What is Wan 2.5?

Wan 2.5 is Alibaba's (Tongyi / Wan team) video generation model, released in September 2025. It turns a text prompt — or a still image — into a clip up to 10 seconds long at up to 1080p (24fps), and generates the audio in the same pass: dialogue with lip-sync, ambient sound effects, and music. That one-pass audio-visual output is what separates it from earlier Wan releases and from most open video models, which produce silent video only.

Wan 2.5 ships as an API-only preview — its weights are not publicly downloadable. (Wan 2.1 and 2.2 are the open-weight, Apache-2.0 versions you can self-host; 2.5 and 2.6 are API-only.) On GPTProto, you call it with the model string wan-2.5, billed per generation by resolution and duration, on the same balance as 200+ other models.

What you can build with the Wan 2.5 API

Short-form social and ads — 5–10s vertical or landscape clips with built-in voiceover and sound, no separate audio editing. At $0.45 per 720p/5s run, A/B-testing a dozen ad variants stays cheap.

Localized video at scale — Wan 2.5 handles multilingual prompts and generates lip-synced speech, so one storyboard becomes versions in several languages without re-shooting or re-dubbing. Strong Chinese support.

Storytelling and explainers — Dialogue, ambient sound, and music generated together keep narration and visuals aligned across the clip — useful for YouTube intros, product explainers, and training segments.

Animating a still image — Feed a single image as the first frame and a motion prompt to animate it.

How Wan 2.5 generates video and audio together

Most video models output silent frames; sound is a separate job you bolt on later. Wan 2.5 was built the other way around. Its generation step produces the visual frames and a matching audio track jointly, so speech lands on the right lip movements, footsteps line up with steps, and music sits under the cut without you editing anything. In practice this collapses a two-tool pipeline (video model + TTS/foley) into one API call.

Two things follow from that. First, your prompt should describe sound as well as motion — name the dialogue line, the ambient bed, and whether you want music (see §6). Second, because audio and video come from one pass, a single run is one billable unit; you are not paying twice or stitching tracks. If you need to supply your own music or voiceover instead, the audio parameter accepts it and the model syncs motion to it.

Choosing inside the Wan family

GPTProto carries several Wan models on one balance. Pick by what the job needs, not by version number:

• Wan 2.5 — text-to-video (this page): 1080p, up to 10s, native audio. The default for a prompt-to-clip with sound.
• Wan 2.5 — image-to-video: start from a still image as the first frame and animate it (the "wan 2.5 animate" case). 
• Wan 2.2 (wan-2.2-plus): silent, 720p, ~5s — but open-weight, so it's the pick when you must self-host.
• Wan 2.6 (wan-2.6): 1080p, up to 15s, plus multi-shot scene planning and reference-based identity retention — step up when you need longer or multi-cut narratives.

This map is the kind of thing competitors leave out: they list models without telling you which to reach for. Use 2.5 for single-shot clips with audio, 2.2 if you need weights, 2.6 for length and multi-shot.

Wan 2.5 vs Sora 2

Both generate video with synchronized audio from text or an image. The practical differences:

Honest take: Sora 2 is a flat $0.4/run and supports longer clips (up to ~25s on Pro) — reach for it when duration matters. Wan 2.5 starts lower at $0.225/run (480p/5s) and lets you trade resolution for cost, with strong multilingual speech and fewer content restrictions. Both run on one GPTProto balance.

Related: Sora 2 → · Wan 2.6 → · Wan 2.2 →

Wan 2.5 prompt recipes

Wan 2.5 generates audio with the video, so good prompts describe sound as well as motion. The platform's own examples put audio inline with a Sound: cue — follow that convention. Structure: subject + action + camera move + lighting + Sound: cue + style. Paste and adjust.

1. Dialogue + lip-sync (shows the audio engine)

Medium close-up of an elderly fisherman on a wooden boat at dawn, soft golden
light, gentle water lapping. He looks to camera and says warmly, "The sea always
keeps its promises." Slow push-in. Sound: soft waves, distant seagulls, a calm
spoken line, no music.

2. Ambient / SFX, no dialogue

Rain on a neon-lit Tokyo alley at night, a cat darts under an awning, reflections
shimmer on wet pavement. Static wide shot, cinematic. Sound: steady rain, distant
traffic, no music, no speech.

3. Product / marketing

Close-up of an iced coffee on a marble counter, condensation beading, morning
kitchen light. A spoon stirs the ice. Slow 180-degree orbit around the glass.
Sound: soft clink of ice, quiet ambient kitchen tone, light background music.

4. Motion / tracking

A skateboarder rolls across an empty concrete plaza at sunset, low-angle tracking
shot alongside, subtle lens flare. Sound: wheels rumbling on pavement, wind
ambience, no dialogue.

Tips: keep dialogue short enough for the clip length (about one spoken line per 5s); name the camera move explicitly; state "no music" if you only want ambient sound; or pass your own track via the audio parameter.