Hailuo Image to Video: Animate a Still Step by Step
To make a video from an image in Hailuo, open image-to-video, upload your still as the first frame, write a one or two sentence prompt that describes movement and camera rather than appearance, pick a model version and duration, then generate. The image fixes the look; the prompt only needs to say what moves and how the camera behaves.
Hailuo (hailuoai.video, made by MiniMax) is known for natural-looking motion: people walk, fabric moves and cameras glide without the jittery, over-animated feel of early video models. Image-to-video is the mode most working creatives use, because starting from a still you already approve keeps composition, colour and character design under your control. This guide walks through the workflow, the prompt style that suits Hailuo, and the fixes for the problems you will meet.
How image-to-video works in Hailuo
You upload an image, which becomes the first frame of the clip, and write a prompt describing the motion. Hailuo then generates a short clip, a few seconds long, that begins exactly on your still and animates it. Several model versions are available in the current interface, and newer ones generally follow motion instructions more faithfully and hold the subject’s identity better; the list changes, so check which version is selected before you generate.
Hailuo also has subject-reference features that let you upload an image of a person or character so their face and appearance carry through text-to-video generations. That is a different mode from image-to-video: image-to-video fixes the whole first frame, while a subject reference fixes the character and lets the prompt build the scene. Use image-to-video when the composition matters, and the reference route when you need the same character in many different shots.
Generation costs credits that vary with model version and duration. Check the platform’s current pricing page rather than older guides.
Step by step: your first clip
- Prepare the still. Use a sharp image at the aspect ratio you want in the video, with the subject clear of the edges so motion has somewhere to go. Remove any text or logos; video models smear lettering.
- Open image-to-video and upload. The still goes in the first-frame slot. Check the preview is not cropped.
- Write a motion prompt. One or two sentences: who moves, how, and what the camera does. "The woman lifts the cup and takes a slow sip. Camera holds still."
- Choose model version and duration. Start with the shortest duration offered and the version you have had the best results with.
- Generate and review at full size. Step through the first and last second frame by frame; that is where warping appears.
- Iterate one change at a time. Adjust the verb, the pace word or the camera, not all three at once, so you learn what each does.
Writing a prompt for a still
The most common error is pasting the image’s original prompt back in. Hailuo can already see the picture. The prompt should add only what the picture cannot show: movement, timing and camera.
- One subject, one action. "The dog shakes water from its coat" is reliable. "The dog shakes, barks and runs off" is three clips’ worth of motion crammed into one.
- Pace words calm the model. "Slowly", "gently", "a subtle breeze" produce natural motion. Without them clips often move too fast for their length.
- Name the camera explicitly. "Static camera", "slow push in", "gentle pan left". Hailuo responds well to film vocabulary; the Hailuo prompting guide lists the terms that work.
- Say what stays still. "Background remains fixed, face stays in frame" reduces drift.
- Skip appearance and style words. They are already in the image. Adding "cinematic, 8k" does nothing useful here.
Settings at a glance
| Setting | What it does | Start with |
|---|---|---|
| Model version | Selects the Hailuo model; newer versions follow motion and hold identity better | The latest version offered |
| First-frame image | The still the clip starts on | A sharp image at the final aspect ratio, no text |
| Prompt | Describes motion and camera | One action plus one camera instruction |
| Duration | Length of the clip | The shortest option, then longer once the motion works |
| Resolution | Output size where the version offers a choice | Standard for tests, highest for the keeper |
| Prompt enhancement | Optional rewriting of your prompt by the platform, where offered | Off when you want precise control |
| Subject reference | Separate mode that fixes a character rather than a frame | Use for multi-shot character work |
Setting names in the interface may differ slightly from these descriptions, and options vary by model version.
Common problems and fixes
- The subject warps or melts mid-clip. Too much motion was requested for the duration. Cut to one action, add a pace word and try a newer model version.
- Nothing moves. The prompt described the scene rather than an action. Use a verb and name a camera move.
- The face changes. Keep the face large in frame, avoid fast head turns, and add "face stays consistent". For a character who must survive many shots, switch to the subject-reference route.
- The camera does something you did not ask for. Hailuo adds camera motion when none is specified. Write "static camera" if you want a locked-off shot.
- Lettering smears. Remove text from the still and add it back in your editor.
- The clip looks fine but the last half-second collapses. Trim it in the edit. Most clips have a usable core even when the tail fails.
Try it
First frame: a woman in a yellow raincoat standing on a wet pier, grey sky, sea behind her.
Prompt: A gentle breeze moves her hair and the hem of the raincoat. She turns her head slowly to look out to sea. Slow push in towards her, static background, natural motion.
Duration: shortest available.
Generate, then replace "slow push in" with "static camera" and run again. The pair shows how much of the feel comes from the camera instruction alone, and tells you which to keep for the final version at a longer duration.