How to use HappyHorse reference-to-video (multi-image AI video)

Image-to-video animates one picture. Reference-to-video composes several. If you want a specific person wearing a specific outfit and holding a specific object — and no single photo has all three — R2V is built for exactly that: it treats each reference image as one ingredient and assembles them into a scene your prompt describes.

This makes it the right tool for combining a character with props, outfits, or accessories from separate shoots, rather than for animating a single existing photo as-is. If you already have one image that shows everything you want, plain image-to-video is simpler and cheaper.

Treat the prompt as stage directions for a small scene, where each reference image is a named prop or performer. Introduce a reference the first time you use it — "the woman in a red qipao from [Image 1]" — so the model has both the visual source and a plain-language description to anchor it to.

Once a reference is introduced, you can describe it acting: "unfolding the fan from [Image 2]", "the tassel earrings from [Image 3] sway with her head movement". This lets you sequence a short scene — an opening shot, an action in the middle, a closing detail — while keeping every element tied back to its source image.

Keep the numbering in your prompt consistent with the upload order. A mismatch between the order you provided images and the order you reference them in the prompt is the most common cause of the wrong element showing up in the wrong place.

Consistent lighting and quality across references matters more than it seems. If one reference is a bright, sharp studio photo and another is a dim, grainy phone snapshot, the model has to reconcile two different qualities of source material in one scene — the result often looks like it's compositing rather than one coherent shot.

Show each subject clearly and separately. A reference image where the outfit or object is partially hidden, at an odd angle, or mixed in with other clutter gives the model less to work with than one where it's the clear focus of the frame.

Fewer, stronger references beat many marginal ones. Three sharp, well-chosen images that clearly show the person, the key prop, and the key detail will outperform seven images where several barely add anything and only add room for confusion.

A reference image doesn't appear in the output: the prompt didn't clearly introduce it, or its image number doesn't match its upload position. Re-check the order and make the reference to it more explicit and specific.

Elements look like they don't belong in the same shot: the reference images differ too much in lighting, quality, or resolution. Replace the weakest one with a cleaner, better-lit image.

Wrong element shows up where another was expected: the numbering in the prompt doesn't match the order the images were provided in. Double-check [Image 1], [Image 2], [Image 3] against your actual upload order.

Output feels cluttered or the focus is unclear: too many references competing for attention. Drop to the two or three that matter most for the shot.

If you have one photo that already shows everything you want animated, plain image-to-video (like Wan) is simpler, cheaper, and has one less thing to get wrong. Reach for reference-to-video specifically when the shot you're after is assembled from parts — a person from one photo, an item from another — and no single image contains the whole scene.

The two techniques also compose: you can use R2V to generate a combined starting frame or short clip, then treat its output as the source image for further single-image animation or chaining into a longer sequence.