Wan 2.7 Review: Alibaba's AI Video Generator Tested

Overview

Wan is Alibaba's answer to the AI video race, and its strategy is the opposite of everyone else's mystique: give the models away, price the rest like a commodity, and let volume do the talking. Built by Alibaba's Tongyi Lab — the research arm behind the Qwen model family — Wan (from Tongyi Wanxiang) detonated its category in early 2025 by open-sourcing its video models under Apache licensing: full weights, free for anything including commercial use, runnable on a good consumer GPU. Overnight, "AI video" stopped being something you could only rent from San Francisco and became something you could own, fine-tune, and ship — and the open-source video ecosystem that exists today, from ComfyUI pipelines to a thousand niche products, runs substantially on Wan's donated foundations. The official platform at wan.video is Alibaba's own international storefront for the newest generations: a browser studio for text-to-video, image-to-video, text-to-image, and image editing, wrapped around models the community can't yet download.

One housekeeping note this catalog owes readers, given how we handled a lookalike earlier in this series: wan.video is the genuine article — Alibaba's official international platform, verifiable down to its infrastructure — while the model's fame has spawned a swarm of third-party wrapper sites with confusingly similar names selling the same models at marked-up subscriptions. If you're paying for Wan, pay Alibaba or an established API provider, not a domain that merely rhymes.

Why does a trillion-dollar commerce giant give away frontier-adjacent models? The strategy is legible and worth understanding, because it predicts how the platform will treat you. Alibaba runs the same play with Wan that it runs with its Qwen language models — now the most-adopted open model family in the world: open-source aggressively to set standards, win developer loyalty, and commoditize rivals' moats, then monetize the surrounding surface — the cloud that serves the models (every API call through DashScope is Alibaba Cloud revenue), the hosted platform with its newest closed generations, and the enterprise relationships that follow. Domestically it's also an arms race: ByteDance's Seedance and Kuaishou's Kling iterate against Wan weekly in the world's most competitive video-AI market, and that pressure — three giants leapfrogging quarterly — is precisely why this tier of models improves faster than Western release cadences. For buyers the implications are concrete: prices here trend down, not up, because cheap is the strategy rather than a promotion; the newest generation will always sit behind the hosted paywall for a while before any open release; and betting on Wan means betting on an ecosystem Alibaba has structural reasons to keep feeding — the most credible such commitment in the category, and still a commitment, not a contract.

The model line moves fast and the naming matters. The open-sourced 2.1 and 2.2 generations remain the self-hosting workhorses — genuinely capable, especially the larger variants, and free forever. The newer generations available through the platform and APIs pushed into frontier territory: native synchronized audio (dialogue, effects, ambience generated with the picture — a capability the open models lack), 1080p output, multi-shot storytelling with continuity, first-and-last-frame control for dictating exactly where a clip starts and ends, and in-video text rendering in both English and Chinese that most Western rivals still fumble. The current flagship generation has drawn "strongest model of the year" chatter from parts of the community — hype worth discounting, but grounded in a real pattern: on independent comparisons, top Wan models now trade blows with the category's leaders while remaining, by a wide margin, the cheapest serious video generation on Earth. Through APIs, Wan generates at roughly seven to twelve cents per second — an order of magnitude under Western flagship rates — which is why it has quietly become the default engine inside countless budget video products whose users have never heard its name.

Calibrating expectations honestly: Wan's ceiling is high, but the polish crown still belongs elsewhere. Veo's cinematic coherence, Sora-class physics simulation, and Kling's motion realism edge out Wan at the top of the quality range, and Wan's output benefits more than theirs from post-processing hands. The trade Wan offers is different — 80–95% of frontier quality at 5–15% of frontier price, plus the only genuine ownership story in the category. And the familiar considerations for Chinese platforms apply in the proportions we've used throughout this catalog: the hosted platform runs on Alibaba infrastructure under Chinese jurisdiction, moderation reflects that environment, and documentation still occasionally assumes Mandarin — while the open weights, as with DeepSeek, offer the exit no Western frontier lab provides: run it yourself and the jurisdiction question dissolves.

Pricing & Plans

Wan's pricing splits across three surfaces — the official platform's cheap memberships, pay-per-second APIs, and free self-hosted weights — with the structural note that free-tier output is lower-resolution, queued, and not commercially licensed, so the modest paid tiers are the real floor for published work.

Two buying notes: verify plan details against the official site, since regional pricing and credit mechanics shift and third-party wrappers quote inflated numbers as if official — and remember the audio capability lives in the newer hosted generations, so self-hosters should budget for adding sound separately.

