alibaba/happyhorse-1.1/reference-to-videoHappyHorse 1.1 reference-to-video drives one shot from one to nine ordered reference images, holding a cast, wardrobe or location steady. Clips run 3 to 15 seconds at 720p or 1080p across nine aspect ratios, with generated sound.
Filtering differs by endpoint: Image to video — Uncensored, Reference to video — Uncensored, Text to video — Filtered.
Workbench
These examples were written for this endpoint. Replace the prompt and sample media URLs with your own accessible inputs, then check the current quote and model availability before submitting.
- Output5s
opera-box-emerald-velvet-smile [Image 1] is the person: a red-haired woman in an emerald-green velvet off-the-shoulder gown holding a gold lorgnette, shot on a plain dark backdrop. [Image 2] is the place: an empty Art Nouveau opera box with a gilded balustrade, looking out over a packed golden auditorium under a crystal chandelier. The clip puts the woman from [Image 1], in the same gown, at the balustrade of the box from [Image 2] with the lorgnette raised to her eyes; she lowers it, turns her head to the camera with a slow, knowing smile and brings a fingertip up toward her collarbone as the camera pushes in. `aspect_ratio` is `16:9`, `duration_seconds` is `5` at `720p`.
- 720p
- 16:9
- 5s
- Output5s
product-in-place [Image 1] is the candle jar and [Image 2] the marble shelf. Naming each one by its tag puts the jar on that shelf in the shelf's own window light instead of blending the two photos.
- 1080p
- 1:1
- 5s
- Output5s
rider-jacket-bike-station Four things have to stay identical from take to take: the rider, the jacket, the machine and the filling station. `reference_image_urls` carries up to nine stills in one request, which is what lets you pin a wardrobe and a vehicle instead of describing them and hoping. Here the first still already holds the rider, the bike and the station, and the second a leather jacket; the third, a woman on a neon-lit street, has nothing to do with this shot and does not appear in the clip. The prompt then only has to name the action and the light — it says what he does, not what he looks like.
- 720p
- 3:4
- 5s
- Output5s
three-plate-corridor Three references, three different jobs — the person, the guitar she carries and the corridor she walks down — resolved into one backward-tracking take. That is the honest test for this endpoint: face, prop and location all have to survive the same shot, and the prompt says which reference is which.
- 720p
- 16:9
- 5s
This example starts at $0.14 per output second. The Playground updates the estimate when you change the settings.