AI Product Video Generator: Turn a Product Photo into a Video Ad

Product pages and paid feeds keep moving toward video, but most small stores still have exactly one asset per product: a photo. An AI product video generator closes that gap in the narrow way that actually works today — you supply the photo of the product you sell, describe the motion, light and camera move, and get back a short clip you can download and publish.

This guide walks through the whole loop on Cyanvid: which photo to pick, how to write a prompt for a shot where the subject is already fixed, which model to render each version with, and where the honest limits are. Nothing here requires an editor, a studio, or footage of your product.

Why a photo is the right input for an ad

Text-to-video invents the product. That is fine for a mood shot, but for an ad the thing in the frame has to be the thing you sell — the same shape, the same label, the same colour. Image-to-video solves this by starting from your photo.

The model uses your image as the reference for the subject, its colour and its framing, and your prompt only has to describe what changes: light, camera, and small ambient motion. You are not describing a product from scratch and hoping it looks like yours; you are animating the product you already photographed.

One honest caveat before the workflow: the photo anchors the subject, but it is not a frame-perfect guarantee. Small details — label edges, fine texture, reflections — can drift during generation. For ad use, review the render at full size before you publish it. The limits section below lists what to check.

The workflow in four steps

1. Upload the photo. PNG or JPEG. A clean, well-lit shot with a plain or simple background works best, because the model carries that background into the clip. The photo also decides the composition, so frame it the way you want the ad framed. See photo to video for the current file limits.

2. Pick the model and its frame. Three current models accept an image: Wan 2.2 5B Fast (9:16), Wan 3.0 (16:9 and 9:16) and Seedance 2.0 Fast (16:9 and 9:16). Decide where the ad will run first — vertical for TikTok, Reels and Shorts, landscape for product pages and website heroes — because the frame is chosen at generation time, not cropped later.

3. Write the motion prompt. The subject is already set by the photo, so your prompt describes only the change: one camera move, one lighting change, one ambient motion. Patterns and ready-made prompts are in the next section.

4. Check the quote, render, review. The studio shows the exact credit cost before you commit, so the price is never a surprise. A render typically finishes in one to three minutes, and the MP4 lands in your private history. Watch it twice — once for the motion you asked for, once for artifacts around the label and edges. If a job fails, the charged credits are refunded automatically.

That is the whole loop. Open the studio or start from the dedicated AI product video page, which is set up specifically for product ad creative.

Prompt patterns for product shots

Because the photo already fixes the subject, a product prompt has four parts instead of six:

1. Camera — one move only: push in, pull back, slow orbit, or locked off.
2. Light — one change: a sweep across the surface, a rim light passing an edge, dawn-to-dusk warmth.
3. Ambient motion — something small the product itself does not do: steam, condensation, drifting dust, a flickering flame beside it.
4. Constraints — what must hold still: the label, the background, no people, no text overlays.

That last line matters more for products than for any other subject. Models tend to distort text, and a warped label is the fastest way to make an ad unusable. Saying "the label stays static and legible" reduces it; keeping the camera off extreme close-ups of the label removes it.

Here are five prompts written for common product types. They are written templates to adapt, not sample renders:

Skincare or cosmetics bottle:

Slow push-in on the bottle, soft window light sweeping from left to right, faint condensation on the glass, the label stays static and legible, clean studio backdrop, shallow depth of field.

Headphones or small electronics:

Locked-off camera, the product rotates slowly about ten degrees on a matte turntable, a thin rim light passes along its edge, soft shadow drifting beneath, dark minimal background, no text overlays.

Jewelry:

Slow push-in toward the ring, a warm light glinting across the facets, black velvet background, shallow depth of field, the camera settles and holds, no hands in frame.

Beverage can or bottle:

Slow quarter-turn orbit around the can, cold mist drifting across the lower frame, condensation glistening, bright daylight look, label stays straight and readable, no text overlays.

Candle or home fragrance:

Static camera, the flame flickers softly, warm evening light on the jar, faint smoke curling upward, cozy minimal scene, label stays static and legible.

Each is one shot, one camera move, one light change — sized for a five-second clip. If the result moves too fast, cut the speed in half by changing only the camera phrase ("very slow push-in") and keep everything else. The full method is in how to make an AI video, and the AI video prompt generator assembles the parts for you if you would rather pick from options.

