Write a motion prompt that stays on the product

The first prompt asked for a slow turn, a sunrise, a hand entering, and a logo at the end. The bottle that came back was a cousin of the bottle in the photo. The tests that fixed it were not poetic. They were shorter. A motion prompt has one job: say what moves, and say what must stay. The still already contains the product.

These are the fixes worth keeping, in the order they usually help. They apply on photo to video whether you type the line or speak it, and whether you generate with Standard, Ultra, or Express. The AI photo to video homepage is the studio. The prompt does not replace it.

Name one motion

A slow push-in. A small turn. Light shifting across the label. A nod. If you need a second beat, that is another clip, then a combine or a next scene. A paragraph of scenes is how the subject gets replaced. Try: slow push-in, keep the product, label, and colors exactly as in the photo, no new objects.

Say what must stay

The model will fill any silence with a new idea. If the label, the color, the shape, the face, or the clothes matter, write that they stay as shown. For a portrait: hold the face, hair, and clothes as shown, add a small nod, camera stays still. Leaving those words out is how you get a smoother clip of a different person.

Do not describe a new product

The photograph is the product. A prompt that says a matte black bottle when the photo is a clear glass bottle is a request to repaint the still. Describe the move. If the photo is wrong, change the photo. Do not argue with it in the prompt.

Keep the camera boring on purpose

A push-in or a locked camera is easier to judge than an orbit, a crane, and a zoom in one line. If the frame flies off the product, you cannot tell whether the label survived. Lock the camera unless the move itself is the point of the shot.

Ban the extras you did not photograph

No new objects is a useful ending to a product prompt. Hands, tables, sparkles, and extra packaging appear when the line is vague and the photo had empty space. If you want a hand, start from a photo that already has the hand. Do not hope the prompt will composite one cleanly onto a floating pack shot.

Read a spoken line before you spend the generation

On the phone, speech is the same instruction as typing. A misheard word is still the prompt. Look at the line. If it names the wrong motion, fix it, then generate. Express is a reasonable pass for that first check. Ultra is the pass when the line is right and the product has to look clean.

Retry shorter, from the same still

When the result replaces the subject, do not add adjectives. Delete a scene. Generate again from the same photo. Switching from Express to Ultra will not rescue a prompt that asked for a different world. Standard will not either. The pass changes the render. The still and the sentence decide the subject.

The line that replaced the bottle, and the line that did not

The failed prompt read like a treatment. Slow cinematic turn, morning light, a hand sets the bottle down, then a logo appears. The bottle in the photo was a clear glass with a blue label, shot at noon. The clip returned a darker glass and a hand that covered the words. Every extra clause had been treated as permission. The retry deleted the hand, the morning, and the logo. Logos are an overlay, not a prompt. The line became: slow push-in, keep the bottle, the blue label, and the glass exactly as in the photo, camera moves closer, no new objects.

That retry used the same still. They did not switch to Ultra to rescue the treatment. They ran the short line on Express, saw the label survive, and only then used Ultra for the file. A spoken version of the short line, captured on the phone while holding a second bottle, worked the same way after they read it back and removed a misheard “spin.” Speech was not the problem. Length was.

A small set you can reuse

Product push-in: slow push-in, keep the product, label, and colors exactly as in the photo, no new objects. Product turn, only when the still shows the side you are allowed to reveal: small turn, keep the product and the label as shown, no new objects. Portrait: hold the face, hair, and clothes as shown, add a small nod, camera stays still. Held product: handheld and close, keep the product and the hand as shown, one small move toward the lens. If you need a second sentence, you need a second clip.

Paste the closest one, swap the noun for your subject, and stop. Standard, Ultra, and Express will render that line differently in finish and in speed. They will not turn a treatment into a photograph. The still does that, or the generation fails and you shorten the line again.

A prompt you can reuse is boring in the best way. One move. One list of things that stay. No new objects. If that clip holds, stop writing and brand it.

They kept the failed line in a note for a week, not as a prompt to reuse but as a list of decorations that had replaced the bottle: sunrise, a hand, a logo, a second scene. Each of those words had seemed harmless alone. Together they gave the render permission to leave the photograph. The line they reused across the next three SKUs had none of them. It named a push-in, the product, the label, the colors, and a ban on new objects. Where a jar needed a turn instead of a push-in, they changed the verb and left the rest. Where a portrait needed a nod, they changed the subject of the hold — face, hair, clothes — and locked the camera. They did not add a mood. Mood is how a pack shot grows a window it never had.

Speech went through the same note. If the phone heard a word that was not in the photograph, they corrected the text before Express spent the test. They did not discover the error in the Ultra file and then blame the cleaner pass. The pass renders the sentence. The sentence must already be willing to live inside the still. Photo to video will take a poetic line and do its best. Its best, on a vague line, is a new product. Read the seven fixes as deletions: one motion, what stays, no repaint, a boring camera, no extras, a spoken line you have read, a shorter retry from the same file.

When the clip holds, stop. The next sentence is a caption, a logo, or another still. It is not a second act squeezed into a frame that has already told the truth. A reusable prompt is a constraint you can hand to someone else without standing over their shoulder. If they need a paragraph to explain it, the line is not reusable yet. Shorten it until the photograph can defend every word.

The person who inherited the line did not watch a tutorial. They swapped the product name for the next SKU only when the photograph itself had changed, and they left the motion alone when the pack shot was the same kind of frame. Three clips later the bottles still matched their pages. The inherited sentence had not grown. Growth is how a reusable line becomes the sunrise prompt again. They kept a failed example next to it, the long one, so the temptation to decorate had a corpse. New writers want to sound specific. Specific, here, means specific about the photograph, not specific about a world the photograph does not include. If a word cannot be pointed at in the still, it does not belong in the line, no matter how precise it sounds.

They printed the short line and the long line on the same page and asked a new teammate which one the photograph could defend. The teammate pointed at the short one without being told the ending. That is the test. If a colleague who has the still in front of them cannot tell which words the picture allows, the prompt is still a wish. Wishes belong in the caption after a true clip exists. They do not belong in the generation that decides what the product is. The short line shipped three more times that month. The long line stayed on the page as a warning, not as a draft to improve.

Point at the still. If the word is not there, delete the word. Then generate. That is the entire craft, and it is enough to keep the next bottle the bottle on the page.

This job is covered in The best AI video generator for a product still. That short line, posted vertical, is an AI reel.

Where this fits

This guide supports the AI photo to video studio on the KuvaVid homepage. The page for this job is photo to video.

Keep reading

KuvaVid · All articles