APIMODELSAPIMODELS

Wan 3.0 First Frame to Video Prompts

Sixty-four Wan 3.0 prompts from the official Wan 3.0 creator handbook, each shown next to the clip it produced: 30-second text-to-video, image, video and audio references, document-to-video, editing, extension and first-frame jobs. Hover to play, one click to copy.

← All 64 Wan 3.0 prompts

Last updated: · Wan 3.0 API — pricing & playground · API docs

TL;DR

Wan 3.0 is Alibaba's all-in-one video model: one model name covers text-to-video, first-frame and first+last-frame image-to-video, and omni-reference generation from up to 10 reference images, 5 reference videos and 5 audio clips, or straight from a document or a public web page. It outputs a native single shot of up to 30 seconds at 30fps with a synced audio track, in 480P, 720P or 1080P. On apimodels.app it is model wan-3.0-video at $0.045, $0.09 and $0.18 per second by resolution, with a prime mode that is about 6x faster for roughly 50% more. This page collects the 64 prompts and sample clips published in the official Wan 3.0 creator handbook, grouped by job type and free to copy.

Source: Wan 3.0 视频创作者手册(通义万相) (the model vendor’s official site)

Hover to preview · click for sound

First frame, ink-wash bamboo: a swordsman in three beats

(0:00 - 0:03) 运镜：中景。 画面：沿用 @Image1 的水墨竹林场景。头戴斗笠、身着墨色长袍的刀客，其剪影伫立在浓雾环绕、竹影婆娑的中心。 动作：刀客左手轻按刀柄，竹叶随风飘落。 氛围：寂静，压抑，水墨晕染的颗粒感清晰可见。 (0:03 - 0:08) 剧情引入，运镜：特写。 剧情：刀客察觉杀气。 画面：镜头快速推进到刀客斗笠下的特写。 细节：斗笠边缘露出一双凌厉、决绝的眼睛，眼神如刀。水墨晕染的“汗珠”从额头滑落。 动作：刀客右手缓缓握紧刀柄，指关节泛白。竹叶飘落的速度加快。 (0:08 - 0:13) 运镜：快速环绕。 画面：镜头以刀客为中心快速环绕。 剧情：两名同样是水墨剪影的刺客从竹林深处杀出，手持双钩和短刃。 细节：刺客的动作也是水墨笔触构成的，如墨迹流动。镜头快速掠过刺客的身影，强调速度感。 (0:13 - 0:18) 运镜：手持感，高速捕捉。 打斗：刀客猛然拔刀，刀光是一道纯白色的水墨裂痕。刺客双钩攻来。 细节： 一击：刀客回身一斩，刀刃与刺客双钩碰撞，迸发出无数黑白相间、带有书法感的火花和墨点。 二击：刀客虚晃一枪，利用竹竿弹射，从高处劈下。刺客翻滚躲避。竹竿被打断，化作飞散的墨痕。 三击：刀客刀尖直指另一刺客咽喉，刺客以短刃格挡，镜头捕捉刀刃与短刃摩擦时细微的墨迹崩裂细节。 **(0:18 - 0:20) 运镜：慢镜头至定格。** * **画面：** 刀客的刀停在刺客颈侧，刺客凝固。 * **细节：** 刀客的斗笠在打斗中略微倾斜。竹林归于平静，更多竹叶飞舞。 * **氛围：** 镜头慢慢拉远，回到中景，定格在刀客收刀、刺客倒地的画面。 * **文字（可选）：** 画面下方显现水墨风文字：“江湖，不过一笔勾销。” **风格说明：** 整个过程保持 `image_0.png` 的高度黑白、水墨画风格，动作必须有毛笔触感，速度要快，细节要足（墨滴、火花、墨痕裂口）。

