이미지: The Decoder
Summary
- Alibaba's video generation model Wan3.0 can take not just text and images but PDFs, PowerPoint files, and even webpages as input, producing videos up to 30 seconds long.
- A 1080p, 30-second clip costs $6 on the Standard plan, while the faster Prime plan runs $8.40.
- The announcement comes as Alibaba carries out the largest share offering in Hong Kong stock market history and reports a 75% drop in quarterly net profit.
- 모델명
- Wan3.0 (퍼블릭 베타)
- 최대 영상 길이
- 30초 (전작 Wan2.5 대비 2배)
- 입력 형식
- 텍스트, 이미지 최대 10장, 영상 최대 5개, 오디오 최대 5개, PDF·PPT·웹페이지
- 이용 경로
- wan.video, 알리바바 클라우드 모델 스튜디오, Qwen Cloud API
- 요금(1080p 30초, 스탠다드)
- 6달러 (현재 30% 할인 적용가)
- 요금(1080p 30초, 프라임)
- 8.40달러
- 알리바바 분기 순이익
- 전년 대비 75% 감소, AI 투자 확대가 원인
Wan3.0 shows up this time with a price tag
As METAL LAB reported on August 8, Alibaba's video generation model Wan3.0 was already available in public beta. The headline features were its ability to generate videos up to 30 seconds in a single pass, and its willingness to accept not just text, images, voice, and video as input, but documents, spreadsheets, slide decks, and webpages too. What The Decoder has now reported is how that beta actually works in practice and how it's priced. Users can reach it through three paths — the wan.video website, Alibaba Cloud Model Studio, and the Qwen Cloud API — with pricing split into two tiers: Standard and Prime.
What's newly confirmed
The 30-second video length is double what the previous model, Wan2.5, could produce. Wan3.0 can look at a user's prompt and recommend an appropriate video length on its own, and it also comes with an extension tool that lets users lengthen a video they've already made. This report also confirmed the specifics of how much material can go into a single prompt: up to 10 images, 5 videos, and 5 audio clips at once. Feed it a document or a slide deck, and it turns that static material into a moving video.
One recurring problem in AI video generation is "drift," where faces or on-screen elements shift subtly from scene to scene. Alibaba says it has reduced this by preserving details from reference material — characters, props, spatial layout — more consistently across a video.
How the pricing breaks down
| Resolution | Standard (per second) | Prime (per second) | 30-sec clip (Standard) | 30-sec clip (Prime) |
|---|---|---|---|---|
| 480p | $0.05 | $0.068 | $1.50 | $2.04 |
| 720p | $0.10 | $0.14 | $3.00 | $4.20 |
| 1080p | $0.20 | $0.28 | $6.00 | $8.40 |
The Standard plan's listed price already reflects a 30% discount currently in effect. Prime processes faster but charges a higher per-second rate.
What it's meant for
Alibaba has pitched a wide range of uses for Wan3.0, from speeding up film production to making short-form dramas and social media clips. It's also floated a scenario where businesses turn nothing but text and images into marketing or training videos. One use case that stands out is generating simulation footage to train autonomous vehicles and robotics systems — the idea being that a large volume of realistic virtual driving or task-performance scenes could be used to train robotic or self-driving models.
How to actually use it
Wan3.0 is accessible through the wan.video website, Alibaba Cloud Model Studio, or via API calls through Qwen Cloud.
- On the wan.video generation screen, enter a text prompt, or attach a PDF, PowerPoint file, or webpage link.
- To include image, video, or audio reference material, you can add up to 10 images, 5 videos, and 5 audio clips.
- The model will recommend an appropriate video length based on your prompt — you can go with that or adjust the length yourself.
- Choose a resolution (480p, 720p, or 1080p) and a plan (Standard or Prime); cost varies accordingly.
- If a finished video feels too short, the extension tool lets you lengthen it by adding on to what you already have.
Enterprise developers can instead go through the Qwen Cloud API to build video generation directly into their own services.
The announcement lands alongside a bigger AI push
The rollout of Wan3.0's pricing structure coincides with Alibaba significantly scaling up its AI investment. The company recently announced the largest share offering by any Hong Kong-listed firm in history to fund its AI business, and last week it reported that its quarterly net profit had fallen 75% year-over-year, largely due to that expanded AI spending. Rolling out a video generation model as a fully priced product, complete with a detailed rate card, looks like part of the strategy to start recouping that investment.
Alibaba has also been busy elsewhere in the Wan lineup: it recently open-sourced its character animation model Wan-Animate-2, and added an Agent Skills feature that lets the community build and share workflows. The whole Wan family is getting updated in rapid succession.
Editor's take
What this pricing sheet really tells us is that Alibaba has stopped treating Wan3.0 as an experimental demo and started treating it as a product people actually pay for. Going from an early-August beta to a fully worked-out per-second rate card in just three weeks reads like a deliberate move to compete head-on, on price, with rivals like OpenAI's Sora and Google's Veo. The ability to ingest entire documents and slide decks is something you rarely see in other video generation models, and it signals that Alibaba is aiming less at consumer content creation and more at the market for turning corporate work documents into video.
Anyone who's tried document-to-video features in practice runs into the same sticking point every time: how accurately the text information survives once it's baked into a video. Feed in a few slides, and the more tables and numbers they contain, the more likely something gets distorted along the way. For marketing teams or e-learning content teams in Korea, it's probably safer at this stage to start testing with lower-text-density material — short internal training videos or product intro clips — rather than anything data-heavy. At $6 to $8 for a single 30-second, 1080p clip, the cost is manageable for small-scale experiments, but anyone planning to scale up production should factor in how quickly the per-second billing can add up.
Given that Alibaba is willing to absorb a 75% drop in quarterly profit to keep pushing its AI investment, it's likely we'll see more news in the coming weeks about robotics simulation use cases for the Wan lineup, or new partner integrations. There are already signs that third-party platforms have started building Wan3.0 into their own services.
Comments