阿里多模态

HappyHorse

阿里巴巴下一代多模态视频模型,原生支持音视频协同生成。一个统一模型,四种生产力场景——文本、图像、多图参考与视频原位编辑。在 FireRed 修图 免费体验。

About

关于 HappyHorse

HappyHorse 是阿里巴巴下一代 AI 视频模型,基于原生多模态架构构建。单一统一模型覆盖四种生产场景——文生视频、图生视频、多图参考生视频以及视频原位编辑,支持原生音视频联合生成、720p/1080p 输出,深度适配广告营销、电商展示、短剧制作与社媒创意等内容生产场景。

关于 HappyHorse

HappyHorse 核心能力

原生多模态、一模型四场景、视频原位编辑、720p/1080p 输出

核心功能概览

多图参考生视频

提供最多 5 张参考图,HappyHorse 会将它们合成为一个连贯镜头——Image 1 出角色、Image 2 出环境、Image 3 出道具等等。模型在所有参考元素间保持身份、材质与光线的一致性,轻松完成复杂的场景搭建。在提示词中使用 Image 1 / Image 2 等标签精确绑定每个元素。

提示词
输出示例 (Output)

Image 1 中的女孩在阳光林间慢跑,Image 2 中的森林精灵像彗星般紧随其后。

双图合成——主体 + 伙伴生灵

Image 1 中的偶像站在 Image 2 的水上舞台,面对巨大的发光月亮。

舞台 + 表演者组合,保留光线与水面反射

视频原位编辑

上传源视频并描述要改变的内容。HappyHorse 会替换主体、服装或整体渲染风格(如把真人画面变成乐高、黏土或动漫),同时保持原始镜头路径、节奏与构图不变。非常适合本地化、创意重混,以及无需重拍就能快速验证视觉方向。

提示词
输出示例 (Output)

把视频中的少年替换为海绵宝宝踩滑板完成 kickflip,高品质写实三维渲染。

主体替换保留运动

把整段视频变成鲜艳的乐高世界,完整保留原有挥手动作与空间布局。

整场风格转换,运动与布局不变

HappyHorse Showcase

12 Real-world Cases

See HappyHorse in action across all four scenes: text, image, multi-image reference, and video editing.

3 Text-to-Video Cases

Generate video from pure text prompts with native audio

Text
1080p

A Pixar-style short about a nervous little traffic cone who dreams of being a finish line pylon at a major race. Other cones mock its ambitions. A construction worker accidentally places it at a marathon finish line. The cone's painted face shifts from terror to joy as runners pass. Confetti falls on its cone head. Other cones watch on TV, inspired. Audio: Traffic sounds becoming crowd cheers, inspirational swelling music.

Duration: 5s

Text
1080p

8mm vintage film style, grainy texture, slight light leaks. A group of friends laughing and running on a beach in the 1970s. Sun-drenched colors, nostalgic atmosphere, handheld camera shaking slightly. Authentic retro look.

Duration: 5s

Text
1080p

First-person POV (GoPro style), a high-speed mountain bike descent through a narrow, rocky forest trail. The camera vibrates with the bumps, trees rushing past in a blur. Intense sunlight filtering through the canopy. Adrenaline-pumping action, immersive sound of tires on gravel.

Duration: 5s

3 Image-to-Video Cases

Animate still images into motion with synchronized sound

Image
1080p
1 Image

Tracking shot as the girl walks gracefully through the meadow. Her dress and hair flutter in the wind, and clouds drift slowly. Cinematic audio of soft footsteps on grass, rustling summer wind, and melodic bird calls.

Duration: 5s

Image
1080p
1 Image

First-person POV. The camera glides smoothly and continuously forward deep into the sci-fi corridor. Glowing neon lights pass by rapidly on both sides. Tiny glowing dust particles float in the illuminated air. Steady tracking shot, immersive atmosphere.

Duration: 5s

Image
1080p
1 Image

Time-lapse effect. The thick morning mist rolls and flows fluidly through the pine trees like a slow-moving river. The bright volumetric light rays shift their angle dynamically as the sun rises. Cinematic slow zoom in.

