跳到主要内容
AI 视频模型
Alibaba
Wan 2.7

Wan 2.7 Video

Wan 2.7 提供文生视频、首帧或首尾帧图生视频,以及已有视频的编辑请求。当前可选 720p 与 1080p 档位;先按素材选择方式,再核对时长、可用参数和预扣积分。

请求方式
文生视频、图生视频、视频编辑
请求档位
720p 或 1080p
请求时长
文生/图生 2~15 秒;视频编辑 2~10 秒,或沿用输入时长
每秒预扣
720p 16 积分;1080p 24 积分
快速了解
  • 根据是否上传图片或视频,分别发起文生、图生或视频编辑请求。
  • 图生可用首帧图,也可同时提供首尾帧图;它们是画面参考,衔接效果需要检查。
  • 请求档位有不同的每秒预扣费率,可结合本轮时长安排试片。
  • 可设置提示词扩写、水印、负面提示词和随机种子;音频处理只用于视频编辑。

相关查询: Wan 2.7 在本站可以怎样生成或编辑视频? · Wan 2.7 怎么使用首帧和尾帧图片? · Wan 2.7 不同方式的请求时长怎样设置?

示例

Cinematic scene

15-second cinematic action sequence inside a massive battlefield filled with destroyed giant mechs half-buried in snow and ash, red warning beacons blinking weakly through blizzards and smoke. 0.0–3.0s: Camera tracks through colossal mechanical wreckage as a lone armored pilot climbs across the rusted skull of a fallen mech carrying an energy spear. 3.0–6.0s: A rival mercenary activates from inside another wrecked titan, glowing reactor core illuminating the storm around him. 6.0–9.0s: Spear clash detonates sparks and snow outward while loose armor plates crash down around the battlefield in slow motion. 9.0–12.0s: Heavy grounded combat through gigantic mech corpses, fighters leaping between wreckage while reactor fires burn beneath the snow. 12.0–15.0s: Huge cinematic pull-back reveals endless fields of fallen titans stretching across the frozen wasteland.

Cinematic scene

A fast paced cinematic fashion montage featuring a stylish young woman wearing a black leather jacket baggy light blue jeans a white crop top and dark sunglasses The video begins with a fast tracking shot of her running past metal shopping carts in a cool toned concrete parking garage The camera transitions rapidly using whip pans and barrel rolls An extreme low angle fisheye shot reveals her standing against a tall concrete building under a bright overcast sky This immediately cuts to a high angle drone shot looking straight down as she spins in an empty parking lot Digital glitch effects and visual warp distortions appear briefly between cuts The sequence features dynamic hand transitions where the subject reaches for the lens It concludes with a low angle tracking shot of her walking confidently forward The lighting shifts between dim fluorescent garage lights and bright overcast daylight Highly realistic motion blur cinematic color grading natural skin textures and authentic clothing physics Upbeat electronic music plays in the background aspect ratio 16 9 duration 12 seconds

