--- title: "Wan 3.0: Specs, Strengths & Prompt Tips | imaginary.app" description: "Wan 3.0 is Alibaba’s video model for lifelike people, with 30 fps output and native audio. Specs, prompt tips, known weak spots and changes from Wan 2.x." url: https://www.imaginary.app/models/wan-3-0 --- Updated November 2026 # Wan 3.0 (Wan3.0) Wan 3.0 is Alibaba’s current-generation video model, released on Alibaba Cloud in August 2026. It generates smooth 30 fps footage with optional sound, and Alibaba pitches it for lifelike people and consistent characters. Alibaba is candid that its audio and on-screen text still lag behind. Key takeaways - Listed on Alibaba Cloud Model Studio on August 6, 2026 and announced as generally available on August 24, 2026. - Alibaba documents 30 fps output in 480P, 720P or 1080P. - Audio is generated by default and can be switched off. - Short prompts are rewritten by a language model before generation by default, which helps quality but adds waiting time. - Unlike Wan 2.1 and 2.2, Wan 3.0 has no open weights as of November 2026; it is API-only. Alibaba Generate with Wan 3.0 Write a prompt, pick Wan 3.0, choose your settings and see the credit cost before you start. Failed generations return their credits. [Open the generator](https://www.imaginary.app/generate)[View Pricing](https://www.imaginary.app/pricing) Sign in with Google to create an account; buy credits only when you need them. ## On imaginary.app These are the settings our generator offers for Wan 3.0 right now, read from our live model catalog. - Type Video, from a text prompt - Resolutions 480p, 720p, 1080p - Aspect ratios 16:9, 9:16, 1:1 - Clip lengths 5 s, 10 s - Audio Optional: switch generated sound on or off - Starting price 2.08 credits (about $0.44–$0.52) at the lowest settings (480p, 16:9, 5 s, sound off) The generator shows the exact credits for your settings before you start. USD ranges depend on which [pack or plan](https://www.imaginary.app/pricing) you bought credits with. If a generation fails, or you cancel it before it starts, its credits go back to your balance. Very rarely, when we can't confirm with the provider whether a generation finished, the credits stay reserved while we check, rather than being charged. Models other than Veo 3.1 need an account that has bought a credit pack or plan, and they have daily usage limits, per account and across the site: if one is reached, the generator tells you and you can try again the next day (limits reset at midnight UTC). ## Lifelike people and fine-grained consistency Alibaba’s launch post says text-to-video aims for “diverse, lifelike faces” with natural micro-expressions tied to body language. In reference work, a character’s “facial features, hairstyle and hair color, body shape, clothing, and accessories stay consistent at fine granularity”. fal summarises the generation as “enhanced motion smoothness, superior scene fidelity, and greater visual coherence”. For a reader choosing a model, this is the one to try for people-centred scenes: a face in close-up, someone walking and talking, small gestures. ## Your prompt is rewritten before generation By default, Wan 3.0 passes your text through a language model that expands it into a fuller shot description before generating. Alibaba calls this prompt\_extend and fal calls it prompt expansion. Alibaba says it improves quality for short prompts, and fal notes that disabling it saves roughly 20 to 60 seconds of latency. In practice, a one-line idea still produces a fully staged scene, and Wan generations can take noticeably longer than lighter models. ## Smooth 30 fps motion Alibaba’s API reference lists a 30 fps output; MiniMax H3, for comparison, is documented at 24 fps. Higher frame rate makes pans, sports and anything with fast hand movement look smoother, and it suits screen-native social clips. If you are cutting Wan footage into a 24 fps film edit, expect to conform it. ## Where it sits in the Wan line Wan 3.0 succeeds Wan 2.7, which Alibaba says ran from 2 to 15 seconds and introduced video editing. A high-speed sibling, Wan3.0-Video-Prime, followed on August 20, 2026 with the same capabilities and faster turnaround. Earlier Wan 2.1 and 2.2 models were released as open weights on Hugging Face; as of November 2026 Wan 3.0 has not been, so it runs only through Alibaba’s cloud and partners such as fal. Alibaba documents up to 30 seconds per Wan 3.0 generation; on imaginary.app the available lengths are shown in the “On imaginary.app” panel. ## Wan 3.0 specs from Alibaba What Alibaba and its API partners document for the model itself. A maker spec is not always a setting our generator offers: what you can choose on imaginary.app is listed in the block above. | Spec | As documented | Source | | ----------------------------------- | -------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------- | | Listed in Model Studio | wan3.0-video listed August 6, 2026 | [Alibaba Cloud Model Studio — newly released models](https://www.alibabacloud.com/help/en/model-studio/newly-released-models) | | General availability | Announced August 24, 2026 | [Alibaba Cloud blog — Wan3.0 at General Availability (Aug 27, 2026)](https://www.alibabacloud.com/blog/603505) | | Status in Alibaba docs | API reference still labels the model “Currently in preview” | [Alibaba Cloud Model Studio — Wan3.0 video generation API reference](https://www.alibabacloud.com/help/en/model-studio/wan3-video-generation-api-reference) | | Length per generation (Alibaba API) | 2 to 30 seconds, or smart duration | [Alibaba Cloud Model Studio — Wan3.0 video generation API reference](https://www.alibabacloud.com/help/en/model-studio/wan3-video-generation-api-reference) | | Resolutions | 480P, 720P, 1080P | [Alibaba Cloud Model Studio — Wan3.0 video generation API reference](https://www.alibabacloud.com/help/en/model-studio/wan3-video-generation-api-reference) | | Frame rate | 30 fps | [Alibaba Cloud Model Studio — Wan3.0 video generation API reference](https://www.alibabacloud.com/help/en/model-studio/wan3-video-generation-api-reference) | | Audio | On by default; can be turned off | [fal — Wan 3.0 text-to-video (llms.txt schema)](https://fal.ai/models/alibaba/wan-3.0/text-to-video/llms.txt) | | Prompt languages | Chinese and English; prompts up to 20,000 characters | [Alibaba Cloud Model Studio — Wan3.0 video generation API reference](https://www.alibabacloud.com/help/en/model-studio/wan3-video-generation-api-reference) | | Prompt expansion | On by default; rewrites prompts with a language model | [Alibaba Cloud Model Studio — Wan3.0 video generation API reference](https://www.alibabacloud.com/help/en/model-studio/wan3-video-generation-api-reference) | | Faster variant | Wan3.0-Video-Prime (August 20, 2026), same capabilities, faster | [Alibaba Cloud Model Studio — newly released models](https://www.alibabacloud.com/help/en/model-studio/newly-released-models) | | Open weights | Not released for Wan 3.0 as of November 2026; Wan 2.1 and 2.2 are open on Hugging Face | [Hugging Face — Wan-AI organisation (open-weight Wan models)](https://huggingface.co/Wan-AI) | ## Who Wan 3.0 is for ### Good fit - People-centred shots: faces, expressions, gestures and characters who need to look the same throughout. - Quick ideas where you only have a sentence and want the model to fill in staging and camera. - Smooth-looking 30 fps social clips with motion, like sport, dance or a moving camera. - Drafting a concept at a lower resolution before re-running at a higher one. ### Look elsewhere if - Clips that depend on readable on-screen text, which Alibaba says is not where it wants it yet. - Projects where the generated soundtrack has to be polished; plan to replace or mix it. - Anyone who needs exact, literal adherence to every word; prompt expansion will reinterpret short prompts. - Self-hosting: Wan 3.0 has no downloadable weights as of November 2026. ## Prompt tips for Wan 3.0 Because prompts are expanded automatically, a short prompt is fine. Lead with the one or two things that must not change (who, what, which camera move) so the rewrite keeps them. Example prompt (written by us) ``` Close-up of a grandmother laughing as she kneads dough in a sunlit kitchen, flour on her cheeks, handheld camera. ``` For people, describe expression and body language together. Alibaba ties Wan 3.0’s realism to micro-expressions that match the body, so give it both. Example prompt (written by us) ``` A young violinist on a dim stage closes her eyes and sways as she plays the final phrase, shoulders relaxing, a single spotlight above her; slow push-in from the front row. ``` Use the frame rate. Movement-heavy subjects show off 30 fps more than static scenes, so give the camera or subject something to do. Example prompt (written by us) ``` Tracking shot beside a golden retriever running along a wet beach at sunrise, spray kicking up from its paws, waves rolling in, wind and surf on the soundtrack. ``` Keep