Key Features & Capabilities

• Frontier-adjacent generation at commodity prices — Text-to-video and image-to-video that lands within sight of the category's leaders on quality while undercutting them by an order of magnitude on cost. The price-performance ratio, not any single output, is the product — the same play DeepSeek ran on language models, executed on video.
• Native synchronized audio — Newer generations produce dialogue, effects, and ambience in sync with the picture — lip movement matching speech included — collapsing a whole post-production step for social-length content. A genuine differentiator at this price point, where most budget rivals ship silent clips.
• The open-source foundation — Apache-licensed weights for the 2.1/2.2 generations: self-host, fine-tune on your own style or product, build commercial products, run offline. No Western frontier video lab offers this; for developers and studios allergic to vendor lock-in, it's the category's only real answer. The ecosystem it seeded is the multiplier: community LoRA fine-tunes for specific styles and subjects, ready-made ComfyUI workflow graphs, control adapters for pose and motion guidance, and quantized variants that squeeze the models onto consumer hardware — an entire craft tradition, free, that hosted-only rivals structurally cannot grow.
• Production controls — First-and-last-frame conditioning (dictate exactly where a shot begins and ends — the underrated key to stitching coherent sequences), multi-shot generation with continuity, reference-driven consistency, and instruction-based editing. Less famous than rivals' equivalents, comparably useful.
• In-video text rendering — Legible, intentional text inside generated video in English and Chinese — titles, signage, packaging — where most competitors produce alphabet soup. A quiet superpower for ads and social formats built around words on screen.
• The full creative platform — wan.video bundles image generation and editing alongside video in one credit pool and browser studio — a lighter suite than Runway's, at a fraction of its price, adequate for the social-content workflows that dominate real demand.
• Ecosystem gravity — Wan models ship through every major API aggregator, run natively in community tools like ComfyUI, and power a long tail of third-party video products. Skills learned here transfer; assets aren't hostage to one storefront.

Verdict

Our verdict places Wan precisely where its pricing does: the value engine of AI video — the right choice whenever cost, volume, or ownership outranks the last increment of polish, and the wrong one only when that last increment is the job.

The clear yes, in three tiers of enthusiasm. Budget-conscious creators first: for social content, ads, product animations, and the high-volume short-form work that constitutes most actual demand, Wan's paid memberships are almost unreasonably cheap — a commercial license and 1080p output for the price of two coffees monthly, and the Premium tier's allowance at $20 embarrasses what the same money buys elsewhere in this catalog. Developers and product builders second: at seven-ish cents per second through APIs, Wan is the engine that makes video features economically viable in products where Western flagship rates would kill the business case — the same arithmetic that made DeepSeek the budget default in language, playing out again. And self-hosters third, most distinctively: studios, researchers, and tinkerers who want models they own, can fine-tune on proprietary style, and can run without any platform's permission have exactly one serious video option, and this is it — the open weights alone would earn Wan a place in this catalog if the hosted platform didn't exist.

The clear no: productions where top-shelf polish is the deliverable — cinematic brand films, footage cut against live-action, anything where the client is paying for the last 10% — still belong with Veo, Runway's marketplace, or Kling's flagship modes, and pretending otherwise wastes revision cycles that erase the price advantage. Also mismatched: buyers needing clean Western jurisdiction for sensitive material on hosted services (the self-hosting exit exists, but it's real work), teams requiring polished English-first documentation and support SLAs, and anyone whose workflow depends on audio while self-hosting the silent open generations. The moderation boundaries drawn in Hangzhou apply to the hosted platform as they do on Kling — test your actual use case on the free tier before subscribing around it.

The competitive map in one pass, since this tier of the market reshuffles quarterly: Kling is the closest rival in spirit — Chinese, aggressive, consumer-polished — winning on motion realism and image-to-video fidelity at prices above Wan's and below the West's; Seedance presses the same budget flank with periodic leapfrogs, making the three-way Chinese race the segment worth benchmarking as a group; Veo holds the polish-plus-audio crown inside Google's bundle economics; Runway sells the professional pipeline around whichever model wins the shot; and the Sora consumer story has faded into a cautionary tale about assuming any leader is permanent. Against all of them, Wan's position is unduplicated on exactly two axes — lowest serious cost and only true openness — and duplicated everywhere else. Buyers who need one of those two axes have their answer; buyers who don't should run a bake-off, because between Wan, Kling, and Seedance, this month's winner is an empirical question with a shelf life.

Practical playbook: evaluate on the free daily credits for a week against your real prompts — at these prices the trial costs nothing but attention. If you publish, take the ~$5 tier immediately for the commercial license rather than gambling on free-tier terms. Prototype compositions with cheap fast variants and spend the flagship credits on finals. Use first/last-frame control to plan sequences shot-by-shot instead of praying for coherence. For products, route through an established API provider with published rates, benchmark against Kling and Seedance quarterly (this tier of the market leapfrogs constantly), and keep the OpenAI-compatible abstraction thin so switching stays a config change. And if ownership is the point, budget a weekend for the self-hosting rabbit hole — the community documentation is excellent, and the capability compounds.

Weighing it: the category's best price-performance by a distance, the only true open-source story among serious video platforms, native audio and text rendering that punch above the price, and ecosystem gravity that makes the skills portable — against a polish ceiling below the crown-holders, hosted-platform jurisdiction and moderation considerations, documentation gaps, wrapper-site brand confusion, and a naming-and-versioning churn that demands buyer attention. That lands Wan solidly high with a clean identity: not the most beautiful video in the catalog, but the most video per dollar — and, for anyone who wants to own their tools rather than rent them, the only game in town. The self-test is the same one the free tier makes costless: run this week's real briefs through Wan's daily credits alongside your current tool, count the keepers, and divide by the price. For a large share of working creators, that arithmetic ends the comparison — and for the rest, it clarifies exactly what the premium elsewhere is buying, which is its own kind of value.