Cinematic scene

Create a 30-second cinematic Korean-drama-style mystery video featuring the same young Korean woman throughout the entire video. She has long dark hair, natural makeup, a beige trench coat, black skirt, white sneakers, and carries a transparent umbrella and the same black smartphone. Set the entire story on one rainy night in Seoul, using the same subway entrance and platform. STRICT CONTINUITY — VERY IMPORTANT This must feel like one continuous unfolding story, not separate disconnected clips. Every action must begin exactly where the previous action ended. Keep the woman's face, hairstyle, clothes, umbrella, phone, environment, weather, lighting, and proportions identical throughout. The subway entrance, stairs, signs, platform, train tracks, and background must remain consistent. No teleporting, no sudden wardrobe changes, no location jumps, no character redesigns, no random cuts, no skipped actions, no unexplained time jumps. Camera movement should naturally follow the character. 0–5 SEC Begin with a cinematic wide shot of a rainy Seoul street at night, neon signs reflecting on the wet pavement. The same young woman walks slowly toward a subway entrance while holding her transparent umbrella. Camera smoothly follows behind her. She reaches the entrance and walks down the first few stairs. 5–10 SEC Continue directly from the stairs into the same subway entrance. Her smartphone vibrates inside her hand. She stops naturally on the landing, closes the umbrella, and looks at the screen. A mysterious message appears: “MESSAGE FROM YOU — TOMORROW” A second line appears: “DON’T GET ON THE NEXT TRAIN.” Her expression changes from confused to concerned. She looks toward the platform stairs. 10–15 SEC She slowly walks down the remaining stairs, still holding the phone. The camera follows her continuously from behind and then moves beside her. She reaches the same platform seen from the staircase. A train approaches from the tunnel. Headlights illuminate her face. She looks at the arriving train, then back at her phone. Her breathing becomes nervous. 15–20 SEC The train stops directly in front of her. She does not board. Instead, she steps closer to the train window and sees her reflection in the glass. At first everything looks normal. Then she notices something impossible: Her reflection is smiling. The real woman remains completely serious. She slowly raises one hand toward the glass. Her reflection raises its hand too—but with a slight delay. She immediately steps backward. 20–25 SEC Her phone vibrates again. Without changing location, she looks down at the SAME phone. A new message appears: “NOW LOOK BEHIND YOU.” Her eyes widen. She slowly turns her head first, then gradually turns her entire body toward the empty end of the SAME platform. Keep the camera following her movement continuously. 25–30 SEC At the far end of the platform stands another woman who looks exactly like her, wearing the identical beige trench coat and holding an identical transparent umbrella.The woman in the foreground freezes.The mysterious double slowly smiles. The phone screen lights up one final time: “YOU SHOULDN’T HAVE COME.” The subway lights flicker once. The double takes one slow step forward. The real woman takes one slow step backward.End on a dramatic close-up of the real woman's terrified eyes as the screen cuts smoothly to black. VISUAL & AUDIO STYLE Premium Korean thriller/K-drama cinematography, realistic Korean city atmosphere, rainy neon reflections, shallow depth of field, subtle handheld camera movement, natural facial expressions, realistic wet surfaces, cinematic lighting, atmospheric suspense music, distant subway announcements, footsteps, rain, train brakes, phone vibration, and realistic ambient sound.

Cinematic scene

Create a highly realistic cinematic horror sequence set inside a quiet apartment building late at night. A young woman enters an empty elevator alone, wearing simple casual clothes, with natural realistic facial features and subtle nervous expressions. The elevator moves strangely, stopping at several floors. Every time the doors open, the exact same long, empty apartment corridor appears outside, dimly lit by flickering ceiling lights. She becomes increasingly confused and frightened, repeatedly pressing the elevator buttons and looking down the corridor. On the final stop, the elevator doors slowly slide open. At the far end of the corridor stands another young woman who looks exactly identical to her, motionless and staring directly at her. The atmosphere becomes completely silent. The woman inside the elevator slowly steps forward, unable to understand what she is seeing. The identical woman suddenly tilts her head at an unnatural angle and takes one slow step toward her. Without warning, the corridor woman violently charges toward her with extremely fast, unnatural movement. The original woman screams and tries to retreat into the elevator, but the other woman tackles her and both fall to the floor. The corridor woman grabs her and aggressively drags her backward across the floor while the original woman desperately tries to hold onto the elevator doorway. The elevator lights flicker rapidly, casting harsh moving shadows across both faces. The original woman manages to reach one hand toward the elevator control panel, desperately trying to press the door-close button. The corridor woman suddenly grabs her wrist and pulls her away. The elevator doors begin closing by themselves, slowly separating the two women. The original woman reaches toward the narrowing gap in panic while the identical woman stares at her with an eerie, emotionless expression. The doors finally slam shut. Cut to a completely empty corridor. No one is visible. A smartphone lies on the floor where the struggle happened, its screen still glowing and displaying 2:13 AM. The hallway lights flicker once, then everything goes completely dark. Ultra-photorealistic live-action horror, realistic human movement, believable apartment building textures, natural skin detail, subtle handheld camera movement, cinematic low-light exposure, practical fluorescent lighting, deep shadows, atmospheric tension, realistic motion blur, disturbing but non-gory horror, no supernatural visual effects, no text overlays, no subtitles, no cartoon appearance.

