How to Use Hailuo Image-to-Video for Product Shots
Image-to-video is the half of Hailuo that gets overlooked. You supply a still and describe how it should move; the model keeps the composition and animates it. For product marketing, that is a much shorter path to usable footage than prompting a scene from nothing, because you already control the frame.
Why start from an image rather than a prompt?
Because composition is the hardest thing to prompt and the easiest thing to photograph. If you have a clean product photo on white or a styled lifestyle shot, the model only has to solve for motion. That is a far smaller problem than inventing subject, framing, and lighting simultaneously.
In practice: text-to-video gave usable frames roughly one in three. Image-to-video gave usable frames closer to three in four, provided the source still met a few conditions.
What does the source image need?
| Requirement | Why it matters |
|---|---|
| At least 1280px on the long edge | Below that, the animation gets soft |
| Clear subject separation from background | Motion artefacts appear at blurred edges |
| No small text on packaging | Generated frames garble text |
| Consistent lighting, no blown highlights | The model amplifies any flicker |
Small text is the killer for consumer goods. If your label has four lines of ingredient copy, the model will animate them into runes. Shoot a version with minimal or no legible text for the animated shot.
How do you write the motion prompt?
Describe what changes, not what exists. The scene is already there.
Weak prompt: “A bottle of serum on a marble counter, soft lighting, premium.”
Working prompt: “Slow orbit to the right around the bottle, gentle steam rising behind, camera height fixed, no zoom.”
The second prompt tells the model what to do with time. That is the only thing it needs from you in this mode.
What motions look convincing and which look fake
Convincing: slow orbit, slow push in, slight parallax, steam, liquid pouring, fabric moving in a breeze, light shifting across a surface.
Fake: the object rotating to show the back label, a hand picking something up, anything where two surfaces intersect, crowd scenes. Object rotation is a specific trap — the model invents the unseen side of the product and it is always wrong.
If a client needs a 360° product view, shoot a turntable. It takes twelve photos and five minutes.
How many clips per product makes a usable sequence
Three is the sweet spot: an establishing orbit, a detail push-in, and a lifestyle scene with motion. Six-second clips at these lengths cut together into a 15-second ad without feeling repetitive.
Use the same source lighting across all three so the cuts land. Teams that generate each shot independently, at different times with different prompts, end up with clips that each look fine and look wrong together. MiniMax Hailuo has no style-locking feature, so consistency is a manual discipline.
For the assembly stage, Pictory Video can take the clips and add captions and a voiceover. If the launch needs a real human demoing the product, record it with Tella (AI Recorder) instead of animating.
Who should skip it
Regulated products where the label must be legally accurate frame by frame. Anything where a minor difference in packaging could be misread as a different SKU. And anyone who needs more than 10 seconds of continuous camera move — the clip length is the hard limit.
Quick Summary
- Image-to-video keeps your composition and animates motion, with far higher hit rates than text-to-video.
- Source stills need 1280px+, clean edges, and no small text.
- Prompt only motion: orbit, push in, steam, light shift.
- Never ask the model to rotate a product to reveal an unseen side.
- Three clips per product, shot with matching light, cut into one sequence.
