Wan 2.5 文本转视频

About Wan 2.5 Text to Video

Wan 2.5 text to video turns a written scene into a five- or ten-second clip with no image to upload. The page runs Alibaba's model through WaveSpeed and exposes the controls that change the result: frame size, duration, negative prompt, prompt expansion, seed and an optional audio track. The credit charge follows the frame and the length you choose, and the form shows it before the request is sent. The form opens at 1280×720 for five seconds, which puts a first run at 640 credits unless you move either control.

There is no ratio menu here, only six pixel sizes: 832×480 and 480×832, 1280×720 and 720×1280, 1920×1080 and 1080×1920. Every one is landscape or portrait, so a square clip is not available on this page. The pairs are separate output sizes rather than a rotation applied afterwards, which means a vertical edit should start at 720×1280 instead of cropping a 1280×720 render. The price ladder matches the sizes. Five seconds cost 320 credits at 480p, 640 at 720p and 960 at 1080p; a ten-second clip costs exactly double. The estimate on the form recalculates when either control changes, so the figure next to the generate button is the figure charged.

Prompts are trimmed to 2,000 characters on the server, which means a long script loses everything past that point; keep the shot description inside that budget. A negative prompt is optional and takes the same limit, and it is the practical place to name flicker, warped hands or washed-out color. Prompt expansion is off by default and lets the model enrich the prompt with cinematic detail, so switching it on trades control of the wording for more added description. The seed defaults to -1, meaning random; a fixed integer makes a run repeatable when the prompt, frame and length are unchanged. The helper beside the field notes that reusing a seed keeps pacing consistent between takes, which matters when two versions of the same shot have to cut together.

The optional audio field accepts MP3, WAV, OGG or M4A files up to 40 MB. The panel describes it as input for aligning motion and lip-sync, which suits a voiceover or a track the clip has to sit on rather than an unrelated extra. Render estimates come from the endpoint and climb with frame size: roughly four minutes at 480p, six at 720p and seven at 1080p, shown as a countdown once the request is accepted. The countdown is a budget rather than a deadline, and the panel keeps working through the final packaging after it reaches zero. The Fast page drops the 480p tier, charges less per five seconds at 720p and 1080p, and reports a shorter estimate.

How Wan 2.5 Text to Video works

1. 1

   Choose a frame

   Six pixel sizes, all landscape or portrait. The pair you pick fixes the bill for a five-second run: 320 credits at 480p, 640 at 720p, 960 at 1080p. Start at 1280×720 if you have no target platform yet.

2. 2

   Write the prompt and set the controls

   Keep the prompt inside 2,000 characters. Add a negative prompt if the shot keeps picking up the same artifact, decide whether expansion stays off, and enter a fixed seed if the run has to be repeatable. Attach narration or music only when the clip must follow it.

3. 3

   Set the length and generate

   Five or ten seconds; ten doubles the charge. Confirm the estimate, submit, and watch the countdown. The result opens in the preview panel, can be enlarged, and downloads as an MP4.

What Wan 2.5 Text to Video is best for

• Vertical social edits

  480×832 at 320 credits for five seconds is the cheapest way to test a hook in the frame a feed actually shows, and 720×1280 is the step up when the draft holds.

• Repeatable character shots

  Fix the seed and the prompt, and a look can be rendered again later instead of described from memory. The same seed also keeps pacing close enough that takes can be compared honestly.

• Voiceover-led clips

  An uploaded track gives the model a timing reference for motion and lip-sync, which is the case the audio field exists for. Files up to 40 MB are accepted in four common formats.

• Draft low, finish high

  Block the scene at 480p, then re-run the same prompt at 1080p once the composition works; the charge scales with the frame, not with the number of drafts you made on the way.

Wan 2.5 Text to Video at a glance

Model

Alibaba Wan 2.5 text to video, run through WaveSpeed

Frames

832×480, 480×832, 1280×720, 720×1280, 1920×1080, 1080×1920; no square

Duration

5 or 10 seconds

Credit cost

320 per 5 seconds at 480p, 640 at 720p, 960 at 1080p; 10 seconds doubles

Controls

Prompt and negative prompt up to 2,000 characters, prompt expansion, seed, optional audio up to 40 MB

Output

MP4, downloaded from the preview panel

Wan 2.5 Text to Video FAQ

How much does a Wan 2.5 text to video run cost?

Five seconds cost 320 credits at 480p, 640 at 720p and 960 at 1080p, and ten seconds cost twice that. The form shows the exact figure before submission, and a run the provider fails to return is refunded. New accounts start with 300 credits and can claim 50 more each day they log in, both valid for 30 days; Starter is $4.90 a month for 300 credits.

Can I add my own audio?

Yes, optionally. The field takes MP3, WAV, OGG or M4A files up to 40 MB and is described as a reference for motion and lip-sync, so narration or a track with a clear beat is the intended input. Leaving it empty is the normal case and does not change the charge, which depends only on frame and length.

Why is there no square aspect ratio?

The page offers six fixed pixel sizes and all of them are landscape or portrait. Square output is not among them, so a square post means cropping in the edit rather than generating at 1:1.

How do I reproduce a clip I liked?

Write down the prompt, frame, duration and seed, and note whether prompt expansion was on. The seed field takes -1 for random or a fixed integer; with the other settings unchanged, that is what makes the run repeatable. Keep the same negative prompt as well, since it is part of the request.