AI Product Video

You have the packshot. What you need is the product moving, lit, in a room your photographer never booked. Drop the photo in, name the parts that may not change, and Wan 3.0 generates the shot around it — sound included. Your first clip is free, an email and no card.

Start here

Create image

JPG · PNG · BMP · WEBP · ≤20MB · 240–8000px · ratio ≤8:1

0 / 20000

The subject keeps the same face, hair, clothing and colours throughout; only the camera and the described motion change.

Model
Aspect ratio
Duration
Resolution
免费注册 · 送一条 Wan 3.0 片子

免费那条就是同一个 Wan 3.0,声音也有,上限是 480P · 3s,走的是共用队列。方案能把上限抬高到 1080P 和三十秒 —— 免费档是一个分辨率,不是另一个模型。

1 free clip on the real model · an email, no card · commercial use is included on every paid plan and every credit pack

Two of Wan 3.0’s own published product runs, each printed with the whole prompt beside the clip it produced · Try this loads the prompt

A premium mixed-nut pack on a kitchen table, turned to show its ingredient panelTHREE SHOTS · PACK COPY LEGIBLE · 2:3

A packshot walked round, with the small print still readable

The difference that decides everything

An AI product video generates footage. It does not edit your photos.

Most tools sold under this phrase cut your existing images against a template. This one shoots something that never existed, around the product you supplied. Knowing which you are buying saves an afternoon.

The back of a snack pack held to camera, ingredient list and nutrition table in frame

Your packshot

Generated around it — the room, the light, the hands, the sound

Your photograph becomes the literal first frame, so the product is on screen in frame one at the shape and colour you shot it. From there Wan 3.0 generates 2 to 30 seconds at 30 fps — the figures are published in Alibaba's API reference — and it generates the whole frame: the room, the light, a hand entering, the surface it stands on, and the audio, all in the same pass. Nothing is composited, nothing is a Ken Burns pan across a JPEG, and nothing else in the frame had to exist before you pressed generate.

Your product, unretouched
Frame 001

Your product, unretouched

Any whole second, at 30 fps
2–30 s

Any whole second, at 30 fps

Ratio read from your packshot
Adaptive

Ratio read from your packshot

Commercial use, every paid plan
Included

Commercial use, every paid plan

  • It is not a slideshow — your other photographs are not lifted into the cut
  • It is not a 3D turntable — there is no model of your product, so the far side is invented
  • It is not a catalogue pipeline — one product, one prompt, one clip. A thousand SKUs is not this
  • It does not remember your product between runs; the packshot goes with every request

Three products, one phrase

Three things sold as product video, and which one this is

They cost different amounts, take different amounts of time, and break in completely different places. Most people discover which one they bought after the first delivery.

  • Tabletop shoot

    What you supply
    The physical product, a studio and a day
    Turnaround
    Days to weeks
    What it is good at
    Legally exact. Every reflection is real
    Where it breaks
    Cost, scheduling, and one more shot means one more booking
    Choose this when
    The label has to be readable and correct in a regulated market

    Nothing here replaces a real shoot for a regulated claim. That is the honest answer.

  • Asset-lift template

    What you supply
    Your existing photographs and a script
    Turnaround
    Minutes
    What it is good at
    Cutting what you already have, on beat
    Where it breaks
    It can only show what you already photographed
    Choose this when
    You have a full asset library and need it assembled

    If your photo library is already complete, a template editor is faster than this and cheaper.

  • Generated footage

    What you supply
    One packshot and a sentence
    Turnaround
    About ninety seconds
    What it is good at
    Scenes, moods and moments you never shot and could not afford to
    Where it breaks
    Approximate rather than exact — it holds a design, it does not guarantee a pixel
    Choose this when
    You need a shot that does not exist and cannot book a studio to get it

    This page. Upload the packshot at the top; the first clip is free.

The line between the first and the third is not quality, it is provenance. A shoot produces a photograph of your product. Generation produces a convincing depiction of it. For a hero loop, a social cut or a lifestyle scene, convincing is what the brief asked for — and for a claim a regulator will read, it is not.

