Skip to main content

HappyHorse 1.1

Alibaba's cinematic video model (Singapore region) — text-to-video, image-to-video, and reference-to-video with up to 9 reference images. Version 1.1 improves physical realism, motion smoothness, and cross-shot consistency over 1.0.

Capabilities​

FeatureSupport
Text-to-VideoYes
Image-to-VideoYes (first frame)
Reference-to-VideoYes (up to 9 images)
Resolutions720p, 1080p
Aspect Ratios16:9, 9:16, 1:1, 4:3, 3:4, 4:5, 5:4, 9:21, 21:9
Duration3–15s continuous (1s granularity)
Frame Rate24 fps
SeedYes
Native AudioNo (silent output)
Negative PromptNo

Reference-to-Video​

Connect up to 9 upstream images and the model uses them as identity and style anchors — characters, products, environments. Use @-mentions to assign roles:

@[Image 1] walks through the rain-soaked market from @[Image 2],
camera tracking alongside.

Duration & Seed​

  • Continuous 3–15s range — like Seedance, any integer duration; note the floor is 3s (Seedance starts at 4s).
  • Seed support — one of the few video models here with deterministic reruns. Lock a seed to iterate on the prompt while keeping the motion signature.

What 1.1 Improves Over 1.0​

  • Physical realism — fewer object interpenetrations, more believable weight and contact.
  • Motion smoothness — less frame-to-frame jitter on fast action.
  • Cross-shot consistency — subjects keep their identity through camera changes within the clip.

Prompting Tips​

  • Describe physics you want to see. The 1.1 physicality upgrade responds to prompts like "cloth billows as she turns", "water splashes on impact".
  • Anchor identity with references, direct action with text. References pin who/what; the prompt should carry what happens.
  • Silent output — plan audio separately. Pair with Seed Audio 1.0 or ElevenLabs Sound Effects for the soundtrack.

Limitations​

  • No native audio
  • No first+last frame interpolation (single first-frame anchor only)
  • No video or audio reference inputs (images only)
  • No negative prompt

See Also​