Start with the provider and operation
A model name is only the beginning of a production decision. Record where you will run it, which model identifier is selected and what the input means. A still used as an opening frame establishes a specific starting composition. A reference set can instead communicate a subject, scene or visual style. Those roles are not interchangeable.
This is a research and preparation guide, not a hosted HappyHorse generator. It does not submit a job, create provider access or demonstrate a tested HappyHorse output. The goal is to make a small request that you can evaluate before expanding a sequence.
Keep first-frame and reference modes distinct
Alibaba’s HappyHorse reference-to-video model page, updated October 11, 2026, identifies happyhorse-1.1-r2v and supports up to nine reference images. That maximum is not a requirement to fill every slot. A focused reference set can be easier to inspect than several images that contradict each other.
The separate first-frame API reference uses happyhorse-1.1-i2v as its example and requires exactly one first-frame image. It documents JPEG, JPG, PNG and WEBP, a minimum of 300 pixels on each side, an aspect ratio between 1:2.5 and 2.5:1, and a file-size limit of 20 MB. Recheck the actual operation before preparing an upload.
Assign one role to every input
| Input role | Prepare | Review |
|---|---|---|
| Opening frame | An approved composition with a clear subject | The beginning matches the source |
| Subject reference | A readable owned character or product | Identity survives the requested action |
| Scene reference | A coherent setting and perspective | Spatial relationships remain understandable |
| Style reference | A deliberate material or visual treatment | The look stays consistent through motion |
These are planning roles, not additional API field names. Use the controls and input format documented by your selected provider. Download the blank reference review sheet to record why each source is present and what you observed after generation.
For a simple invented source, download the original lamp planning image. It is code-authored illustration, not a real product photograph or a HappyHorse result.
Brief one visible action first
Use the approved original desk-lamp image as the opening composition. Keep the lamp body, base, shade color and table placement recognizable. Show one gentle camera push while the lamp remains still under soft studio lighting. Preserve the background arrangement and finish with a clear view of the lamp. No dialogue, new accessories or scene cut.
This is an original untested creative brief. It describes an observable action and a visual acceptance condition. It does not set an exact lens path, guarantee object preservation or establish which audio controls are available in a particular wrapper. Put actual duration, quality and input settings in the selected workflow rather than hiding them in prose.
Follow one job through to its actual result
Alibaba’s first-frame API documentation requires the model, endpoint and key to belong to the same region and uses asynchronous task creation followed by polling. Preserve the returned task identifier. A slow response is not, by itself, a reason to create a duplicate paid task.
Record the accepted request and chosen settings alongside the result. Watch the whole exported file before extracting attractive stills. Inspect the base, shade edge, contact shadow and background at the opening, middle and end. If one feature drifts, simplify the move and compare the next result with the same approved source.
Compare alternatives by the job you need
Clout’s current first-party image/video catalog checked for this guide does not list HappyHorse. The Clout Image and Video Studio is an alternative creation workflow, not a link to HappyHorse. Select an available model, inspect its supported controls and review the current credit quote before generating.
For a source-frame exercise in Clout, the Wan shot planner helps prepare a separate brief. For organizing several supporting shots, use the B-roll planner. If you need HappyHorse specifically, use a provider that explicitly identifies the requested model and operation.
Expand only after a small test is accepted
A good next step is a second shot with the same subject and a different framing, not an immediate long sequence. Keep the accepted reference set stable and change one production decision at a time. Separate image-to-video, multiple-reference work and any editing operation in your notes.
Do not infer commercial terms, regional availability, output quality or benchmark position from a model label. Check current provider terms for the intended use and judge your own sample. This guide establishes neither a speed comparison nor a preservation success rate; it supplies a clearer way to make and review the first test.
Record the whole wait
The generation timing guide separates queue, processing and delivery so a browser timeout does not become an accidental duplicate request.



