Video
HappyHorse 1.1
Light creative shorts with snappy motion and pacing.
Available on Ofrai — chat, images, and video in one place with shared credits.
Example prompts
Choose an example to load its prompt and recommended settings into the HappyHorse 1.1 creator above.
How it works
Create with HappyHorse 1.1 in three steps
- 01
Choose your inputs
Keep HappyHorse 1.1 selected, then add a prompt and any supported reference assets.
- 02
Set the output
Choose the aspect ratio, resolution, and available Video options before generating.
- 03
Generate and refine
Review the result, adjust the prompt or settings, and generate another version from the same place.
HappyHorse 1.1 hides four different jobs behind a single model id. Text-to-video, first-frame image-to-video, reference-image generation and video editing are all served by the same endpoint, and which one you are doing is decided by an action field rather than by a separate model name. If you omit that field the endpoint infers rather than errors: a supplied first-frame image is treated as image-to-video, and the absence of one is treated as text-to-video.
The consequence is that the parameters behave differently depending on the route — aspect ratio is honoured when you are generating from text and ignored when a first frame is supplied, the prompt is required on three routes and merely recommended on the fourth, and resolution defaults to 1080P, which is the opposite of the usual bargain-baseline assumption most people bring to a new video endpoint.
One model id, four jobs, one routing field
The action field is the whole API surface story on this endpoint. Set it and you choose explicitly between text-to-video, first-frame image-to-video, reference-image-to-video and video editing. Leave it out and the documented inference kicks in: if a first-frame image is present the request is handled as image-to-video, otherwise as text-to-video.
That is convenient and it is also the first place a bug can hide, because a request you meant as text-to-video but accidentally attached an image to will not be rejected — it will be routed as image-to-video, with all the different parameter consequences that follow.
The rule of thumb is to set the action field explicitly on every call, including on the paths where the inference would have guessed correctly, because the cost of being explicit is one field and the cost of being wrong is a render that silently used the wrong defaults for ratio and prompt handling.
1080P is the default, which inverts the usual assumption
Resolution defaults to 1080P here, with 720P available as the lower option. That default matters more than it looks. Most endpoints in this catalogue default to the lower tier so that iteration is cheap and the high tier is something you opt into once a shot is settled; this one defaults to the high tier, so an unconfigured request renders at full resolution and an iteration loop that never sets the field is a loop that never saves anything. Two habits are worth forming immediately.
Set resolution explicitly on every call, including to the value you would have got anyway, so that the behaviour of a request is visible in the request. And note the value spelling: the options are written with a capital P, 1080P and 720P, which differs from the lower-case convention some other endpoints use — worth checking against your own validation layer before assuming a rejected value is a server-side problem.
Duration runs three to fifteen seconds
Duration is measured in seconds with a documented range from 3 to 15, and the reference names 3 and 15 as the concrete values illustrated. A 3-to-15 window is wide enough to cover almost every short-form deliverable: a three-second loop-friendly beat for social, a mid-range beat for an explainer, and a fifteen-second take for a shot that genuinely needs to be continuous. The trade-off is the usual one, and here it is sharpened by the resolution default.
A fifteen-second 1080P render is the most expensive thing this endpoint can produce, so length is not just an editing decision here, it is a spend decision. Plan the cut points in the brief, render each beat separately, and reserve a single long generation for the cases where splitting would break the shot. Three seconds is the shortest request available, so if your design calls for a sub-second stinger, this endpoint will not produce it and that requirement needs to move elsewhere in your pipeline.
The first frame sets the frame, and it has hard image limits
When a first-frame image is supplied, the picture's geometry comes from that image rather than from the aspect ratio field — the documented ratio list governs text-to-video, and image-to-video follows the frame you uploaded. Supplying a landscape still and requesting 9:16 will not re-compose the shot.
The image itself carries hard constraints that are easy to violate by accident: formats are JPEG, JPG, PNG, BMP and WEBP; file size must not exceed 10MB; width and height must each be at least 300px; and the aspect ratio must fall between 1:2.5 and 2.5:1. Anything outside those bounds is a rejected request rather than a degraded one, so an asset-pipeline check before submission saves a render cycle.
One further detail is easy to get wrong: only the first image in the list is used for the first frame, and several alternative field names are accepted for the same input. Whichever alias you use, verify your file satisfies the limits above.
Prompt requirements change with the action you pick
The prompt is required for text-to-video, for reference-image-to-video and for video editing. It is optional for first-frame image-to-video, though the documentation recommends supplying it there as well, and that recommendation is worth taking: the prompt is the only channel you have for controlling movement and style once the frame is fixed.
Writing a prompt for an anchored frame is a different exercise from writing one from scratch, because restating what is already visible in the image competes with the image for the model's attention. Put the subject's action in one sentence, the camera move in another, and style terms last so they modify rather than compete. For the editing route the prompt describes the transformation rather than the scene, and for the reference-image route it describes the shot that the references inform.
Treating all four routes as the same writing task is the most common source of wasted renders here.
Async task IDs shape how you build a pipeline
Jobs on this endpoint are asynchronous and results are retrieved by task ID, which is an ordinary and well-understood pattern that nonetheless has one real consequence: your request and its result are separated in time, so the only record of what produced a given clip is what you stored at submission.
The practical implication is that the submission layer should capture the full request — action, prompt, duration, resolution, aspect ratio, and which image field you used — alongside the task ID, because the model will not hand that back to you later. Two related planning points. Because the endpoint serves four jobs under one name, task IDs from different routes will interleave in whatever queue you poll, so store the action with the ID rather than inferring the route from the result.
And since the model id does not change across routes, there is no cheap way to compare routes with identical parameters — comparing them means holding everything else constant deliberately, one variable at a time.
FAQ
- What is the default resolution, and which values are available?
- 1080P is the default, with 720P as the lower option. This is the reverse of the common pattern where the cheap tier is the default, so an unconfigured request renders at full resolution. Set the field explicitly on every call if you want iteration to be cheap, and note that both values are spelled with a capital P.
- What happens if I do not send the action field?
- The endpoint infers instead of erroring: if a first-frame image is present the request is handled as image-to-video, otherwise as text-to-video. That inference means a mistakenly attached image will not be rejected, it will silently change the route and with it the handling of aspect ratio and prompt — so set action explicitly on every call.
- What are the limits on a first-frame image?
- Formats are JPEG, JPG, PNG, BMP and WEBP; file size must not exceed 10MB; width and height must each be at least 300px; and the aspect ratio must be between 1:2.5 and 2.5:1. Only the first image in the list is used. Violations are rejected outright rather than silently degraded, so validate assets before submitting.
- Which durations can I request?
- Whole seconds from 3 to 15, with 3 and 15 named as the illustrated values. Because resolution defaults to 1080P, length here is also a spend decision — render beats separately and reserve a single long generation for shots that genuinely have to be continuous. Sub-second stingers are not available on this endpoint.
FAQ
- Can I use HappyHorse 1.1 directly on this page?
- Yes. The creator above is the working Ofrai studio with HappyHorse 1.1 selected by default.
- Where are my finished results saved?
- Signed-in generations are kept in your Ofrai library so you can download or revisit them.
- Which prompts work best with HappyHorse 1.1?
- Describe the subject, setting, visual style, lighting, and camera or motion clearly. The examples above provide practical starting points.
- Can I edit the example prompts before generating?
- Yes. Load any example, then change its wording, aspect ratio, references, or other supported settings in the creator.
Related models
Compare other Video models that can handle similar creative work to HappyHorse 1.1.
Related guides
Tools
Commercial use depends on the selected model and model terms.





