Wan 3.0: 30-Second Shots in a Single Generation

Wan 3.0 on VdoBloom is Alibaba’s newest Wan video model, and its headline feature is duration: one generation can run anywhere from 2 to 30 seconds, in one-second steps, at 480p, 720p or 1080p, with native audio — which is double the 15-second ceiling of the Wan 2.6 and Wan 2.7 video models it sits next to. It costs credits per second and per resolution, from 7 credits for a 2-second 480p clip to 384 credits for 30 seconds at 1080p, and the exact price appears on the button before you generate.

Updated October 2026.

Disclosure: VdoBloom is our platform.

What Wan 3.0 is

Wan 3.0 is the newest generation in Alibaba’s Wan line on VdoBloom’s video tools, released in August 2026. It collapses what used to be two separate jobs — text-to-video and image-to-video — into a single model with a single settings panel, so you pick a duration, a resolution and an aspect ratio and the input type follows from whether you supplied a source image.

Three things define it. First, the duration range: every whole second from 2 to 30 is selectable, which is unusual granularity and a real ceiling change, since the Wan 2.6 and Wan 2.7 video models both stop at 15 seconds. Second, native audio: sound is generated in the same pass as the picture rather than added afterwards. Third, a per-second price that does not penalise the longer ceiling — at a matching 5-second 720p clip, Wan 3.0 costs the same 32 credits as Wan 2.7.

As with the other recent Wan releases, Alibaba has not published Wan 3.0’s weights, so VdoBloom’s hosted access is the practical way to run it — no install and no GPU. It carries the platform’s HOT content label, the most permissive filtering tier, with the usual platform-wide rules on illegal content still applying.

What it is good at

• Shots that develop. A 30-second ceiling means a shot can build — a walk that becomes a run, a slow reveal, a product demonstration with a beginning and an end — instead of landing in a single beat.
• Long takes without cutting. Because the length is one generation rather than several clips stitched together, there is no seam in the middle and no match-cut to hide.
• Sound in the same pass. Dialogue-free ambience, footsteps, room tone, traffic and music beds arrive attached to the clip.
• Either input, one model. Start from a prompt or from an image without switching models or learning a second settings panel.
• Cheap first drafts. A 2-second 480p clip is 7 credits, which makes prompt iteration genuinely affordable before you commit to a long render.

Wan 3.0 credit costs

Pricing scales with both duration and resolution, in one-second steps. These are the numbers VdoBloom charges, taken from the platform’s own price table.

Duration	480p	720p	1080p
2 seconds	7	13	26
5 seconds	16	32	64
10 seconds	32	64	128
15 seconds	48	96	192
20 seconds	64	128	256
30 seconds	96	192	384

In per-second terms that is roughly 3.2 credits per second at 480p, 6.4 at 720p and 12.8 at 1080p. A new account starts with 10 free credits and no card, which covers a 2-second 480p test and leaves a little over; the pricing page lists the plans, including Lite at $15 a month for 300 credits (about $0.05 a credit), or $10 a month billed yearly, and one-time credit packs from $10.39 for 200 credits that never expire.

Wan 3.0 Video Prime

Prime is the same capability set — same 2 to 30 second range, same three resolutions, same five aspect ratios, same native audio — with faster generation, and the whole documented difference is speed. That speed is priced: Prime runs roughly 55 to 60 percent more credits per second, so a 5-second 720p clip is 51 credits on Prime against 32 on the base model, and a 30-second 1080p clip is 605 against 384.

The sensible workflow is to draft on the cheaper base model to lock the prompt, duration, resolution and framing, then switch to Wan 3.0 Video Prime for the final delivery pass if turnaround matters. Since the cost is shown before every generation, the premium is a deliberate choice rather than a surprise.

