Scenes with complex physics
Sports, acrobatics, object interactions: HappyHorse accurately simulates body and object motion physics on screen without losing their geometry.
Video generation with HappyHorse 1.0
Video generation with HappyHorse 1.0
Flami lets you create videos with HappyHorse 1.0 by Alibaba ATH, a model that outperforms Seedance 2.0, Kling 3.0, and Grok Imagine on public Artificial Analysis benchmarks for motion stability and physics accuracy.
Why HappyHorse 1.0
A new model from Alibaba ATH that topped public AI video benchmarks thanks to stable motion physics, accurate interaction simulation, and support for up to nine characters in one scene.
The best motion stability in the industry
HappyHorse keeps object geometry and body proportions intact through complex motion. A hula hoop stays in frame during a squat, a ball rolls along the right path, with no drifting or the usual AI distortions.
Realistic reflections and optics
Accurate reflection physics in mirrors, chrome, and water. The reflected image moves in sync with the character, keeping proportions and lighting right, something most competitors can't pull off.
Accurate liquid and texture simulation
Stable pours, coffee foam, latte art, water droplets, smoke. Liquid follows real surface physics instead of jittering the way it does in models without dedicated simulation.
Up to 9 characters in one scene
Reference-to-Video mode lets you submit up to nine reference images and use them as different characters in one clip. Each one keeps its look throughout the scene.
What HappyHorse 1.0 can do
Duration
3-15 seconds per clip (5 by default)
Quality
720p, 1080p
Formats
16:9, 9:16, 1:1, 4:3, 3:4
Inputs
Text plus up to 9 reference images, plus an optional finished video to edit
Sound
Generate with or without sound, whichever fits the task
Speech
Multilingual support for global content
Creation methods
From a description, from a source image, from references tied to characters, or by editing an existing video
HappyHorse 1.0 video for any task
Scenes with multiple characters
Up to nine different characters in one scene, each tied to a reference image: for short narrative clips, group scenes, and multi-character ads.
Ad clips with realistic reflections
Accurate reflections in mirrors, chrome, glass, and water: for premium products, fashion clips, and lifestyle shoots where realistic surfaces matter.
Multilingual content for global campaigns
Video for international markets with support for different languages: for brands, localized campaigns, and content for multiple regions.
Editing a finished video
Swap objects, transfer style, or make local edits to an already-shot clip, all with one text instruction, no regenerating from scratch.
How it works
Step 1. Write a prompt
Describe the video you want: the scene, mood, camera movement, characters. If there's more than one character in frame, describe each one using "character1", "character2", and so on.
Step 2. Add an image
Upload up to 9 reference images: one for a single character, several in order for complex scenes with multiple characters. You can also upload a finished video to edit it with a text instruction.
Step 3. Get your finished video
Click Generate, and Flami will create your HappyHorse 1.0 video with your settings in a few minutes.
Create your first video with HappyHorse 1.0
One workspace, all the world's best video models