HappyHorse I2V Config

HappyHorse's image-to-video node is the simplest in the whole pack, and that's not a criticism. HappyHorseBailianCNI2VConfig takes one image, one prompt, and a handful of settings, and turns them into a MEDIAHUB_VIDEO_REQUEST for Alibaba's new HappyHorse model served over Bailian/DashScope. No mode picker, no keyframe choreography, no subject-reference options - just "make this picture move." If you want a focused, low-friction first frame animation, this is the one to grab.

Chain it the usual way: config → MediaHubSubmitVideo → MediaHubResolveVideo → SaveVideo, with a DASHSCOPE_API_KEY supplying the auth.

Inputs that matter

• first_frame_image - the required anchor. JPEG-encoded to base64 and uploaded with the request. This is the entire creative input, so the better this image is, the better the clip.
• prompt - the motion/action description. With a strong first frame, a short prompt goes a long way.
• model - one choice, happyhorse-1.0-i2v.
• resolution - 720P or 1080P.
• duration - default 5 seconds, range 3–15.
• watermark - on by default (the HappyHorse quirk - every other provider here defaults it off).
• seed - -1 = random; set a value for repeatable takes.

Output: request → MediaHubSubmitVideo.

Why you'd reach for it

Because sometimes the right tool is the one with the fewest decisions. If your job is "I have a picture, I want a video of it," this node is the path of least resistance through a brand-new model - and using it through the official Bailian API means you're not gambling on sketchy third-party mirrors of the weights.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/vantang/ComfyUI-MediaHub.git
/path/to/ComfyUI/python -m pip install -r ComfyUI-MediaHub/requirements.txt

Restart ComfyUI; under MediaHub/Provider/HappyHorse Bailian CN/Video. Dependencies are requests/Pillow/numpy, plus a DashScope key.

Common gotchas

Same watermark trap as the T2V sibling: it defaults to on, unlike every other provider in the pack, so check it or your "clean" clips carry a mark. And because there's no first_last_frame mode here, you can't bookend clips the way you can with the Seedance or Wan nodes - if your idea depends on a fixed ending, that's the wrong node. Input image quality is everything, and since the frame is JPEG-encoded before upload, keep it clean and well-lit. Then let ResolveVideo do its patient polling thing - minutes of queue time is normal.