Example prompts

The service accepts prompts between 2 and 20,000 characters, which is room to describe a whole shot rather than a caption. Five worked examples:

• Single continuous take, 30 seconds. “A single unbroken take: a chef walks into a small kitchen, ties an apron, sets a pan on the hob and turns to camera. Natural window light from the left, handheld camera slowly pushing in. Ambient kitchen sounds: the hiss of the pan, a radio in the next room, footsteps on tile.”
• Product reveal, 10 seconds. “A matte black watch on a dark stone surface. The camera orbits slowly from the left as a soft key light sweeps across the case. Tight product-photography framing, shallow depth of field. Sound: a low room tone and a single soft click as the second hand moves.”
• Landscape, 15 seconds. “A wide valley at sunrise, mist sitting in the treeline. The camera drifts forward at walking pace. Golden light rakes across the hillside. Sound: birdsong close, wind in grass, no music.”
• Image start, 20 seconds. “Use the uploaded photograph as the opening frame. Hold the framing for the first two seconds, then slowly tilt up to reveal the sky as the light changes from grey to gold. Keep the building’s architecture exactly as in the still. Sound: distant traffic and wind.”
• Deliberate slow beat, 8 seconds. “A girl at a bus stop looks up from her phone as the bus arrives. Static camera, 35mm, overcast light, rain on the shelter glass. Sound: rain on glass, an engine idling, a door hissing open.”
• Iteration prompt, 2 seconds. “Test shot: a hand turns a brass tap, tight framing, hard side light.” A cheap 480p test that tells you whether the motion and framing behave before you pay for a long render.

Limits worth knowing

• Duration: 2 to 30 seconds, in one-second steps. There is no 1-second option and no rounding to scene length.
• Resolutions: 480p, 720p and 1080p only. There is no 4K tier for Wan 3.0.
• Aspect ratios: five — 16:9, 4:3, 1:1, 3:4 and 9:16.
• Prompt length: between 2 and 20,000 characters.
• No reference-image syntax. Multimodal reference inputs exist upstream but are not exposed on VdoBloom yet, so a prompt refers to no numbered reference images.
• No timed multi-shot prompts. The platform documents timestamped, shot-by-shot prompt formats only for the Seedance 2 family, so a Wan 3.0 prompt describes one continuous shot.
• Frame control: Wan 2.7 Image-to-Video keeps a first- and last-frame control that Wan 3.0 does not list. If you need to pin an exact ending frame, that older model is the right tool.

Frequently asked questions

How long can a single Wan 3.0 generation be?

Thirty seconds, in one-second steps — so you can ask for 7, 13 or 27 seconds rather than rounding to a preset length. That is double what Wan 2.6 and Wan 2.7 allow, and it is the main reason to choose Wan 3.0 over them.

Does Wan 3.0 generate sound?

Yes. Audio is generated natively in the same pass as the picture, at no separate charge, and the per-second rate above is the whole price. There is no audio toggle, no dialogue field and no music picker on the model — the prompt is how you steer the sound.

Is Wan 3.0 more expensive than Wan 2.7?

Not at matching settings. A 5-second 720p clip is 32 credits on both. What changed is the available range: Wan 3.0 adds a 480p tier at roughly 3.2 credits a second and extends the ceiling to 30 seconds, where a 1080p clip is 384 credits.

What is the difference between Wan 3.0 and Wan 3.0 Video Prime?

Generation speed, and the price of it. Everything else is identical. Prime costs roughly 55 to 60 percent more credits per second, so it is worth it for a final pass or a tight deadline, and rarely worth it while you are still exploring prompts.

Can I use the output commercially?

Paid output on VdoBloom is watermark-free and comes with commercial rights. Check the platform’s terms for the specifics of your plan, and remember that the underlying model is hosted rather than open-weight, so the licence you rely on is the one you are given at generation time.

Open Text to Video, pick Wan 3.0, and start short: a 2-second 480p test costs 7 credits and tells you more about your prompt than any amount of rewriting. Lock the shot on the cheap settings, then climb the duration and resolution ladder one step at a time. If you are weighing it against another family, WAN vs Seedance is the direct comparison, common prompt mistakes is worth reading before a long render, and Wan 2.7 Image-to-Video is the model to open instead when a pinned final frame matters more than length.