Before you spend

Five ways a product video comes back unusable

Every one of these turns up on somebody's first packshot. Four are fixed by a sentence; the fifth is a decision about what you are willing to publish.

  • The label turns into a smear of letters

    Small high-contrast type is the first thing to go, and it goes fastest while the camera moves

    Lock the label explicitly, keep the camera slow, and shoot at 720P or above. For type that must be legally correct, composite the real label in afterwards

    Read the frame before publishing
  • The shape changes as it turns

    There is no 3D model. The far side of your product was never photographed, so it is invented as it comes into view

    Keep the rotation under a quarter turn, or supply a second reference from the other side. A full 360 is what a real turntable is for

    One sentence
  • Glass and metal look like plastic

    Material words are weak. "Premium" and "luxury" mean nothing to the model; a finish does

    Name the finish and the light together: "ribbed clear glass, single soft key from the left, no specular blowout"

    One sentence
  • The hand holding it looks wrong

    Hands are the classic failure and they get worse the closer the crop

    Frame wider, or write the product alone on the surface. If a hand is essential, say what it does — "one hand lifts the lid and leaves frame" — rather than leaving it to improvise

  • Shot three does not match shot one

    Every generation is independent. Nothing about your product carries from one request to the next

    Send the same reference set every time — up to ten images. Reference to video is the mode for a full set of PDP shots

A run that fails outright is refunded automatically. The five above do not fail — they come back finished and wrong, and those are billed. That is the argument for drafting at 480P: an iteration that teaches you something costs a quarter of a keeper.

Pricing · 1080P on every paid plan · Native audio

What a product video costs against the shoot it replaces

The comparison that matters is not against another AI tool, it is against a half-day of studio time. Budget three or four generations for a hero shot that has to look right — first a draft, then the thing that slipped added to the clause, then the keeper. A failed run is refunded automatically, audio costs nothing extra, and commercial use is included on every paid plan and every credit pack.

How this is billed

Billed by
output second
Commercial use
included
Failed generation
refunded
Your first clip
free · 480P

Find the move at 480P, then run the keeper once at 720P or 1080P — labels and fine detail need the higher tier, motion tests do not. Full pricing.

两种付法,每月积分一样

  • 免费的第一条不用卡

    在真正的 wan3.0-video 上生成一次。拿你自己的镜头看看 Wan 3.0 会做成什么样 — 480P,三秒,声音同一次出。

    $0一次性

    全程不用卡

    生成免费的那条

    一个邮箱,不用卡

    1 条,仅此一次

    480P · 最长 3 秒 · 带声音
    一个邮箱,不用卡

    和付费套餐跑的是同一个 `wan3.0-video` — 480P 是一个分辨率,不是一个缩水的模型

    • 真正的 wan3.0-video,不是 2.7
    • 声音和画面一起生成
    • 收据上有 model ID 和 task ID
    • 跑失败永远不扣你的
    • 720P 与 1080P付费
    • 超过 3 秒的片子付费
    • 参考、文档与编辑模式付费
    • 第二条以及以后付费

    只回答一个问题:Wan 3.0 会把你的想法做成什么样。1080P、三十秒和全部输入模式,从 $12.90 起。

    一眼看完

    每月条数
    1 条,仅此一次
    每次出几条
    1
    最长时长
    3s
    最高分辨率
    480P
  • Starter最多省 20%

    一周一条发布贴。够你摸清 Wan 3.0 AI 视频生成器合不合你的工作方式,而不用先对用量下判断。

    $12.90/ 月$15.90

    按年收 $154.80

    随时取消

    640 积分 / 月 · 约 16 条

    按 5 秒 480P 算 · 720P 8 条 · 1080P 4 条

    套餐积分在当月月底过期

    • 每月 640 积分 — 大约 16 条成片
    • 同一条提示词每次出 2 条
    • 全部分辨率 — 480P、720P 与 1080P
    • 全部时长 — 一次 2 到 30 秒
    • 全部输入 — 文字、图片、参考、文档、网页
    • 声音写在同一次生成里,无水印
    • 含商用授权 — 发布、售卖、开票都可以
    • Wan 3.0 Prime 快速档,随时可用
    • 跑失败永远不扣积分
    • 积分未动用时,7 天内可退
    • 两步取消 — 你加入时的价格锁死