by Wan 3.0 视频创作者手册 · 通义万相
Hover to preview · click for sound

First frame, rose garden: a 35mm memory

镜头 1：美好的过往（00:00 - 00:08） 35mm电影胶片，复古阳光花园，一位戴草帽、黑色项链和白色衣裙的优雅女子，站在盛开的玫瑰花墙前。风吹拂着她的发丝，她轻触帽檐，眼神温柔而带有一丝沉思。镜头缓慢拉近，自然光斑，柔和暖色调，浅景深，极其细腻的动态，4k高画质。 镜头 2：情绪的转折（00:08 - 00:15） 电影特写镜头，女子缓缓摘下头上的草帽，低下头，表情带有淡淡的怅惘与眷恋。柔和的风掠过，画面光影微变化，呈现出故事感与叙事氛围，35mm胶片质感，深沉优雅，电影感光线。 镜头 3：回归现实/车内（00:15 - 00:23） 行驶的车内后座视角，同一个女子侧颜，手枕在车门上，凝视着挂满雨滴的车窗外。窗外模糊的绿植与光影迅速退后，车内光线阴沉而温暖，侧面轮廓精致，极具电影情绪感，真实胶片颗粒，慢速摄影，优雅忧郁。 镜头 4：最后的凝望（00:23 - 00:30） 极近面部特写，女子在车窗旁缓缓闭上双眼，流露出沉静而释怀的表情。雨滴在车窗玻璃上缓缓滑落，焦距逐渐转移到玻璃雨滴上，背景人物虚化，画面逐渐淡出黑屏，王家卫风格电影感，深情且流畅。

by Wan 3.0 视频创作者手册 · 通义万相
Hover to preview · click for sound

First frame, West Lake: a mineral-pigment one-take

【一镜到底，慢速唯美运镜，无剪辑，15秒长镜头，岩彩画矿物颜料质感与3D渲染碰撞，青蓝×鎏金撞色】 全局设定：国风岩彩画质感，矿物颜料颗粒肌理，金箔质感，青蓝与鎏金强烈撞色。场景为清晨薄雾中的西湖：湖面呈哑光青蓝晕染，金色晨光穿透云层形成丁达尔光柱，远处三潭印月的石塔与雷峰塔在雾中若隐若现，近处有荷叶与荷花。画面主体是西湖水神"西子"：岩彩画质感，鹅蛋脸，柳叶眉，丹凤眼，眉心一点朱砂花钿，肌肤如羊脂白玉泛着柔光，头后悬浮一轮淡金色圆光；高挽发髻插一支鎏金荷花发簪，她的裙摆与披帛由流动的湖水构成，青蓝色渐变，表面水光粼粼。 配音与音效：背景为轻柔的古筝与竹笛，环境音有细微的水波声与清晨鸟鸣。第8秒起，一位空灵温柔的女声缓慢吟诵："欲把西湖比西子——"；回眸段接下半句："淡妆浓抹，总相宜。"语速缓慢，带轻微回响，余韵悠长。 [0-3秒：一滴水] 极近微距，清晨的荷叶边缘，一颗水珠凝聚饱满，映着金色晨光，缓缓坠落。镜头跟随水珠下摇（Tilt down），水珠滴入湖中，激起一圈金蓝色涟漪，涟漪层层扩散。 [3-7秒：水聚成形] 涟漪中心，湖水缓缓隆起，凝聚成神女"西子"的身形：湖水化作她的青蓝色裙摆与半透明披帛，细小水流在她周身环绕流动。她从湖心缓缓升起，双眸轻闭又缓缓睁开，鎏金荷花簪亮起微光。镜头缓慢环绕（Orbit）她半垂的眼眸与流水织成的裙摆。 [7-11秒：一步一莲] 她赤足踏在湖面上缓缓前行，每一步脚下都漾开一圈金色涟漪，涟漪中荷花次第绽放。镜头后撤拉远成中全景：薄雾散开，远处三潭印月的三座石塔亮起温暖灯光，雷峰塔在晨光中显现，岸边垂柳拂水。第8秒，女声旁白起："欲把西湖比西子——"她的披帛在身后拖出长长的金蓝色水光。 [11-15秒：回眸·总相宜] 她停下脚步，回眸看向镜头，镜头缓缓推近至半身特写：晨光洒在她的脸上，湖水披帛在身侧轻轻飘动，几滴发光的水珠从发梢滴落，在空中化作细碎光点。旁白接下半句："淡妆浓抹，总相宜。"她浅浅微笑，身后是晨光中的雷峰塔与朦胧的三潭印月。最后一秒画面完全定格在她的回眸特写，定格为完美的电影海报构图。

