Alibaba's newest video generation model, available on SharkFoto in Text to Video and Image to Video (Standard and Prime). Pick any length from 5 to 15 seconds and a 480P, 720P or 1080P render, start from a prompt in one of five framings or from a first and optional last frame, and keep native audio on or switch it off.
On SharkFoto, set any whole number of seconds from 5 to 15. Earlier Wan models on SharkFoto only offered fixed 5, 10 and 15 second steps, so a 7-second cut meant generating 10 and trimming. Here you ask for 7 and pay for 7.
16:9, 9:16, 1:1, 4:3 and 3:4 cover landscape, vertical, square and classic-camera shapes in SharkFoto's Text to Video. Alibaba's API adds 21:9 and Adaptive, which hands the framing decision to the model instead of pinning it to a fixed ratio.
Wan 3.0 adds a 480P option below the usual HD tiers. Because credits scale with resolution, a 480P pass is the cheapest way to test whether a prompt reads the way you intended before committing to a 1080P render.
SharkFoto's prompt field accepts up to 2,500 characters - 500 more than Wan 2.6 here - and Alibaba's API takes up to 20,000. That is enough to spell out subject, wardrobe, lighting, lens, camera move and pacing in one go instead of compressing the whole idea into a sentence.
Wan 3.0 is the current generation of Alibaba's Wan video model family, released publicly in August 2026. Through Alibaba's API it covers text-to-video, image-to-video, all-in-one reference generation and video editing, with clips of up to 30 seconds and native audio. On SharkFoto it runs in Text to Video and Image to Video: you write a prompt - plus a first frame and an optional last frame for image-to-video - choose the resolution, length and (for text-to-video) the shape of the clip, and the model renders the whole thing in one pass, with native audio on or off. No storyboard, no editing timeline.
On SharkFoto, the practical difference from Wan 2.6 is control granularity. Duration is a continuous 5-15 second range rather than a set of preset buttons, the aspect ratio list grows from two options to five, and a 480P tier sits underneath 720P and 1080P so that iteration costs a fraction of delivery. The prompt window is larger as well, which matters more than it sounds: a video prompt has to carry subject, motion and camera direction all at once.
Even from a single prompt, Wan 3.0 plans the shot itself and holds subject appearance, spatial layout and motion consistent across the clip. That makes it a good fit for short social cuts, product and marketing b-roll, and quick visual concepts - anywhere the idea exists as a description or a still image rather than as footage you already have.
On SharkFoto, duration is a slider across a 5-15 second range (Alibaba's API goes up to 30), and every whole second is a valid request. A 6-second loop for a feed, an 8-second product beat, a 15-second full cut - each is asked for directly rather than trimmed out of a longer render.
Five fixed ratios cover the formats delivery actually requires: 16:9 for landscape and YouTube, 9:16 for Reels, Shorts and TikTok, 1:1 for feed posts, 4:3 and 3:4 for the more photographic shapes - all five are available in SharkFoto's Text to Video. Alibaba's API adds 21:9 for cinematic widescreen and Adaptive, for when the prompt should decide the frame.
480P, 720P and 1080P are separate renders, not an upscale chain. Credits scale with the tier, so the same prompt can be explored cheaply at 480P and then re-rendered once at 1080P when the wording is settled.
A long prompt window - up to 20,000 characters through Alibaba's API and 2,500 on SharkFoto - lets you direct rather than describe: subject and wardrobe, the environment and time of day, the lighting, the lens, how the camera moves, and what changes between the first and last second of the clip.
Wan 3.0 is designed to hold character detail, spatial relationships and motion coherent for the whole duration rather than drifting as the clip runs on, so faces, props and background geometry stay recognisable from the first frame to the last.
The model handles multi-shot structure natively, so a prompt that describes a small sequence does not have to be split into separate generations and cut together by hand afterwards.
| Resolution | 480P, 720P or 1080P (default 720P) |
|---|---|
| Clip Length | 2-30 seconds via API; 5-15 seconds on SharkFoto, any whole second (default 5s) |
| Aspect Ratios | Adaptive, 21:9, 16:9, 9:16, 1:1, 4:3, 3:4 via API; 16:9, 9:16, 1:1, 4:3, 3:4 on SharkFoto text to video (default 16:9) |
| Generation | Single pass - no stitching or upscale step |
| Prompt | Required; up to 20,000 characters via API, 2,500 on SharkFoto |
|---|---|
| Reference Image | API: first/last frame or reference (up to 20 references in total); SharkFoto Image to Video: first frame + optional last frame |
| Reference Video | API: yes (reference, editing, extension); not on SharkFoto |
| Reference Audio | API: yes (as a reference); not on SharkFoto |
| Documents | API: documents and web pages as references; not on SharkFoto |
| Tool | AI Video - Text to Video, Image to Video |
|---|---|
| Versions | Standard and Prime (Prime uses more credits) |
| Developer | Alibaba |
| Modes | Text to video; image to video from a first frame (optional last frame); native audio on or off |
| Credits | Scale with both resolution and clip length |
| Other Modes | Use Wan 2.6 for video-to-video |
Generate 9:16 clips sized for Reels, Shorts and TikTok without cropping a landscape render. Short durations keep the cost of testing several hooks low, and the same prompt can be re-run at 1:1 or 16:9 for the other placements.
Fill the gaps in an edit - an establishing shot, a texture close-up, an ambient transition - without a shoot. Ask for the exact number of seconds the timeline needs rather than trimming a longer clip down.
Test whether an idea reads on screen before anyone books a camera. Iterate the wording at 480P where a pass is cheap, then render the version that works at 1080P.
A few seconds of motion at the top of a page or post does more than a static hero image, and 4:3 or 3:4 suits editorial layouts that a 16:9 clip would not sit in cleanly.
One written concept, rendered into landscape, square and vertical versions, so a campaign ships in every placement without re-framing work in an editor.
Both Wan models are available here. This is what actually differs between them in the generator, not on paper.
| Feature | Wan 2.6 | Wan 3.0NEW |
|---|---|---|
| Clip length | Fixed 5s, 10s or 15s | Any whole second, 5-15s |
| Resolution | 720p, 1080p | 480P, 720P, 1080P |
| Aspect ratios | 16:9, 9:16 | 16:9, 9:16, 1:1, 4:3, 3:4 (text to video) |
| Prompt limit | 2,000 characters | 2,500 characters |
| Draft tier | None below 720p | 480P |
| Generation modes | Text, image and video to video | Text and image to video (first + optional last frame) |
| Credit model | Preset duration steps | Charged per second |
On SharkFoto, Wan 3.0 takes a written prompt plus, in Image to Video, a first frame and an optional last frame. Alibaba's all-in-one reference mode (up to 20 images, videos, audio clips, documents or web pages) and its video editing are not available here, so work that has to start from existing footage belongs on Wan 2.6's Video to Video.
Alibaba's API allows up to 30 seconds per generation and can continue a clip with video extension, but on SharkFoto a single generation tops out at 15 seconds and extension is not offered. Longer pieces have to be assembled from multiple clips in an editor, and continuity between separate generations is not guaranteed.
Credits scale sharply with resolution: a 1080P clip costs four times the same-length 480P clip, and 720P costs twice as much. Iterating at 1080P gets expensive quickly - settle the prompt at 480P first.
Alibaba notes that audio texture and on-screen text rendering accuracy are still being refined, so check dialogue, sound and any lettering in the frame before publishing a clip.
Describe the shot or start from an image, set the length to the second, pick your framing, and let Wan 3.0 render it with native audio. Start at 480P to find the prompt, finish at 1080P.
Try Wan 3.0 now