Which model to render each version

The three image-to-video models split the work cleanly:

• Wan 2.2 5B Fast — 2 credits, 9:16, five seconds at 720p, silent. The cheapest way to iterate: at 2 credits a clip, you can afford several retries before the good one.
• Wan 3.0 — 24 credits, 16:9 and 9:16, five seconds at 720p, with generated audio. The pick for final renders, and the only current i2v model that produces sound.
• Seedance 2.0 Fast — 32 credits, 16:9 and 9:16, five seconds at 720p, silent. The alternative when you need the landscape frame without audio.

A practical split: draft vertical ads on Wan 2.2 Fast until the prompt is right, then spend the larger credit cost on Wan 3.0 for the version you publish. New accounts start with 8 welcome credits — four free Wan 2.2 drafts before you pay anything.

Free credits and no watermark

Two questions come up first, so here are direct answers.

Is it free? New accounts receive 8 welcome credits with no card required. Wan 2.2 5B Fast costs 2 credits per clip, so that is four free product videos before any purchase. The exact credit quote is shown before every render, and failed jobs are refunded automatically.

Is there a watermark? No. Downloads from every plan — including free-credit renders — are clean MP4 files without a watermark, and they stay private in your history rather than on a public feed.

If you are validating whether video ads are worth producing at all, that makes the trial loop concrete: generate four drafts free, cut the best one into your feed, and only then decide about paid credits on the pricing page.

Fit the clip to the platform

• 9:16 for TikTok, Reels, Shorts and Stories. Wan 2.2 covers vertical natively; Wan 3.0 and Seedance 2.0 Fast also have verified vertical variants.
• 16:9 for product pages, marketplace listings and website heroes. Wan 3.0 and Seedance 2.0 Fast.
• Five seconds is the current i2v duration. Treat it as one shot in a larger edit: most product ads are two or three five-second clips — opening, detail, pack shot — cut together in any editor. The free product video script generator drafts that three-beat structure, and there is a dedicated Amazon product video script generator for listing-style ads.
• 720p is the current output for all three i2v models. Fine for feeds; upscale or reshoot if a campaign needs 4K masters.

Compose for the frame you will publish in — a 16:9 render cropped to vertical loses the edges, and usually the product with them.

What this cannot do

The limits, so you do not spend credits finding them:

• Five-second single shots. One continuous take per generation, no scene cuts inside a clip.
• 720p output on the current i2v variants. Check the model picker for any newer options before planning a campaign.
• Text and logos can distort. The model redraws fine detail probabilistically. Keep the camera away from extreme label close-ups, add "label stays static and legible" to the prompt, and review at full size before publishing.
• Details can drift, not just text. The photo anchors subject, colour and framing, but fine texture and reflections are re-generated. If your product's identity lives in a micro-detail, check that detail first.
• Audio only on Wan 3.0. The other two produce silent clips; add music in your editor.
• No editing, lip sync, or person replacement. There is no timeline here — download the MP4 and finish elsewhere.

FAQ

Does the generated video actually look like my product? The photo sets the subject, colour and framing, so it starts from your product rather than inventing one. Small details can shift during generation — check the label and edges at full size before you use a render.

How much does it cost? Wan 2.2 5B Fast is 2 credits per five-second clip, Wan 3.0 is 24, Seedance 2.0 Fast is 32. The studio quotes the exact price before you confirm, and failed renders are refunded. Eight welcome credits mean four free Wan 2.2 clips.

Can I use the clips in ads? The MP4 downloads without a watermark and is yours to use. Before running paid campaigns, confirm your plan terms and the ad platform's own policies on AI-generated content.

How long does a render take? Typically one to three minutes. The finished MP4 appears in your private history.

Which file types can I upload? PNG or JPEG, up to the current limits on the image-to-video page. Pick a sharp, well-lit photo — the clip inherits the lighting and background of the shot.

---

If the product-photo workflow fits, the adjacent guides cover the rest of the pipeline: 12 annotated AI video prompt examples for other verticals, and the step-by-step AI video walkthrough for text-to-video shots where no photo exists. When you are ready to try it, open the AI product video page or go straight to the studio.