HappyHorse 1.1: The Model Behind Your Videos
txtovid runs on HappyHorse 1.1 (happyhorse-1.1-t2v), a text-to-video model served through Alibaba Cloud Model Studio. It's engineered for two things above all: physical realism and motion smoothness — the two qualities that separate 'AI-looking' clips from footage people believe.
Technical specifications
- Model: happyhorse-1.1-t2v (successor to happyhorse-1.0)
- Input: text prompts up to 5,000 characters, any language
- Output resolutions: 720P and 1080P tiers
- Aspect ratios: 16:9, 9:16, 1:1
- Typical generation time: 1–5 minutes per clip
- Serving: asynchronous task API on Alibaba Cloud Model Studio
What HappyHorse 1.1 does exceptionally well
Physics first: water, cloth, hair, smoke and particle motion behave believably — the model was clearly optimized against the classic AI-video artifacts of floating objects and morphing limbs. Camera language compliance is strong: dolly, orbit, crane and tracking instructions translate into recognizable cinematography. And motion stays smooth across the full clip length instead of degrading toward the end.
Stylistically it spans photorealism, cinematic film looks, anime and illustration styles, and handles text-light scenes best — like most video models, it's not built for rendering long readable text.
How txtovid uses it
We submit your prompt with your chosen resolution, ratio and duration to the HappyHorse task API, poll until rendering completes, then immediately archive your video to permanent storage (the raw model output URL expires after 24 hours). Every generation passes Alibaba's content moderation on both input and output — prompts violating content policy are rejected before any credits are consumed permanently (blocked generations are refunded automatically).
By default we randomize the generation seed for every run ('Surprise me'), so identical prompts explore fresh variations; fix the seed to reproduce a result.