How to Make an AI Product Video From One Photo
Put your real product into AI-generated footage: what photos to use, how to describe the scene, what to check before you approve the clip, and how to get vertical versions for social media.
A short video of your product in use (poured, worn, unboxed, sitting on a sunny shelf) sells better than a still photo. Shooting one means props, lighting and an afternoon. With reference-to-video, you give the AI a photo of the real product, describe the scene, and get a clip with your product in it, with sound.
This guide covers what to photograph, how to describe the scene, what to check before you approve, and how to get versions for each platform.
How it works in one paragraph
You attach one or more photos to your message and say what should happen. The agent looks at each photo, names what is in it (“the amber glass bottle with a gold dropper cap in @Image 1”), and writes a shot-by-shot prompt around it: where the scene is, what the camera does, what happens, and the sound. You see that prompt, the photos and the clip’s settings on a card, and nothing is generated until you approve it. The clip then renders in the background and lands in your Files.
Step 1: choose the right photos
The photos matter more than anything you write. More is not better: one to three good images beat six mediocre ones, because too many make it hard for the model to tell which details matter.
- One product per photo, filling most of the frame, on a plain background.
- Even, soft light that shows the true colour and shape. No harsh reflections hiding the label.
- A second angle only if the back or side matters in the scene.
- No collage or multi-angle sheet in one image. The model reads several views as several products and may show two of them.
- No watermarks, captions or price stickers. Text in a photo tends to show up in the clip.
Want a person in the clip too? A close-up headshot holds a face far better than a full-body photo where the face is small. Keep the number of people to two or three.
Step 2: describe the scene like a director
You do not have to write the technical prompt. The agent does. But the more concrete your idea, the better the clip. Cover four things:
| What | Vague | Concrete |
|---|---|---|
| Place and light | “a nice bathroom” | “a white marble shelf, soft morning light from a window” |
| Action | “someone uses it” | “a hand picks it up, turns it, light glows through the glass” |
| Camera | “cinematic” | “slow push-in, then a close-up of the dropper” |
| Sound | — | “soft, airy electronic music; a glass clink” |
Sound is generated with the picture: music, sound effects, even a line of dialogue if someone speaks. Say what you want to hear, or say “no music” if you will add your own.
Step 3: pick length, shape and quality
- Length: 5 to 15 seconds per clip. For an ad, 8-10 seconds with one clear action is usually stronger than 15 seconds of wandering camera.
- Aspect ratio: 9:16 for Reels, TikTok and Shorts; 1:1 or 3:4 for feeds; 16:9 or 21:9 for a website banner or YouTube.
- Resolution: 720P for the final version, 480P to try an idea first.
If you leave these out, the agent asks.
Step 4: check the card before you approve
The approval card shows your photos numbered @1, @2 and so on, the settings, and the full prompt. Read it for three things:
- The product is described by what is really in your photo: colour, material, cap, label.
- The prompt asks to keep the label and colours exactly as in the photo.
- The action is something a product can plausibly do. Liquids pour, fabric moves, a box opens; a bottle does not dance.
Not right? Press “Change something” and say what to change. The agent rewrites the prompt and shows you a new card.
Step 5: get the clip, and make variations
A clip takes a few minutes. You can keep chatting or close the tab; the Tasks tab shows progress, and the finished MP4 appears in Files, ready to play or download. If a clip fails, it is refunded automatically.
Variations are one message away: “the same scene in 1:1”, “now outdoors at golden hour”, “a version where a woman holds it and smiles at the camera”. Each one comes back as a new card to approve.
Honest limits
- Small print on a label can come out soft or slightly wrong. Large logos and short brand names hold up best; check the label in every clip before you publish.
- Hands and fine mechanical parts are the hardest things for video models. Keep actions simple and the product large in frame.
- A clip is one continuous scene. For a longer ad with several scenes, ask for a long video: the agent plans it as a storyboard and films it clip by clip, as described in the post on turning a story into an animated drama.