Duration: 5s

3 Multi-Image Reference Cases

Combine up to 5 reference images into a coherent scene

Reference
1080p
ref 1
ref 2

The girl from Image 1 is jogging lightly through a sunlit forest. The glowing forest spirit from Image 2 playfully flies closely behind her like a small comet, leaving a faint luminous trail in the air. Golden light filters through the dense trees. Cinematic audio of soft, quick footsteps on grass, a gentle magical whoosh, and distant bird calls.

Duration: 5s

Reference
1080p
ref 1
ref 2

Place the cotton doll from Image 1 into the vintage room from Image 2. The doll sits on the wooden workbench, gently swinging its legs, looking around curiously. Keep the lighting of Image 2 and the plush texture of Image 1 strictly consistent.

Duration: 5s

Reference
1080p
ref 1
ref 2

The idol from Image 1 stands on the water stage from Image 2, directly in front of the giant glowing moon. The idol steps forward slowly, creating gentle ripples in the water, and raises the microphone to sing. The soft blue light from the moon reflects perfectly on the idol's outfit.

Duration: 5s

3 Video Edit Cases

Replace subjects, styles, or elements while keeping camera motion

Video Edit
1080p
Source Video

Replace the teenage boy in the video with SpongeBob SquarePants. He should retain his classic iconic look: a yellow rectangular sea sponge with large blue eyes, wearing a white collared shirt, red tie, and brown square pants. SpongeBob should be riding the skateboard naturally and performing the kickflip. Render him in a high-quality 3D realistic style to match the lighting and shadows of the real-world park background. Keep the original camera tracking and motion exactly the same.

Video Edit
1080p
ref 1
Source Video

Replace the grey hoodie and pants with the floral silk skirt from the reference image. The skirt should flow and sway naturally with the woman's walking and spinning motion. Keep her face, hair, and the living room background exactly the same.

Video Edit
1080p
Source Video

Transform the entire video into a vibrant Lego world. The person, the desk, and every object in the room should be constructed from high-quality plastic Lego bricks. Keep the original waving motion and spatial layout perfectly. The lighting should be bright and clean, like a professional Lego toy commercial.

FAQ

HappyHorse 常见问题

HappyHorse FAQ

HappyHorse 是阿里巴巴下一代多模态视频模型,原生支持音视频协同生成,并在单一统一模型中提供四个生产就绪场景:文生视频、图生视频、多图参考与视频原位编辑,深度适配广告、电商、短剧与社媒创意。

HappyHorse 支持 720p 与 1080p 输出;常用时长为 5/8/10 秒;视频编辑场景使用源视频的时长。

参考生视频与视频编辑场景最多可使用 5 张参考图。请在提示词中使用 Image 1 / Image 2 等标签精确绑定每个元素。

上传源视频并描述要改变的内容,HappyHorse 会替换主体、服装或整体风格,同时完整保留原始镜头路径、节奏与构图。适合本地化、创意重混与快速验证视觉方向。

提供每日免费生成额度。定价按时长与分辨率计费:720p 为 31 积分/秒,1080p 为 51 积分/秒。

无需注册即可试用。创建账号可保存历史记录、解锁更长时长并追踪积分余额。

Testimonials

创作者们对 HappyHorse 的评价

HappyHorse 让我们用同一份 brief 产出四种风格的产品视频——多图参考真是效率神器。

林梅

电商创意总监

林梅: “HappyHorse 让我们用同一份 brief 产出四种风格的产品视频——多图参考真是效率神器。

汤玛斯: “原生音视频协同生成正是短剧制作所需——不再需要单独配音与拟音环节。

佐藤莉佳: “视频原位编辑是真正的亮点,一个午饭前就能试完五种视觉方向,完全不用重拍。

朴丹尼: “文生/图生/参考/编辑一体化,使团队工作流高度紧凑。HappyHorse 已成为我们管线中的常驻模型。

探索更多AI视频模型

开始用 HappyHorse 创作

体验 HappyHorse——阿里巴巴的多模态视频模型,在线免费使用

user 1
user 2
user 3
user 4
user 5

10,000+ users