Tongyi Wanxiang 通义万相
Alibaba's open-weight image/video generator — now on Wan 2.7
Overview
Tongyi Wanxiang is Alibaba Cloud's text-to-image and video model family (Wan 2.1 through April 2026's Wan 2.7), with open Apache 2.0 weights plus a hosted DashScope API. It spans styles from watercolour to anime in Chinese and English, billed pay-as-you-go, not subscription.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 4 Sep 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
What you can produce with Tongyi Wanxiang 通义万相
- A 4K print-ready image (up to 4096x4096) generated from a text prompt, with precise HEX brand colours locked and multilingual text rendered accurately in up to 12 languages (Wan 2.7 cloud/API only) — including Chinese, Arabic, and Korean
- A 12-panel storyboard sequence with consistent character identity across all panels, produced from a set of reference images using Wan 2.7's multi-reference fusion and batch generation
- A self-hosted, commercially deployable generation pipeline using the open Apache 2.0 weights (Wan 2.1/2.2 only — Wan 2.7 has no open weights); the 1.3B text-to-video model needs ~8.19 GB VRAM, so an 8 GB RTX 3070 is borderline and requires VRAM offloading/optimization — zero per-image API fees
- A short narrative video clip (up to 15 seconds at 1080p) with synchronised audio, subject-voice cloning, and first/last frame control for cinematic scene transitions
- Style-transferred marketing visuals in traditional Chinese ink-wash, retro, or 3D cartoon aesthetics generated from a product sketch or reference photo
- An instruction-edited version of an existing video asset — changing background, lighting, or object appearance via text command without re-generating from scratch
- Batch-generated consistent character assets (up to 12 images per run) for a game, animation, or brand campaign, maintaining skeletal pose and identity across varied angles and lighting
Tongyi Wanxiang has grown well past a text-to-image tool: Wan 2.1 topped VBench, the open-weight Wan 2.1/2.2 releases (Apache 2.0) are genuinely popular in the open-source community, and April 2026's Wan 2.7 adds a "Thinking Mode" that plans a composition before rendering it. Backed by Alibaba Cloud's infrastructure and a documented DashScope API, it's a sensible pick for businesses already in that ecosystem, and the open weights give technical teams an option closed rivals don't.
The catch is access and polish: the consumer web app leans Mandarin-first, Wan 2.7's newest features are cloud/API-only with no open weights yet, and pay-as-you-go billing adds onboarding overhead compared to flat subscriptions. Clip length still caps around 15 seconds against competitors offering multi-minute output.
What people say
VBench doesn't lie about raw capability: Wan 2.1 topped the benchmark at 86.22%, and the open Apache 2.0 weights for Wan 2.1 and 2.2 have pulled hundreds of thousands of monthly downloads across HuggingFace and ModelScope — a rare case of a Chinese model the global open-source crowd actually adopted rather than just benchmarked. ChatForest rates it 4/5, and reviewers consistently praise the permissive licensing, non-expiring credits, and storyboard-level structural controls most closed competitors don't expose.
Wan 2.7, released in April 2026, is a different animal: cloud/API-only so far, priced around $0.10/second, and built around a "Thinking Mode" that plans composition before generating rather than reacting straight to the prompt. Its open weights haven't shipped yet, so anyone wanting the new 12-language text rendering or Thinking Mode locally is out of luck for now — that capability lives behind Alibaba Cloud's DashScope API only.
The consistent gripes: a steep learning curve on the structural controls, clips capped around 15 seconds against Kling's multi-minute output, physics simulation that still trails Runway Gen-4, and slow inference on consumer GPUs even with the open weights downloaded locally. Billing is pay-as-you-go rather than subscription — roughly ¥0.2 per image and ¥0.6-1 per second of video — which suits high-volume or API-first users better than someone paying per generation.
Summary of public user & expert reviews, compiled by RECATOOLS.
About this listing
This entry was compiled from publicly available data including Tongyi Wanxiang 通义万相's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Tongyi Wanxiang 通义万相 unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Tongyi Wanxiang 通义万相 directly →
Spotted something out of date? Suggest an update →
Alternatives to Tongyi Wanxiang 通义万相
More in Image Generation
