to start from a specific image.
Download the clip, extend it, or send it through further editing. Prompt, model, and settings stay attached in your history.
One account, the full image and video generation workflow.
Run Wan 3.0 on the free tier's daily credits, or upgrade for premium models, faster generation, and 4K output.
Perfect for getting started
Billed annually · $240/year
Billed annually · $380/year
Long-form takes and controlled shots: clips up to 30 seconds with native audio, a first-and-last-frame mode that pins exactly where a shot starts and ends, and a reference mode that reads images, video, and audio together in one request.
Per second of output: 40 credits at 480p, 80 at 720p, 160 at 1080p — the same rate for text-, image-, and reference-to-video. Reference video seconds are added to the billed duration; reference images and audio are free. The exact cost shows before you generate, and failed runs refund automatically.
Text-to-video from a prompt alone, image-to-video from a first frame (with an optional last frame), and reference-to-video with up to 10 images, 5 video clips, and 5 audio tracks cited in the prompt as Image 1, Video 1, Audio 1.
2 to 30 seconds, any whole second — the longest range of any video model on this platform. Sound is generated with the clip at every length.
Wan 3.0 doubles the length ceiling (30s vs 15s), adds a 480p budget tier and last-frame control, and extends references from 5 images to mixed image + video + audio material. Wan 2.7 remains the lower-rate pick at 720p (40 vs 80 credits per second) for shorter multi-shot work.
Credits are refunded automatically for failed video generations — you only pay for clips that finish.
One-click tools for everything around generation — polish, restyle, and share.
Wan 3.0 (Alibaba Tongyi Wanxiang 3.0) integrated — clips up to 30 seconds with native audio, first-and-last-frame control, and reference-to-video that reads up to 10 images, 5 video clips and 5 audio tracks in one request. 480p/720p/1080p at 40/80/160 credits per second.