Cinematic scene

CAMERA: DV 16mm tape camcorder, handheld POV by CHASE, occasionally propped on vanity. Natural shake, imperfect framing, delayed focus, clumsy zooms, face occasionally cropped. Camera never visible. LOOK: Soft blurry tape, faint tape noise, bloomed vanity bulbs, flickering auto-exposure, muted contrast, natural skin. STYLE: Casual beauty vlog. Relaxed, chatty, slow pacing with quiet mirror moments. CHARACTER: CHASE, Korean idol (20s), long straight black hair, dewy skin, coral lips, robe/makeup cape. Unseen stylist's hands occasionally enter frame. SETTING: Bulb-lit vanity, styling chair, hair tools/products on counter, handheld mirror. STORYBOARD: 1. Propped: Waves. "Hey guys, getting my hair done." 2. Handheld CU: Hair tug. "This always takes forever." 3. Insert: Stylist clips hair. Ambient only. 4. Propped: Smiles. "We're getting to the good part." 5. Handheld CU: Watches mirror, smiles. Ambient. 6. Handheld: Checks back with mirror. "Let me see the back—" 7. Tight: Grins. "I really like this." 8. Selfie: Waves/thumbs up. "Hair's done—see you guys!"

Cinematic scene

15-second cinematic action sequence inside a rotating zero-gravity prison block orbiting a dystopian planet, flickering emergency lights, shattered debris drifting through spinning corridors. 0.0–3.0s: Camera tumbles weightlessly through a damaged prison corridor as a bald cybernetic man launches himself off walls using magnetic gloves. 3.0–6.0s: Riot-control soldiers enter through an exploding bulkhead with illuminated riot shields and compact pulse weapons. Gravity flickers violently. 6.0–9.0s: Mid-air fight erupts in zero gravity, bodies spinning through floating debris while muzzle flashes illuminate the dark corridor. 9.0–12.0s: Combat intensifies through rotating prison modules as gravity repeatedly shifts direction, throwing debris and fighters across the environment. 12.0–15.0s: Pull-back reveals the entire prison ring rotating slowly above the planet while explosions spread across multiple sections.

一句话了解

这个模型怎么样?

不上传图片或视频时,从文字起稿;上传 1 张图片时,将它作为首帧参考;上传 2 张图片时,分别作为首帧和尾帧。上传视频后进入编辑请求,可再加 1 张参考图。各方式的时长和画幅规则不同,首尾帧也不能保证成片逐帧贴合参考。

模型定位
适合需要在文字起稿、图片动画化和视频编辑之间切换的短片流程。选型依据是已有素材和本次要调整的内容;镜头衔接、动作和细节仍应通过小样检查。
适合谁用
可用于尝试商品动效、人物镜头、社交短片和概念预演。需要首尾画面参考时选图生方式,需要基于原片调整时选视频编辑,再逐项核对结果是否适合用途。
本站使用说明
Kyeo 将不同请求方式的素材要求、参数和预扣积分放在同一工作台,方便提交前检查,也便于按相同素材条件比较其他模型。

关键参数一览

根据这些信息,判断模型是否适合你的任务。

模型类别
AI 视频模型
厂商名称
Alibaba
模型家族
Wan 2.7
任务形式
异步生成
文生视频
不上传图片或视频时,从文字发起请求
图生视频
1 张首帧图,或 2 张首尾帧图
视频编辑
1 个视频,可附加 1 张参考图
画幅设置
文生与视频编辑可选 16:9、9:16、1:1、4:3、3:4;图生请求不传此项

它还叫什么

同一个模型在公开资料中可能有不同名称,这里整理相关称呼,方便查找和比较。

Wan 2.7
通义万相 2.7
Wan 2.7 Video

用户常见问题

从任务、输入条件和结果要求出发,整理选择模型时需要确认的问题。

Wan 2.7 在本站可以怎样生成或编辑视频?
Wan 2.7 怎么使用首帧和尾帧图片?
Wan 2.7 不同方式的请求时长怎样设置?
Wan 2.7 图片和视频需要满足哪些上传条件?
Wan 2.7 720p 与 1080p 怎样预扣积分?
Wan 2.7 视频编辑怎样选择原声处理?
Wan 2.7 与 Wan 2.5 的输入方式有什么区别?
怎样为商品或人物镜头准备 Wan 2.7 小样?

