Skip to content
Bud's Wiki

Wan 2.5 live on WaveSpeed — 5s/10s clips up to 1080p (AI Pulse correction tweet)

TL;DR

Section titled “TL;DR”

A short correction tweet from AI Pulse (@youraipulse) on 23 Sep 2025 noting that Wan 2.5 — which AI Pulse had previously teased as “coming tomorrow” via an Alibaba 2.5-Preview leak — was already serving via the WaveSpeed AI API: 5-second and 10-second clips, up to 1080p. The tweet itself is a one-line product-availability nudge; the underlying release (Wan 2.5) is Alibaba’s natively multimodal text/image/video/audio model with one-pass A/V synchronization, positioned as an open-API competitor to Veo 3.

Key claims

Section titled “Key claims”
  • Wan 2.5 is reachable via WaveSpeed AI’s API at the time of the tweet, supporting 5s and 10s generation durations and up to 1080p output [tweet body].
  • AI Pulse’s prior tweet (quote-RT’d here) had teased Wan 2.5-Preview as “coming tomorrow” based on an Alibaba teaser image — this tweet is the correction that it’s already live [tweet body, quote-RT].

Method

Section titled “Method”

Product-availability tweet, no methodology. The substantive content is in the WaveSpeed product page and accompanying launch posts (linked under “See also”), which describe the underlying Wan 2.5 model: a unified text/image/video/audio backbone with one-pass A/V synchronization (lip-sync + sound effects + background audio), 480p / 720p / 1080p at 24fps, durations up to 10s, six aspect-ratio options, and multilingual prompting (English + Chinese particularly strong).

Results

Section titled “Results”

Not applicable — this is a tweet announcing API availability, not a release with measured results. WaveSpeed’s own marketing materials position Wan 2.5 as “up to 3× cheaper than Google Veo 3” with comparable A/V-sync quality, but no third-party benchmarks are cited in the tweet itself.

Why it’s interesting

Section titled “Why it’s interesting”

Wan 2.5 is the version of the Alibaba Wan video stack that ships native one-pass audio-video generation — placing it in the same architectural slot as the joint-A/V work Luma already tracks. It complements the open-weights UniForm: A Unified Multi-Task Diffusion Transformer for Audio-Video Generation (UniForm — unified A/V diffusion transformer) and Ovi: Twin Backbone Cross-Modal Fusion for Audio-Video Generation (Ovi — twin-backbone cross-modal fusion) by being the closed-API datapoint for the same capability at frontier scale, and contrasts with UniVerse-1: Unified Audio-Video Generation via Stitching of Experts (UniVerse-1 — A/V via stitching of separately-trained experts) on the unified-vs-stitched axis. On the release-trajectory side, it slots between Wan2.2: World's First Open-Source MoE Video Generation Model (launch announcement) (Wan2.2 — MoE video, no audio) and Wan 2.7 upcoming features preview (video editing, V2V, real-time) (Wan 2.7 preview — V2V editing, real-time) in the Wan family’s continued-development lineage; the fact that 2.5 ships first as a WaveSpeed-hosted API rather than an open-weights drop (unlike 2.2) is itself a signal worth tracking for Open foundation-model releases.

See also

Section titled “See also”