Alibaba just opened public API access for HappyHorse 1.0, the model currently ranked #1 on Video Arena’s blind tests.
What caught my attention isn’t only the ranking. It’s the shape of the API:
-
text-to-video
-
image-to-video
-
reference-to-video with up to 9 refs
-
natural-language video editing
And the pricing is simple enough to reason about: 0.9 RMB/sec at 720P, 1.6 RMB/sec at 1080P.
What I’m wondering is this:
Does controllability matter more than raw quality now?
The launch examples suggest HappyHorse is very good at following camera language and multi-shot structure. That feels more important for real products than one-off impressive outputs.
For example:
-
“Shot 1 / Shot 2 / Shot 3” prompt structure
-
explicit camera motion
-
style-first prompts for anime or cinematic looks
-
image-to-video workflows that preserve subject identity better than pure t2v
That sounds like a model designed for pipelines, not just experimentation.
I’m also curious whether teams here would prefer:
-
one strong general video endpoint, or
-
separate endpoints like HappyHorse has for t2v / i2v / r2v / edit
My current take: splitting the API by workflow is the right call. It reduces ambiguity and makes production integration easier.