选型指南

根据输入素材、任务要求和积分成本,逐项确认模型是否适合。

1
按已有素材确定方式

只有文字时从文生开始;已有开场图或首尾画面时用图生;需要基于原片调整时用视频编辑。先确认素材类型,再写这次需要的变化。

2
先控制试片时长与档位

主体、动作和构图尚未确认时,可先用 720p 与较短请求时长试片。同等时长下该档位的预扣更低,但是否需要返工仍取决于实际结果。

3
区分输入检查与成片效果

文件通过上传检查,只说明满足当前输入条件。首尾画面、人物外观、声音与画面衔接都需看结果,不能根据相关模型或其他平台的功能推断本站表现。

模型对比

把常见候选放在同一任务下比较,帮助你确认输入方式、控制项和成本差异。

Wan 2.7 与 Wan 2.6 的请求设置有什么区别?
只有文字或图片时,怎样比较 Wan 2.7 与 Wan 2.5?
视频编辑与多素材参考任务,怎样比较 Wan 2.7 与 Seedance 2?
Wan 2.7 720p 与 1080p 的预扣费率有何差别?

模型横向对比表

按同一组维度比较当前模型和候选模型的输入、设置与使用边界。

本站输入方式
Wan 2.7 Video
文字、首帧/首尾帧图片,或视频编辑
Wan 2.6
文字、图片或视频参考
Wan 2.5
文字或图片输入
文生/图生时长设置
Wan 2.7 Video
2~15 秒整数
Wan 2.6
选择预设时长档位
Seedance 2
在允许范围内填写整数秒数
分辨率设置
Wan 2.7 Video
720p、1080p 请求档位
Wan 2.5
在表单中选择分辨率档位
Seedance 2
可选规格还受参考视频条件限制
参考素材
Wan 2.7 Video
图生最多 2 张图;编辑用 1 个视频,可加 1 张图
Wan 2.6
图片输入或视频参考,按方式使用
Seedance 2
多素材参考支持图片、视频与音频
预扣依据
Wan 2.7 Video
分辨率费率乘计费时长后向上取整;编辑沿用时长时读已核实的输入秒数
Wan 2.5
按请求分辨率和时长档位预扣
Seedance 2
有参考视频时,其输入时长也参与预扣

适用情况

依据公开资料与当前站内设置,说明模型适用的任务及需要注意的限制。

首尾帧提供参考,动作仍要写清

首尾图分别提供开场和收束画面参考。提示词应说明中间希望发生的动作与镜头变化;跨度较大的变化可拆开测试,再检查衔接是否合适。

把返工也纳入试片预算

请求档位和时长决定本轮预扣,重新提交时也需为新任务的预扣安排余额,最终费用按各任务的结算结果确认。先验证主体与运动方向,再决定是否换档位或延长请求,便于安排试片预算。

能力详解

文字起稿

无需上传图片或视频,可提交 2~15 秒整数时长的生成请求。可选画幅和分辨率,并设置负面提示词、扩写、水印与种子。

首帧或首尾帧图生

1 张图作为首帧参考,2 张图分别作为首尾帧;支持 JPG、PNG、WebP,单张不超过 30 MiB。图生请求不单独传画幅参数。

基于原片编辑

上传 1 个 MP4 或 MOV,可加 1 张参考图。输入视频需为 2~10 秒,单个不超过 30 MiB;可请求自动处理音频或保留原声,结果仍需试听。

适合做什么

商品短片试样

从商品图尝试轻微运镜、材质表现或场景变化。包装文字、标志和关键细节应逐项检查,再决定是否继续制作。

人物与角色镜头

用首帧图提供人物起始外观参考,或用首尾图描述动作起止。大幅动作可拆成短镜头测试,并检查人物外观是否出现偏差。

原片画面调整

在满足输入限制的原片基础上提出画面调整要求。需要原声时选择对应音频处理,再核对成片声音、节奏和画面是否符合用途。

提示词技巧

先写主体与动作

先说明谁在做什么,再补环境、镜头和需要保持的细节。多个相互冲突的动作宜分开尝试,便于判断结果。

给镜头一个主运动

推近、拉远、平移或环绕可先选一种作为主运动。需要组合运动时分轮测试,记录每次改动与结果之间的关系。

有参考图时说明接下来的变化

参考图已提供外观与构图线索,提示词可重点写动作、镜头变化和希望保持的细节。写明这些要求后,仍要检查输出是否做到。

为什么选它

同一模型提供文字起稿、首尾帧图片参考和视频编辑请求,便于按已有素材切换。
首尾帧输入可分别提供开场与收束画面参考,适合据此规划试片。
文生和视频编辑可选横屏、竖屏、方形及 4:3、3:4 画幅;图生不单独传此项。
预扣按分辨率费率和计费时长计算,沿用原时长时采用已核实的输入视频秒数。

