Character, coat, and street in one push-in
Image 1 holds the person, image 2 holds the coat, and image 3 holds the street. The move is a slow push-in. Check that the coat and the signature color stay recognizable from the first frame to the last.




Turn up to three reference images into one AI video. Guide the subject, scene and motion, review your clip, and download it with Happy Horse 2.
Each setup is a separate shot: its own pictures, its own move, and its own frame. Use this setting copies that ratio into the console. It leaves an action you already wrote untouched, and it does not generate.
Image 1 holds the person, image 2 holds the coat, and image 3 holds the street. The move is a slow push-in. Check that the coat and the signature color stay recognizable from the first frame to the last.




The front view and the side view describe one shoe. The move is a small orbit in a vertical frame. Check that the silhouette and the colorway stay the same shoe.



Three views of the same backpack agree on the strap count and the color. The shot sits on a table in a wide frame. Check that a second bag does not appear and that the straps do not multiply.




Reference to video is a short shot built from pictures you already have. You upload one to three images, put them in order, and write one action. The model returns a single clip. It does not animate a stack of unrelated photos into a montage.
Image 1 is the subject that should stay recognizable: a person, a shoe, a bag. Later images are clues, not extra heroes. A coat detail tells the model which garment to keep. A street tells it where to stand. A second angle of the same object tells it which straps and colors belong.
That is different from a first-and-last-frame shot, which treats two pictures as the start and the end of a timeline. It is also different from the homepage flow that animates one still. Reference to video is the looser brief: several pictures, one moment, one move.
Order the pictures, name what each one is for, write one action, pick the frame, then generate. The work happens in the console above. Nothing on this page sends you back to the homepage to finish.
Arrange one to three images. The first thumbnail is image 1, the subject. Image 2 and image 3 are the object, the material, or the place.
Write one action in the prompt. Include the camera in that sentence: a slow push-in, a slight orbit, or a short lateral move across a table.
Choose 16:9 or 9:16. The length stays at 8 seconds on Veo 3.1 Fast. The Generate button shows the credit cost before you click it.
Watch the clip on this page. Check the subject at the start, the middle, and the end, then download the MP4.
If you refresh or lose the connection, open My Creations and download the same task. That lookup does not start another job.
The page treats image 1 as the main subject and the later images as support for the roles you describe. You do not fill a separate camera box. Put the move in the action.
Use this setting, under each case, fills the ratio and, only when the prompt is empty, a starting action. It never starts a generation. If you already wrote an action, that text stays.
Pick the start that matches the pictures you have. Reference to video wins when the subject, the object, and the place live in different files and still need to become one shot.
You bring one to three pictures and one action. Image 1 leads. The other pictures supply an object or a place. Happy Horse 2 runs that shot on Veo 3.1 Fast for 8 seconds, in 16:9 or 9:16, and keeps the preview and the download here.
The whole shot is already in a single still. Use it when you do not need a separate product angle or a separate location. It cannot carry three references that disagree about what the subject is.
Animate one photoYou already know the opening picture and the closing picture. The model fills the time between them. That is a timeline. It is the wrong tool when the second picture is a detail, not a later moment.
Bridge two framesYou have no pictures. The model invents the person, the product, and the room. Useful for a blank start. Weak when a real coat, shoe, or bag has to survive the shot.
Start from textA useful result keeps one subject, one product, and the colors you uploaded. Play the whole 8 seconds. Extra arms, a duplicated shoe, or a bag that trades places with the person are failed reads of the references, even when the file itself plays.
Fix the input, not the wording alone. Move the clearest view of the subject to image 1. Remove a picture that shows a different object. Ask for one camera move. Changing the prompt or the ratio keeps the last file and does not spend credits. Clicking Generate again is a new charge.
Inputs, the Fast reference setup, failed reads, and where the download goes.