    前两行以下的一切,$12.90 这一档就都有。更大的套餐买的是秒数和条数 — 从来不是更好的模型。

    一眼看完

    每月条数
    5 秒 480P 下约 12 条
    每次出几条
    2
    每美元积分
    基准
    退款窗口
    7 天
  • 多数人落在这里
    Pro

    每个工作日一条成片,每条还有三个备选。480P 打草稿、1080P 定稿的那个循环,按一个工作周的量配好。

    $39.90/ 月$49.90

    按年收 $478.80

    随时取消

    2,240 积分 / 月 · 约 56 条

    按 5 秒 480P 算 · 720P 28 条 · 1080P 14 条

    套餐积分在当月月底过期

    • 每月 2,240 积分 — Starter 的 3.5 倍
    • 每月约 56 条成片,或者 14 条满 1080P
    • 同一条提示词每次出 4 条 — 挑一条,而不是重摇一次2× 条数
    • 每美元积分比 Starter 多 13%更划算
    • 同一周内既够 480P 打草稿,也够 1080P 定稿
    • 全部分辨率 — 480P、720P 与 1080P
    • 全部时长 — 一次 2 到 30 秒
    • 全部输入 — 文字、图片、参考、文档、网页
    • 声音写在同一次生成里,无水印
    • 含商用授权 — 发布、售卖、开票都可以
    • Wan 3.0 Prime 快速档,随时可用
    • 跑失败永远不扣积分
    • 积分未动用时,7 天内可退
    • 两步取消 — 你加入时的价格锁死

    从 Starter 往上,涨的是量、条数和单价。模型、分辨率和那三十秒,本来就已经是你的了。

    一眼看完

    每月条数
    5 秒 480P 下约 44 条
    每次出几条
    4
    每美元积分
    比 Starter 多 21%
    退款窗口
    7 天
  • Studio最划算

    接客户量的档,也是你不再省着用 1080P 的那一档。每月三十一条满分辨率成片,单位积分是我们卖过最低的价。

    $99.90/ 月$119.90

    按年收 $1,198.80

    随时取消

    6,240 积分 / 月 · 约 156 条

    按 5 秒 480P 算 · 720P 78 条 · 1080P 39 条

    套餐积分在当月月底过期

    • 每月 6,240 积分 — Starter 的 9.75 倍,Pro 的 2.8 倍13×
    • 每月约 156 条成片,或者 39 条满 1080P
    • 每美元积分多 26% — 我们卖过最划算的最划算
    • 1080P 多到你不必先用 480P 打草稿
    • 同一条提示词每次出 4 条
    • 按一整份客户名单配的量,不是一个频道
    • 任何一个月都能加购,不用换套餐
    • 全部分辨率 — 480P、720P 与 1080P
    • 全部时长 — 一次 2 到 30 秒
    • 全部输入 — 文字、图片、参考、文档、网页
    • 声音写在同一次生成里,无水印
    • 含商用授权 — 发布、售卖、开票都可以
    • Wan 3.0 Prime 快速档,随时可用
    • 跑失败永远不扣积分
    • 积分未动用时,7 天内可退
    • 两步取消 — 你加入时的价格锁死

    这是为一整份客户名单配的档,不是为一个频道:每条提示词出四条、1080P 不用省着用,还留得出重拍一场戏的余量,不用盯着余额往下掉。

    一眼看完

    每月条数
    5 秒 480P 下约 124 条
    满 1080P 的话
    每月约 31 条
    每美元积分
    比 Starter 多 28%
    退款窗口
    7 天

The whole procedure

How to make an AI product video from one photo

Four steps. The second is the one that separates a usable clip from a pretty clip of something that is not quite your product.

  1. Upload the cleanest packshot you have

    Plain background, even light, product filling the frame. The checker confirms the file in a second, and a clean input leaves the model less to reinterpret.

    Checked free
  2. Name the parts that may not change

    "Gold foil label, ribbed clear glass, matte black cap." Nouns and finishes, not adjectives — the appearance box at the top builds the clause for you.