使用前要知道

当前工作台没有视频续写或独立音频参考入口;视频编辑的音频处理选项不等于上传独立音频。
文生与图生请求为 2~15 秒整数;编辑请求为 2~10 秒整数,或设为 0 沿用符合输入限制的原片时长。这些是请求规则,不保证实际成片时长。
首尾帧、画幅与编辑音频都是请求条件,不能保证逐帧衔接、人物一致或原声保留效果。
请求含固定开启的安全检查,前台没有关闭选项;输出细节、内容是否合适及素材权利仍需人工核对。

可调参数

根据这些信息,判断模型是否适合你的任务。

请求分辨率
可选
参数类型: 选择
默认 720p
720p
1080p
请求时长
可选
参数类型: 整数
默认 5 秒
文生/图生为 2~15 秒整数;视频编辑为 2~10 秒整数,0 表示沿用输入视频时长
画面比例
文生/视频编辑
参数类型: 选择
默认 16:9
16:9
9:16
1:1
4:3
3:4
负面提示词
可选
参数类型: 文本
最多 500 字符(含首尾空白),发送前去除首尾空白
提示词扩写
可选
参数类型: 开关
默认开启
开启
关闭
AI 生成水印
可选
参数类型: 开关
默认关闭
关闭
开启
编辑时的音频处理
仅视频编辑
参数类型: 选择
默认自动处理
自动处理
保留原声
随机种子
可选
参数类型: 整数
0~2,147,483,647,可留空

积分消耗

每秒预扣:720p 16 积分;1080p 24 积分

提交时按请求时长预扣积分:720p 每秒 16 积分,1080p 每秒 24 积分,费率乘以时长后向上取整。视频编辑将时长设为 0 时,使用已核实的输入视频时长计算。成功任务只有在状态已核实、实际费用有效且向上取整后不超过预扣时,才按实际费用结算并退回差额;费用缺失、无效、超出预扣或状态待核时,会保留预扣等待核对,不追加扣费。

使用建议

准备多次生成时,建议先用少量素材测试,核对输入设置和结果是否适合任务,再决定是否继续。每次生成均按工作台显示的计费规则扣点。

常见问题

相关模型

按你的任务比较这些模型的输入要求、功能和当前状态,再查看对应详情。

Wan 2.6

不附媒体时使用文字生成视频,恰好 1 张图片时使用图片生成视频,1–3 个视频时使用视频编辑;图片与视频不能混用。 默认为 1080p、5 秒、镜头结构关闭、内容过滤开启。

AI 视频模型
70–315 积分 / 次

Wan 2.5 Video

了解 Wan 2.5 Video 在 Kyeo AI 的文生与单图视频流程、5/10 秒、720p/1080p、提示词差异和 60–200 点计费。 Kyeo 按分辨率和时长精确计费:720p 5 秒 60 点、720p 10 秒 120 点、1080p 5 秒 100 点、1080p 10 秒 200 点;默认 720p 5 秒为 60 点。Wan 2.5 的 60/100/120/200 点如何计算? — 了解 Wan 2.5 Video 在 Kyeo AI 的文生与单图视频流程、5/10 秒、720p/1080p、提示词差异和 60–200 点计费。

AI 视频模型
每次 60/100/120/200 点

Seedance 2.0

从文字描述、首帧或参考素材出发制作视频。标准版提供文字、首尾帧和多模态参考场景;先选与素材相符的路径,再核对分辨率、预计积分和结果中需要检查的内容。

AI 视频模型
预计预扣 164–3,840 积分/次

信息来源

页面内容依据公开模型资料与 Kyeo AI 的模型信息整理。能否在线使用,请以当前模型状态为准。

来源说明

本页说明 Kyeo 当前的请求方式、输入检查和预扣积分。分辨率、时长及首尾帧均为请求设置或参考条件,不能据此保证成片规格、画面衔接或编辑效果。

内容复核日期: 2026-08-29
Wan 2.7 站内请求规则
平台

依据当前模型配置、素材校验、请求构造和预扣实现说明本站用法;相关项目或平台的功能不能据此视为本站已开放,名称也不表示官方直连或素材授权。