Guides

Wan 2.7 vs Wan 3.0: Release Dates, Specs and Where to Use Each

Wan 2.7 vs Wan 3.0 from Alibaba's own docs: release dates, modes, clip length, resolution, audio and open weights, with Wan 2.6 and 2.2 for context.

By Anna Krejcova — Editor at FansCreatorPublished 11 min read
Article summary
  • Wan 3.0 makes clips of 2 to 30 seconds at 480P, 720P or 1080P with sound. Wan 2.7 stops at 15 seconds and offers 720P or 1080P.
  • Wan 2.7 shipped as four models (text-to-video, image-to-video, reference-to-video, editing). Wan 3.0 folds generation into one model and takes up to 10 images, 5 videos, 5 audio clips and a document or web page as references.
  • Neither Wan 2.7 nor Wan 3.0 has downloadable weights. Both run as paid services on Alibaba Cloud Model Studio. The newest open Wan video weights are Wan 2.2, under Apache 2.0.
  • The platform decides where a clip can go: on OnlyFans only labelled AI content of you, the verified creator; AI personas on Fanvue or MYM with AI labels; no photorealistic AI on Fansly.

Wan 3.0 is the newer and bigger model. Alibaba’s documentation lists clips of 2 to 30 seconds at up to 1080P with sound, where Wan 2.7 stops at 15 seconds. Both are hosted, paid models on Alibaba Cloud Model Studio, and neither has downloadable weights. If you want to run Wan on your own GPU, the newest open video weights are still Wan 2.2.

Wan is Alibaba’s family of video generation models. Everything below comes from Alibaba’s own pages, read on 5 October 2026. Alibaba writes the names without a space (Wan2.7, Wan3.0); they are the same models.

Wan 2.7 vs Wan 3.0 at a glance

Wan 2.2 Wan 2.6 Wan 2.7 Wan 3.0
Announced July 2025 16 December 2025 7 April 2026 Beta 7 August 2026; open API access 24 August 2026 (API reference still says preview)
Modes Text-to-video, image-to-video, combined text-image-to-video Text-to-video, image-to-video, reference-to-video Text-to-video, image-to-video, reference-to-video, video editing One model: text-to-video, image-to-video (first frame or first and last frame), reference-to-video, plus editing and extension of a video you pass in
Clip length Hosted: 5 seconds. Open weights: the press release cites a 5-second 720P clip from TI2V-5B 2 to 15 seconds (reference: 2 to 10) 2 to 15 seconds (reference and editing: 2 to 10) 2 to 30 seconds
Resolution Hosted: 480P to 1080P. Open A14B: 480P, 720P 720P, 1080P 720P, 1080P 480P, 720P, 1080P
Audio No Yes Yes Yes, on by default, can be switched off
Open weights Yes, Apache 2.0 No No No
Where to use it Download, or Model Studio Model Studio, Wan website Model Studio, Wan website, Qwen App Model Studio, Qwen Cloud

The clip lengths, resolutions and audio support come from Alibaba’s video generation model overview, which lists the hosted Wan 2.5, 2.6, 2.7 and 3.0 models at 30 fps in MP4. The open Wan 2.2 TI2V-5B model runs at 24 fps, per its repository.

Release dates: how the Wan line got here

The Wan family moved from open releases to hosted ones in about a year.

  • Wan 2.2, July 2025. The Wan 2.2 repository released the inference code and weights on 28 July 2025. Alibaba’s press release followed on 29 July. The weights went to Hugging Face, GitHub and ModelScope.
  • Wan 2.6, 16 December 2025. Alibaba’s Wan 2.6 announcement added a reference-to-video model and clips “of up to 15 seconds”.
  • Wan 2.7, 7 April 2026. The Wan 2.7 video announcement came a few days after the Wan 2.7 image model.
  • Wan 3.0, August 2026. Alibaba opened a public beta on 7 August. Its product news dated 24 August says the API “can be called without the need to apply”.

Model Studio also still lists Wan 2.5 as preview models with 5 or 10 second clips and synced audio.

What changed from Wan 2.7 to Wan 3.0

Five things changed in practice: clip length, the model structure, the inputs, the resolution tiers and audio control. Editing carried over.

Clips twice as long

Wan 3.0 makes clips of 2 to 30 seconds in one generation. Alibaba’s beta announcement calls 15 seconds “the maximum clip duration of Wan2.7-Video”. The Wan 3.0 API reference adds a smart duration mode, where the model picks a length from your prompt. With video input, input and output together must not pass 30 seconds.

