Kling O3 Reference-to-Video on fal: Put a Character From Photos Into a New Scene (October 2026)
Kling O3 reference-to-video builds a 3 to 15 second shot around up to four references for $0.14 per second. What it accepts, how elements differ from plain reference images, and how it compares with HappyHorse 1.1.
Reference-to-video is the family of models that builds a new shot around people or objects you show it in photos, instead of animating one still or editing an existing clip. Kling O3 reference-to-video is the Kling version, and it is one of the two serious options for "put me in a scene I describe" at a working price: $0.14 per second, or $0.42 for the 4K endpoint. It sits beside HappyHorse 1.1 reference-to-video at the same rate. This guide covers what O3 accepts, where it differs, and how to pick.
What it costs
| Clip length | Standard and Pro ($0.14) | 4K ($0.42) |
|---|---|---|
| 5 seconds | $0.70 | $2.10 |
| 10 seconds | $1.40 | $4.20 |
| 15 seconds (the maximum) | $2.10 | $6.30 |
Read from fal on 18 October 2026. The same price for Standard and Pro follows a pattern across the Kling lineup; the Turbo and Pro comparison explains why a tier at the same price as Pro is rarely the right pick.
What it accepts
- References:
elementsandimage_urlstogether, up to four. An element is a character defined by a frontal image plus extra angles; plain images describe the style, the object or the setting. - Length: 3 to 15 seconds.
- Frame shape: 16:9, 9:16 or 1:1.
- Audio: an option to generate sound with the video.
The differences from HappyHorse 1.1 reference-to-video are mostly in the references and the frame. HappyHorse takes up to nine photos and offers nine aspect ratios including 4:5 for feed posts; Kling O3 takes four references and three shapes. Kling's model has a cinematic bias in motion and camera language; HappyHorse's selling points are the extra references and the vertical formats.
How to choose between them
| If you need | Pick |
|---|---|
| More than four reference photos | HappyHorse |
| A 4:5 feed-format result | HappyHorse |
| Voice and lip-sync in the same call | HappyHorse |
| A cinematic camera feel | Kling O3 |
| To stay inside a Kling-based workflow | Kling O3 |
| A product held by a person (a face plus an object) | Kling O3, which treats an object as an element too |
The last row is the one to try. A product ad needs two things kept consistent, a person and a product, and defining both as elements is a documented feature in Kling's schema. We have not yet verified how reliably HappyHorse handles an object as a second character, so for "me holding this bottle" Kling is the safer first attempt; see UGC ads from a product photo.
What decides the result
Reference-driven models are matching a named subject across angles. A single frontal selfie leaves the model inventing the profile, and viewers read the invention as "not quite them". Give it a frontal, a three-quarter and a profile in similar lighting with the whole head in frame, no sunglasses. If the outfit or build matters, add a full-body frame in the clothes you want. Four good references beat nine bad ones, and four is the ceiling here anyway.
Write the prompt as a scene, not a list of attributes: where the person is, what they do first and what they do next, the light, the camera. Our comparison of HappyHorse and Kling O3 goes through prompt shape and defaults in detail.
What it does not do
It builds a new shot. It does not copy the choreography of a clip you like (use Motion Control), and it does not keep the background or the cuts of a source video (use a video-edit model such as Kling O3 edit). It accepts real faces in our experience, with the usual caveat that Kling enforces restrictions on public figures; a rejection is a policy call.
Honest limits of this guide
The price, the reference cap and the formats are from fal's schema on 18 October 2026. It is an active option in our model catalogue, but we have not run a controlled head-to-head against HappyHorse 1.1 on one set of photos, so "cinematic feel" is a description of the vendor's positioning, not a measured result. If this job matters to you, run the same four photos and the same prompt through both for about $1.40 each at ten seconds and judge on your own phone.
Common questions
How many references can I give it?
Up to four in total, counting elements and plain images together. HappyHorse 1.1 takes up to nine photos, which is the main reason to prefer it when a person needs many angles.
Can it hold a product as well as a person?
Kling documents objects as elements, so a person plus a product is a supported pattern. We have not yet verified how reliably HappyHorse handles an object as a second character.
What aspect ratios are available?
16:9, 9:16 and 1:1. For a 4:5 feed post, HappyHorse offers the ratio and Kling does not.
Does it accept real faces?
In our experience yes, with the usual Kling restrictions on public figures. A rejection is a policy decision, not a bug.
How to do this without code: Putting a character from photos into a described scene is the Me in scene tool. In AutoHustle this is the Me in scene tool: a few photos of you and one sentence describing the scene (about 2 credits per second, quoted before you press Generate). See How to put yourself in any scene and Put yourself in a movie scene.
Related reading: HappyHorse 1.1 Reference-to-Video: How Many Photos You Really Need and What It Costs (October 2026) · Kling 3.0 Pro Image-to-Video on fal: Price, Elements and Limits (October 2026)
For developers: the fal endpoints
| Model | fal endpoint |
|---|---|
| Kling O3 Pro reference-to-video | fal-ai/kling-video/o3/pro/reference-to-video |
| Kling O3 Standard reference-to-video | fal-ai/kling-video/o3/standard/reference-to-video |
| Kling O3 4K reference-to-video | fal-ai/kling-video/o3/4k/reference-to-video |
| HappyHorse 1.1 reference-to-video (comparison) | alibaba/happy-horse/v1.1/reference-to-video |
Prices read on 18 October 2026: $0.14 per second (Pro and Standard), $0.42 (4K). Inputs: elements and image_urls (4 in total), generate_audio, aspect ratio 16:9 / 9:16 / 1:1.