2026-04-28 09:56:44 +08:00
2026-04-28 09:56:44 +08:00
2026-04-28 09:56:44 +08:00
2026-04-28 09:56:44 +08:00
2026-04-28 09:56:44 +08:00
2026-04-28 09:56:44 +08:00

ComfyUI-Happyhorse-Wrapper

ComfyUI custom nodes for HappyHorse 1.0 video generation. The nodes create a task, poll for completion, then return the temporary result URL and task ID.

Two endpoint styles are supported:

  • DashScope official endpoints such as https://dashscope.aliyuncs.com, using /api/v1/services/aigc/video-generation/video-synthesis.
  • AiHubMix OpenAI-compatible endpoints such as https://aihubmix.com/v1, using /v1/videos.

Nodes

  • ComfyUI-Happyhorse-Wrapper Text To Video
  • ComfyUI-Happyhorse-Wrapper Image To Video
  • ComfyUI-Happyhorse-Wrapper Reference To Video
  • ComfyUI-Happyhorse-Wrapper Video Edit
  • ComfyUI-Happyhorse-Wrapper Preview Video

Setup

Install dependencies in the same Python environment used by ComfyUI:

python -m pip install -r ComfyUI/custom_nodes/ComfyUI-Happyhorse-Wrapper/requirements.txt

Create config.local.json in this folder:

{
  "api_key": "sk-...",
  "endpoint": "https://dashscope.aliyuncs.com",
  "poll_interval": 15,
  "request_timeout": 60,
  "upload_timeout": 120
}

The plugin also reads DASHSCOPE_API_KEY or AIHUBMIX_API_KEY from the environment. Use https://dashscope-intl.aliyuncs.com for Singapore-region keys.

With the local AiHubMix endpoint tested during development, happyhorse-1.0-t2v and authenticated content download worked through /v1/videos. Official DashScope endpoints keep the documented media URL payload for I2V/R2V/video-edit.

Media Uploads

HappyHorse media inputs require public HTTP/HTTPS URLs. The image nodes accept ComfyUI IMAGE tensors, save them as PNG files, upload them to tmpfiles.org, and send the resulting public URL to the API.

Generated video URLs expire, so connect generation nodes to Preview Video to save the MP4 under the ComfyUI output directory.

Model Notes

  • Text To Video uses happyhorse-1.0-t2v and supports ratio.
  • Image To Video uses happyhorse-1.0-i2v; the output aspect ratio follows the first frame, so there is no ratio input.
  • Reference To Video uses happyhorse-1.0-r2v; image batch order maps to character1, character2, and so on in the prompt.
  • Video Edit uses happyhorse-1.0-video-edit; provide one public MP4/MOV URL and optional reference images.
S
Description
No description provided
Readme
50 KiB
Languages
Python 90.9%
JavaScript 9.1%