brain/
← all entities
entitygenericai-video-generation

Wan 2.7

Notes

Wan 2.7

One-line summary: Alibaba Tongyi Lab's April 2026 general video suite (t2v / i2v / r2v / videoedit), 2–15 s at 720p/1080p; still the Wan row on the 2026-08-18 AA leaderboards, superseded in length by wan-3-0.

What it is

A four-model hosted suite that unifies text, image, video, and audio for generation, editing, continuation, and referencing. Official Community post (7 Apr 2026): 2–15 s, 720p/1080p, natural-language edits (action, dialogue, appearance, camera), lip-sync when dialogue is rewritten, and "cross-video consistency for up to five distinct characters." From 2026-08-18-autoresearch-best-ai-video-generation-tools.

Why it matters to ai-video-generation

This is the general-video successor to the thread's wan-2-2 parent family. Reference-to-video plus five-character consistency is the commercial-line analogue of wan-animate's identity lock — but this source does not re-benchmark Wan-Animate.

Key facts

  • Vendor: Alibaba (Tongyi Lab); available on Model Studio and the Wan site / Qwen App.
  • Models: Wan2.7-t2v, Wan2.7-i2v, Wan2.7-r2v, Wan2.7-videoedit.
  • AA T2V with audio (2026-08-18): Wan 2.7 Elo 1,109 / $9.00/min; variant Wan2.7-260612 Elo 1,157 / $9.00/min (4th overall).
  • AA I2V with audio: Wan 2.7 Elo 1,090 / $9.00/min.
  • License of weights is not stated in the official Community posts fetched this run (do not assume Apache 2.0 from wan-2-2).

Strengths (from our perspective)

  • Strong mid-frontier Elo at $9/min — well below Veo 3.1 Standard and Kling 3.0 Pro on AA's meter.
  • Editing and multi-character consistency are first-party claims, not SEO blogs.

Weaknesses (from our perspective)

  • 15 s cap; wan-3-0 doubles that (vendor).
  • No independent avatar-task scores.
  • Weight license unverified in this source.

Sources

Related

Referenced by