Happyhorse-1.0-Video-Edit-20260618

Model IntroductionVideo

HappyHorse-1.0-Video-Edit is a state-of-the-art video editing model from Alibaba that topped the LM Arena video editing leaderboard with an Elo score of 1299, marking the first time a Chinese team achieved first place in this category.[reference:0]

HappyHorse-1.0-Video-Edit is a video-to-video (V2V) editing model developed by Alibaba's ATH business group, available through Alibaba Cloud Model Studio.[reference:1] It accepts text, images, and video as input, and outputs edited video.[reference:2] The model supports natural language instructions for local or global editing of video elements, using up to 5 reference images, while precisely preserving the original motion dynamics for superior expressiveness.[reference:3] It supports 720P and 1080P output resolutions.[reference:4] The model is built on a 15-billion-parameter, 40-layer single-stream self-attention Transformer architecture that enables native synchronized audio-video generation.[reference:5]

Its key features include instruction-based video editing via natural language, multi-image reference support for precise element replacement, and strong preservation of original motion and composition.[reference:6] The model excels at style transfer and local replacement tasks, allowing users to reshape visual style, atmosphere, color, lighting, or subject details while retaining the structural skeleton of the source footage—including camera framing, dynamics, and scene composition.[reference:7] It natively supports lip-sync across seven languages: English, Mandarin, Cantonese, Japanese, Korean, German, and French.[reference:8] The model has been deployed on multiple platforms including Alibaba Cloud Model Studio, ComfyUI, and WaveSpeedAI.[reference:9][reference:10]

Benchmarks

Benchmark (Area)Score*
MechVerse (Physical Motion Consistency) (Overall)94.95 (Eval)
MechVerse (Background Consistency) (Overall)96.03 (Eval)
MechVerse (Motion) (Overall)99.43 (Eval)
MechVerse (Temporal Flicker) (Overall)99.41 (Eval)
MechVerse (Dynamics) (Overall)18.73 (Eval)
MechVerse (Aesthetics) (Overall)50.83 (Eval)
MechVerse (Imaging) (Overall)58.68 (Eval)
MechVerse (Instruction) (Overall)1.95 (Eval)
MechVerse (Physics) (Overall)3.70 (Eval)
MechVerse (Common) (Overall)1.91 (Eval)
MechVerse (Sum) (Overall)7.56 (Eval)

API request example

Endpoint /v1/video/generations. Replace <API_KEY> with a key from the console.

Async: poll `GET /v1/video/generations/{task_id}` with response `id`. Allowed size: 720p / 1080p.

curl https://api.ravenrouter.ai/v1/video/generations \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <API_KEY>" \
  -d '{
  "model": "Happyhorse-1.0-Video-Edit-20260618",
  "prompt": "a cute cat walking on grass",
  "duration": 5,
  "size": "720p"
}'

More models: API Reference → Examples by model

Happyhorse-1.0-Video-Edit-20260618 API — Video Generation Endpoint | RavenRouter AI