One model instead of four

Wan 2.7 shipped as four models: text-to-video, image-to-video, reference-to-video and video editing. Wan 3.0 is described on its model page as an “All-in-One video generation model” that handles text-to-video, image-to-video and reference-based generation in one model. There are two versions: wan3.0-video is the standard model, and wan3.0-video-prime is the speed-optimized one, which Model Studio says “significantly improves generation speed while maintaining high quality”. Both have the same modes, lengths and resolutions.

More references, plus documents

Wan 3.0 takes up to 10 reference images, 5 reference videos and 5 audio clips in one request, according to Alibaba’s launch news. The API reference caps reference video and reference audio at 15 seconds in total each.

It also reads documents and web pages. Alibaba’s Wan 3.0 overview lists doc, xls, ppt, pdf, txt, key, pages, numbers and md, one file or link per request, up to 100MB and 50 pages. That matters more for training videos and ads than for creator content.

A 480P draft tier and per-second pricing

Wan 2.7 outputs 720P or 1080P. Wan 3.0 adds 480P. The Wan 3.0 model page lists $0.05 per second at 480P, $0.10 at 720P and $0.20 at 1080P (Singapore region, read 5 October 2026). At those rates a 480P draft costs a quarter of a 1080P render.

Audio you can switch off

Both versions generate sound with the picture. In Wan 3.0, audio is on by default and an audio setting turns it off; the API reference says this “does not affect pricing”. The Wan 2.7 text-to-video reference lets you pass your own WAV or MP3 file. Without one, the model adds background music or sound effects that match the video.

What carried over, and what is still rough

Editing came with Wan 2.7. Alibaba’s Wan 3.0 overview says these capabilities, “first introduced in Wan2.7, carry forward into Wan3.0”: you can “modify visuals, plot, and dialogue on generated videos”, including story re-creation, without regenerating from scratch. The Wan 3.0 API reference also lists video editing and video extension modes, where you pass in a video with a prompt. Alibaba also names two weak spots of Wan 3.0 itself: “audio texture and on-screen text rendering accuracy” are still improving. One status note: Alibaba’s product news says the API opened to everyone on 24 August 2026, but the API reference, last updated 28 September 2026, still labels the model “Currently in preview”.

What Wan 2.7 still does well

Wan 2.7 is not retired. Its image-to-video model lists first frame, first and last frame, video continuation and audio-driven generation, and its text-to-video model lists multi-shot narrative. Alibaba’s announcement says it keeps up to five characters consistent, each with its own voice and look.

Wan 2.7 also keeps a dedicated editing model. Wan 3.0 can edit a video you pass in, but for effect or camera movement replication on existing footage, Model Studio’s guide points to wan2.7-videoedit, which works on clips of 2 to 10 seconds. For clips of 15 seconds or less that don’t need Wan 3.0’s extra inputs, either model works.

Wan 2.6: the 15-second ceiling

Wan 2.6 set the limits Wan 2.7 kept: 2 to 15 seconds, 720P and 1080P, and audio. Its December 2025 announcement introduced Wan2.6-R2V, which lets you upload a character reference video “with both appearance and voice” and generate new scenes with that character. Model Studio still lists the Wan 2.6 models, including faster “flash” versions of image-to-video and reference-to-video.

Open weights: Wan 2.2 is the newest downloadable version

On 5 October 2026, Alibaba’s Wan-AI organisation on Hugging Face, its ModelScope page and the Wan-Video GitHub organisation had weights for Wan 2.1 and Wan 2.2, and none for Wan 2.5, 2.6, 2.7 or 3.0. The newest general text-to-video and image-to-video weights there are Wan 2.2. Newer open releases there include Wan2.2-Animate-2-14B, a character animation model whose card lists its weights as released on 7 August 2026, under Apache 2.0.

None of the Alibaba pages read for this article say whether Wan 2.7 or Wan 3.0 weights will ever be released. So there is no open licence for either. You use them as a paid cloud service.

Wan 2.2 is different. The Wan2.2-I2V-A14B model card puts the weights under Apache 2.0. The team claims “no rights over” what you generate, and you are “fully accountable” for your use. The open A14B models make 480P and 720P video without audio. A separate open model, Wan2.2-S2V-14B, makes audio-driven video from an image and an audio track.

The hosted models also run content checks. The Wan 2.7 text-to-video reference says: “Input and output undergo content moderation.”

Where you can use each version

  • Wan 3.0: the Model Studio API, and Qwen Cloud, which the beta announcement names. The Alibaba pages read for this article do not say whether Wan 3.0 is on the Wan website or in the Qwen App.
  • Wan 2.7: Model Studio, the official Wan website and the Qwen App, per the April announcement.
  • Wan 2.6: Model Studio and the Wan website, with Qwen App integration announced.
  • Wan 2.2: download the weights and run them yourself, or call the hosted wan2.2 models on Model Studio.

