Wan 2.7 Is Here — What Alibaba's Newest Open Video Model Actually Adds

Tongyi Lab shipped Wan 2.7 in April 2026, the third major jump since Wan 2.2 — Wan 2.6 landed first in December 2025 with multi-shot storytelling and clone-level character consistency, then 2.7 added Thinking Mode, a planning pass that interprets and structures a prompt before generation starts. Both releases stayed true to Wan's open-weights approach: 2.7 ships at 27B total parameters (14B active via a mixture-of-experts architecture) under Apache 2.0.

Generate with Wan 2.2

What's actually new since 2.2

Wan 2.6 introduced native audio-visual sync and up to 15-second single-run generation, enough for a complete narrative beat instead of a short loop. Wan 2.7 builds on that with Thinking Mode (prompt planning before rendering), first-and-last-frame control, multi-reference consistency across a whole generation, voice cloning for reference-to-video output, and instruction-based video editing — text-to-video, image-to-video, and editing now sit in one suite instead of separate tools.

Where Uncutly's Wan lineup stands today

Uncutly's video pipeline runs Wan 2.2 across its templates — the budget-friendly, high-volume tier the platform is built around for fast NSFW iteration. Wan 2.6 and 2.7 endpoints exist in the underlying Novita infrastructure Uncutly uses, but no template is currently configured against either version.

Why the jump would matter for uncensored video

Thinking Mode and multi-reference consistency both attack the exact failure points that show up most in NSFW generation today: a character's face, outfit, or pose drifting across a sequence, or a prompt getting only half-followed because the model didn't parse every constraint. Voice cloning would also open reference-to-video with matched audio, a combination most adult-content generators don't offer at all. An open-weights, Apache 2.0 release is also the kind of model uncensored platforms can typically adopt without waiting on a vendor content policy.

What we're watching

Wan has shipped a new major version roughly every few months since 2.2 — 2.5, then 2.6, then 2.7 — which makes it one of the fastest-moving open video lines available. We track each release as Novita brings the endpoint online and evaluate it against Uncutly’s existing Wan 2.2 templates before any migration.

FAQ

Does Uncutly.AI run Wan 2.7?

Uncutly's templates currently run Wan 2.2. Wan 2.6 and 2.7 endpoints exist in the underlying infrastructure but aren't wired into any live template yet.

What's the difference between Wan 2.6 and Wan 2.7?

Wan 2.6 (December 2025) added multi-shot storytelling and up to 15-second native generation. Wan 2.7 (April 2026) added Thinking Mode prompt planning, voice cloning, and instruction-based video editing on top.

Is Wan 2.7 open source?

Yes — Wan 2.7 ships under Apache 2.0 at 27B total parameters (14B active via mixture-of-experts), continuing the open-weights approach of earlier Wan releases.

Can I generate NSFW video with Wan on Uncutly right now?

Yes — Wan 2.2 templates on Uncutly run uncensored for verified 18+ creators generating fictional AI characters, at the lowest credit cost of any video model on the platform.