HappyHorse Image to Video

HappyHorse Image to Video is the first-frame animator of the HappyHorse family: one image in, a moving shot out, with the output's aspect ratio derived from that image rather than chosen from a menu. That last part is the quietly handy feature - you frame your shot exactly how you want it in the image, and the video respects it, no ratio guessing. It's a cloud Wan-family model (partner/video/Wan category) running through Comfy's API, so no local GPU, no checkpoint files, just a prompt, an image, and credits.

It sits between the other two HappyHorse nodes in complexity: simpler than Reference-to-Video (no character references to manage), more controlled than pure Text-to-Video (an image anchors the composition). If you've already got a frame you love - generated locally, cropped to taste - this is the node that gets it moving.

The inputs

• first_frame - the image to animate. Because the output aspect ratio comes from this image, what you feed is what you get: a portrait crop yields a portrait video, a 16:9 crop yields 16:9.
• model - the dynamic combo that holds the real controls. Choose happyhorse-1.1-i2v or happyhorse-1.0-i2v to reveal:
  • prompt - what happens in the video (English or Chinese supported, same as the family).
  • resolution - 720P or 1080P.
  • ratio - offered, but the first-frame-derived ratio is the one that counts; keep them consistent or the image gets framed against a mismatched canvas.
  • duration - 3 to 15 seconds.
• seed - generation seed; standard disclaimer (input, not reproducibility guarantee).
• watermark - add an AI-generated watermark or not.

What comes out

A single VIDEO output.

Gotchas

• Crop before you send. The node doesn't reframe for you - the aspect ratio rides along in the image pixels. If you feed a weirdly-cropped frame you get a weirdly-framed video.
• The prompt still matters. An image anchors the scene, but motion and action come from text; a blank prompt gives a static, breathing-still look.
• Pick the draft tier. 720P / the 1.0 model for tests, 1080P / 1.1 for the final - the standard two-tier credit discipline.
• The model dropdown hides the prompt - same dynamic-combo behavior as the rest of the family; expand the model to edit the text.

The workflow to steal: make and crop a strong first frame locally, drop it in here with a motion-focused prompt, and let the cloud do the animating - with the aspect ratio you actually framed, not the one a menu gave you.