FansCreator lists Wan 3.0 and Wan 2.7 among the models it is built on, and FansCreator’s Wan page covers the open-weight Wan 2.2. Its NSFW AI video generator starts from a picture of your creator, so the clip keeps the face and body of your photo sets. It is built for adult creators, with consent rules in its Terms of Use.

Which Wan version fits a creator workflow

Pick by what the clip needs.

  • Longer than 15 seconds, or many references: Wan 3.0. It doubles Wan 2.7’s 15-second limit and is the only Wan version that reads documents.
  • Replicating an effect or camera move on existing footage: wan2.7-videoedit, the model Alibaba’s guide recommends for that job.
  • Short clips from one approved picture: either one. Both have first-frame image-to-video at 720P and 1080P. The trade-offs between starting from a prompt and starting from a picture are in text-to-video vs image-to-video.
  • Your own hardware and full control: Wan 2.2. You accept 480P or 720P, no audio and a large GPU. For swapping a character into existing video, see Wan Animate character swap.

For how Wan compares with Seedance, Veo, Kling and others, read the best AI image to video generators.

Posting Wan clips on OnlyFans, Fanvue, MYM and Fansly

The model does not decide where a clip can go. The face in it and the platform’s AI rules do. Only use your own face or voice, your own AI persona, or a person whose consent you can prove. Policies were read on 5 October 2026.

OnlyFans. The Terms of Use allow AI content only if “it clearly features the verified Creator” and carries a signifier such as #ai or #AIGenerated. The Help Centre says all content must belong to a verified creator. Its content guide lists content that is “wholly AI-generated” or “features heavily altered images” among what cannot be shared. Use Wan clips of yourself there, labelled. The OnlyFans AI content policy page has the full clauses.

Fanvue. Its AI content article allows “fully AI-generated content” with a disclosure that is “clear and prominent” in the bio, caption or as a watermark. Face-swapped content is allowed too, but “you MUST verify the real model behind the AI”. A persona built from scratch fits here; how to create an AI influencer covers building one.

MYM. The MYM AI article requires “Virtual Image”, #AI or #AIgenerated, and MYM marks the account with an “AI profile” badge. Deepfakes are allowed only with “written permission from the person being deepfaked”.

Fansly. The Fansly AI article says “Photorealistic AI-generated content is not allowed on Fansly”, and a label does not change that.

Wan 2.7 or Wan 3.0: the short answer

Wan 3.0 is the upgrade if you need 30-second clips, a 480P draft tier or many references in one request. Wan 2.7 is still a solid choice for clips up to 15 seconds and for effect or camera replication on existing footage. If you need weights on your own machine, Wan 2.2 is the route.

Specs move fast on hosted models, so check Alibaba’s model overview before you plan a batch around a number in this article.

Frequently asked questions

Is Wan 3.0 open source?

No. On 5 October 2026 Alibaba's Wan-AI pages on Hugging Face and ModelScope and the Wan-Video GitHub organisation had weights for Wan 2.1 and Wan 2.2, and none for Wan 2.5, 2.6, 2.7 or 3.0. Wan 3.0 runs as a paid API on Alibaba Cloud Model Studio. The newest open Wan video weights are Wan 2.2, under Apache 2.0.

How long can a Wan 2.7 video be?

Wan 2.7 text-to-video and image-to-video make clips of 2 to 15 seconds. Its reference-to-video and editing models go from 2 to 10 seconds. Output is 720P or 1080P at 30 fps, according to Alibaba Cloud Model Studio.

What is the difference between Wan 2.6 and Wan 2.7?

Both make clips of 2 to 15 seconds at 720P or 1080P with audio. Wan 2.7 added a video editing model, and Alibaba says editing was first introduced in Wan 2.7. Its image-to-video model lists first and last frame control, video continuation and audio-driven generation.

Does Wan 3.0 generate audio?

Yes. Audio is on by default and can be switched off, which Alibaba says does not change the price. Wan 3.0 also accepts up to 5 reference audio clips, 15 seconds in total. Alibaba says audio texture is still being improved.

Where can I use Wan 3.0?

Through the Alibaba Cloud Model Studio API and on Qwen Cloud. Alibaba's product news says the API can be called without applying since 24 August 2026, though the API reference still labels the model as in preview. Alibaba's pages read for this article do not confirm it on the Wan website or Qwen App. FansCreator also lists Wan 3.0 among its models.

Sources

Read as Markdown

Bring your next idea to life.

Create images, videos and content for your audience with FansCreator.

Start for Free