by Wan 3.0 视频创作者手册 · 通义万相

How to write great Wan 3.0 prompts

  1. 1Number your materials and address them by slot. Wan 3.0 reads references positionally: "figure 1", "video 1", "audio 1", with images, videos and audio counted separately. The handbook prompts do exactly this, for example "the girl in figure 1 wears the blouse from figure 2". Name the slot, never the file.
  2. 2Write a timeline for anything over 10 seconds. Nearly every 30-second case is written as segments: "0-4 s wide shot", "shot 2 (4-9 s) close-up". The model follows a timeline far better than a paragraph of adjectives, and saying "one continuous take" keeps it from cutting.
  3. 3Write the sound, Wan 3.0 generates it. Every clip carries a synced audio track. The handbook marks effects inline as <cloth rubbing the bar> and dialogue in {braces} or quotes; ambience, a BGM cue and a spoken line each land as written. A prompt with no sound gets generic audio.
  4. 4For edits, say what changes and what must not. Editing prompts open with "edit the video", name the element, and always close with a lock: "the woman's action and the rest of the frame stay unchanged". Without that lock the model re-imagines the whole frame.
  5. 5References carry identity, words carry the action. With a character card or portrait attached, the prompt repeats the traits that must not drift (hair, wardrobe, accessories) and spends the rest on blocking and camera. Two people need two references; two voices need two audio clips.
  6. 6Give an extension a script, not just a length. Extension prompts state the direction ("extend forward 15 s") and then write the new beats with the characters named from the existing footage, dialogue included. A bare "continue" produces filler.
  7. 7Frame mode and reference mode do not mix. On the API first_frame_url / last_frame_url and reference_* / file_url / link_url are mutually exclusive. A first-frame prompt describes motion out of that frame; a reference prompt talks to figure 1 and friends.
  8. 8State duration and aspect ratio up front. The handbook prompts open with "10 s, 16:9, 1920x1080" or "vertical 9:16, 15 s". Wan 3.0 takes 2-30 seconds and 480P / 720P / 1080P; put the numbers in the prompt and match them with the duration parameter.

Prompt formula

Duration + aspect ratio + reference slots (figure 1 / video 1 / audio 1) + timeline segments + subject with locked traits + camera move + light and palette + sound tags + what must stay unchanged

Example: 15 s, 16:9. Figure 1 is the only reference for the woman; keep her short black hair and beige trench. 0-5 s: wide shot, she pushes open a glass café door onto a cobbled street, <heels on stone>. 5-10 s: slow dolly in to a mid shot as she unfolds a paper map. 10-15 s: close-up, she looks up and says "It should be this way." Warm afternoon backlight, film grain, street ambience throughout.

Frequently asked questions

Where do these Wan 3.0 prompts come from?

All 64 prompts and clips come from the Wan 3.0 creator handbook that the Tongyi Wanxiang team published for the model launch, linked above. Every prompt is kept verbatim in its original Chinese, and the sample clips are re-hosted so they keep playing; nothing was regenerated. The reference images and documents some cases used were not published with the handbook, so those cards carry the prompt and the result only.

Where can I run Wan 3.0?

On apimodels.app as model wan-3.0-video, on the same unified video endpoint as VEO, Kling and Seedance: $0.045 per second at 480P, $0.09 at 720P, $0.18 at 1080P, billed only on success, with mode "prime" for the faster tier at roughly 50% more. Reference images, audio, documents and web pages cost nothing extra; a reference video adds its own duration to the billed seconds. No Alibaba Cloud account or mainland real-name verification is needed. The model is also available on Alibaba Cloud Model Studio and the Wan web app.

Do the prompts work in English?

Yes. Wan 3.0 accepts English and Chinese; what matters is the structure the handbook uses: numbered material slots, time segments, sound tags and an explicit "everything else stays unchanged" on edits. Translate the words and keep the skeleton. We leave the originals in Chinese because a translated prompt is not the prompt that produced the clip.

How do the editing and extension cases pass the input video?

Through reference_video_urls, with the prompt addressing it as "video 1". Edits keep the input length; extension prompts state the direction and the added seconds. Upstream caps input plus output at 30 seconds, and billing counts both.

Can I use these clips commercially?

The clips belong to their publisher and are shown here as examples of what each prompt produced. Anything you generate yourself is governed by Alibaba's terms for Wan. The prompt text you can copy, edit and reuse freely.

Run these prompts on apimodels.app

Wan 3.0 is live here alongside 160+ other image, video, audio and language models — one API, one key, pay as you go, with $0.10 of trial credit on sign-up.

Try Wan 3.0 →API docs

Other prompt libraries

Seedance 2.5 promptsMiniMax H3 promptsGPT Image 2.5 promptsClaude Opus 5.5 Video promptsClaude Fable 5.2 promptsGemini 4 Carbon promptsMuse Video promptsGPT Image 2 prompts