any required text off the screen, or add it in editing afterwards. Do the same for any sound that has to be exact. ## Limitations - Alibaba’s own words: “audio texture and on-screen text rendering accuracy are improving but not yet where we want them.” - Prompt expansion adds waiting time; fal estimates roughly 20 to 60 seconds of extra latency when it is on. - Alibaba’s API reference still marks the model as in preview, despite the general-availability announcement. - The 30-second ceiling is a maker capability; on imaginary.app the available lengths and resolutions are shown in the “On imaginary.app” panel. - Wan 3.0 is made by Alibaba; imaginary.app is not affiliated with Alibaba. ## Related models - [Seedance 2.5](https://www.imaginary.app/models/seedance-2-5) - [MiniMax H3 Max](https://www.imaginary.app/models/minimax-h3) - [Kling 3 Pro](https://www.imaginary.app/models/kling-3) - [Veo 3.1 Lite](https://www.imaginary.app/models/veo-3-1-lite) - [Every model on imaginary.app](https://www.imaginary.app/models) - [AI video generator overview](https://www.imaginary.app/ai-video-generator) ## Frequently asked questions ### Is Wan 3.0 free? Not on imaginary.app; each generation uses credits, and the starting cost is shown in the “On imaginary.app” panel on this page. Unlike Wan 2.1 and 2.2, Wan 3.0 has no free open weights to run yourself, as of November 2026. ### Is Wan 3.0 open source? No, not as of November 2026. Alibaba has not published Wan 3.0 weights. Its Hugging Face organisation hosts open Wan 2.1 and 2.2 models, but Wan 3.0 is available only through Alibaba Cloud and partner APIs. ### Does Wan 3.0 have audio? Yes. Audio is generated by default and can be turned off. Alibaba says its audio texture is still improving, so for finished work you may want to replace or remix the soundtrack. ### What resolution and frame rate does Wan 3.0 output? Alibaba documents 480P, 720P and 1080P at 30 fps. The resolutions offered on this site are shown in the “On imaginary.app” panel. ### Wan 3.0 vs Wan 2.x: what changed? Alibaba cites more lifelike faces and stronger character consistency, and on its API Wan 3.0 raises the single-generation ceiling to 30 seconds, from Wan 2.7’s 15 (the lengths offered here are in the “On imaginary.app” panel). Wan 3.0 is API-only, whereas Wan 2.1 and 2.2 were released as open weights. ## Sources Prices, model specs and dates on this page were checked on November 10, 2026. Third-party prices change; check the source before you rely on a number. - [Alibaba Cloud Model Studio — newly released models](https://www.alibabacloud.com/help/en/model-studio/newly-released-models) - [Alibaba Cloud Model Studio — Wan3.0 video generation API reference](https://www.alibabacloud.com/help/en/model-studio/wan3-video-generation-api-reference) - [Alibaba Cloud Model Studio — Wan2.7 text-to-video API reference](https://www.alibabacloud.com/help/en/model-studio/text-to-video-api-reference) - [Alibaba Cloud blog — Wan3.0: 30-second AI video generation from any input (Aug 13, 2026)](https://www.alibabacloud.com/blog/wan3-0-30-second-ai-video-generation-from-any-input_603452) - [Alibaba Cloud blog — Wan3.0 at General Availability (Aug 27, 2026)](https://www.alibabacloud.com/blog/603505) - [fal — Wan 3.0 text-to-video (model page)](https://fal.ai/models/alibaba/wan-3.0/text-to-video) - [fal — Wan 3.0 text-to-video (llms.txt schema)](https://fal.ai/models/alibaba/wan-3.0/text-to-video/llms.txt) - [Hugging Face — Wan-AI organisation (open-weight Wan models)](https://huggingface.co/Wan-AI) - [MiniMax — MiniMax H3 open-source release (24 fps reference)](https://www.minimax.io/news/minimax-h3-open-source) - [imaginary.app pricing](https://www.imaginary.app/pricing) How we checked: model specs come from Alibaba's and its API partners' pages listed above; the imaginary.app settings and starting price come from our live model catalog. Example prompts are ours, not Alibaba's. Commercial use: licence and usage terms for Wan 3.0 are set by Alibaba. Check them, and our [Terms of Service](https://www.imaginary.app/terms-of-service), before you publish what you make. Alibaba ## Try Wan 3.0 on your own prompt One account and one credit balance for Wan 3.0, Veo 3.1 and every other video model we offer. The price shows before you generate. [Open the generator](https://www.imaginary.app/generate)[View Pricing](https://www.imaginary.app/pricing) Sign in with Google to create an account; buy credits only when you need them.