    The step that decides it
  3. Describe the move and the light, not the product

    The model can see the product. Say what the camera does, where the key light is, what enters frame, and what it sounds like.

    Sound in the same field
  4. Draft at 480P, commit at 1080P

    Check the label and the silhouette, add whatever slipped to the clause, run again. Re-run the keeper at the higher tier with the same prompt.

    Up to 1080P

Need somebody on camera holding it instead of the product alone? The UGC ad generator puts a creator, your packshot and the spoken line in the same generation, vertical by default.
Need the same product across a whole set of PDP shots rather than one hero? Photograph it from every angle the set needs and send that group with each generation — reference to video takes up to ten.

Questions people actually ask

AI product video questions

What it does with your packshot, whether the label survives, what it costs, and whether you can sell the result

Does it use my product photos, or generate new footage?

It generates. Your packshot becomes the literal first frame, and everything after it — the room, the light, the hands, the sound — is new. It is not a slideshow of your image library and it will not pan across photographs you already have. If assembling an existing library is what you need, a template editor does that better than this.

Can it make videos for my whole catalogue?

No, and if that is the job you should buy something else. This page makes one clip, from one packshot, from one prompt. The bulk tools connect to your store, run a recipe across every SKU and push the results back onto the product pages as assets — a thousand listings in an afternoon. That is a different product and they are better at it than we are ever going to be. What is left over for us is the single shot a recipe cannot produce: the scene that was never photographed. One thing to check before you generate anything aimed at a listing — Amazon's own specification for detail-page video asks for .mp4 or .mov, up to 1080p, under 5 GB and between 1 and 12 minutes, and it only shows in the main media block when the listing carries fewer than six images. A six-second generated hero clip clears every one of those bars except the length, which is why it belongs in an ad unit or in A+ content rather than in the main block.

Will the label and logo stay readable?

Larger marks hold well; small type is the first thing to degrade, especially while the camera moves. Lock the label in the appearance clause, keep the move slow, and render the keeper at 720P or 1080P. If the type has to be legally exact, generate the shot and composite the real label over it — that is what the accounts doing regulated work here do.

Can it do a full 360° turntable of my product?

Not reliably, and it is worth knowing why: there is no 3D model of your product. The back was never photographed, so it is invented as it rotates into view. Under a quarter turn is safe; a full rotation is what a real turntable rig or a CAD render is for.

Can I use the clip in a paid ad?

Yes. Commercial use is included on every paid plan and every credit pack — you may publish, sell and license the output. Free-tier clips are for evaluation. You hold the rights to the packshot you upload, and you are responsible for the finished frame. Generation changes nothing about what an advertisement is allowed to imply: the FTC's guidance for online advertising is written around what the ad shows a viewer, not around how the footage was produced. A generated scene in which your product does something it cannot do is the same violation it would have been with a camera, and the fact that no photograph was faked is not a defence.

What kind of packshot works best?

Plain background, even light, the product filling most of the frame, shot at the angle you want the clip to open on. Busy lifestyle photographs work too, but everything else in them becomes something the model has to keep — and keeping things is where credits go.

Can I get several matching shots for one product page?

Yes, with reference images rather than a first frame. Photograph the product from every angle the set needs — up to ten — and send that same group with each generation. Matching is close rather than exact, and the two input families cannot be combined in one request. Reference to video is the page for it.

How many runs does one usable hero shot take?

Three or four, on the evidence of the accounts here doing commercial work. Draft at 480P so an iteration costs a quarter of a finished clip. A run that fails outright is refunded automatically, and there is a 7-day refund on unused credits.

Can I test it on my own product before paying?

Yes. Every account gets one free clip on the real model at 480P — an email, no card — and the signup credits cover another. Ninety seconds after you upload the packshot you will know whether your product survives being generated, which is the only question worth answering before you budget a campaign.

One packshot. The shot you could not book.

Upload the photo, name what may not change, and see your product moving in ninety seconds. First clip free at 480P — an email, no card — and commercial use is included the moment you pay for anything at all.

由 wan-3.run 编辑团队撰写与维护发布于 最后更新 工具 版本 2026.09.4